Conference PaperPDF Available

Abstract

The agricultural and farming industries have been widely influenced by the disruption of the Internet of Things. The impact of the IoT is more limited in countries with less penetration of mobile internet such as sub-Saharan countries, where agriculture commonly accounts for 10 to 50% of their GPD. The boom of low-power wide-area networks (LPWAN) in the last decade, with technologies such as LoRa or NB-IoT, has mitigated this providing a relatively cheap infrastructure that enables low-power and long-range transmissions. Nonetheless, the benefits that LPWAN technologies enable have the disadvantage of low-bandwidth transmissions. Therefore, the integration of Edge and Fog computing, moving data analytics and compression near end devices, is key in order to extend functionality. By integrating artificial intelligence at the local network layer, or Edge AI, we present a system architecture and implementation that expands the possibilities of smart agriculture and farming applications with Edge and Fog computing and LPWAN technology for large area coverage. We propose and implement a system consisting on a sensor node, an Edge gateway, LoRa repeaters, Fog gateway, cloud servers and end-user terminal application. At the Edge layer, we propose the implementation of a CNN-based image compression method in order to send in a single message information about hundreds or thousands of sensor nodes within the gateway's range. We use advanced compression techniques to reduce the size of data up to 67% with a decompression error below 5%, within a novel scheme for IoT data.
Edge AI in Smart Farming IoT:
CNNs at the Edge and Fog Computing with LoRa
T. Nguyen Gia1, L. Qingqing1,2, J. Pe ˜
na Queralta1, Z. Zou2, H. Tenhunen3and T. Westerlund1
1Department of Future Technologies, University of Turku, Finland
2School of Information Science and Technology, Fudan Universtiy, China
3Department of Electronics, KTH Royal Institute of Technology, Sweden
Emails: 1{tunggi, jopequ, tovewe}@utu.fi, 2{qingqingli16, zhuo}@fudan.edu.cn, 3hannu@kth.se
Abstract—The agricultural and farming industries have been
widely influenced by the disruption of the Internet of Things.
The impact of the IoT is more limited in countries with less
penetration of mobile internet such as sub-Saharan countries,
where agriculture commonly accounts for 10 to 50% of their
GPD. The boom of low-power wide-area networks (LPWAN) in
the last decade, with technologies such as LoRa or NB-IoT, has
mitigated this providing a relatively cheap infrastructure that
enables low-power and long-range transmissions. Nonetheless, the
benefits that LPWAN technologies enable have the disadvantage
of low-bandwidth transmissions. Therefore, the integration of
Edge and Fog computing, moving data analytics and compression
near end devices, is key in order to extend functionality. By
integrating artificial intelligence at the local network layer, or
Edge AI, we present a system architecture and implementation
that expands the possibilities of smart agriculture and farming
applications with Edge and Fog computing and LPWAN technol-
ogy for large area coverage. We propose and implement a system
consisting on a sensor node, an Edge gateway, LoRa repeaters,
Fog gateway, cloud servers and end-user terminal application. At
the Edge layer, we propose the implementation of a CNN-based
image compression method in order to send in a single message
information about hundreds or thousands of sensor nodes within
the gateway’s range. We use advanced compression techniques
to reduce the size of data up to 67% with a decompression error
below 5%, within a novel scheme for IoT data.
Index Terms—IoT; Internet of Things; Smart Agriculture;
Edge Computing; Fog Computing; Edge AI; LoRa; LPWAN;
Low Power Wide Area Networks; CCN; Deep Learning; ML;
I. INTRODUCTION
In order to perform activities such as watering or fertilizing,
farmers need to visit their plants frequently (e.g., every day or
every few days depending on the plant and trees). In some
cases, farmers need to stay close to their remote farms in
order to protect the crop and their resources. When the farmed
areas are large, it becomes increasingly difficult and more
human resources are required to perform these tasks. This
can cause a significant increase in operational costs with a
limited impact on productivity. In the era of the Internet of
Things (IoT), a solution is to deploy remote monitoring and
management systems for these remote farms. Farms adopting
IoT technologies are often referred to as smart farms. However,
remote farm monitoring and controlling have limited impact in
areas with unstable or poor Internet connectivity. This occurs
not only in developing but also in developed countries [1].
The Internet of Things (IoT) can be defined as a platform
where virtual and physical objects are interconnected and
communicate with each other. IoT systems consist of different
technologies such as wireless sensor networks, cloud comput-
ing and embedded intelligence. These systems offer advanced
services such as real-time remote monitoring, online analytics
and remote management. IoT is applied in many remote
monitoring applications in vast domains from healthcare to
smart factories, and including smart homes, smart cities and
smart farming, improving productivity and reducing costs [2]–
[6]. Some of the benefits of the IoT can be utilized to improve
the quality of services for automated and remote farming
systems. However, relying only on traditional cloud-centric
IoT architectures for remote farm monitoring and management
cannot guarantee that the systems work properly because
the IoT still presents several challenges. For instance, cloud-
centric IoT applications cannot be deployed in remote areas
where the Internet is not stable or coverage is limited. In such
cases, data cannot be real-time monitored and actions toward
abnormalities might not be executed on time. For instance, if
a sudden fire occurs or a group of wild animals raid the crops,
the system cannot react on time.
Edge and Fog computing can be illustrated as a mini cloud
which is closer to the edge of the network. In other words,
Edge and Fog computing represent the convergence of differ-
ent network layers into interconnected smart gateways. Edge
and Fog computing can help to overcome some limitations
of the traditional could-centric IoT systems. For instance,
Edge and Fog computing offer many advantages such as
energy efficiency, distributed local storage, interoperability and
enhanced security. In more detail, Edge and Fog computing
can help to reduce the network load and the computational
and storage burden of cloud servers. This is done by moving
many computationally intensive processes from the cloud to
the Edge and Fog layers and gateways, at the same time
enabling more power-efficient sensor nodes as they rely more
on local network smart gateways. Compared to traditional IoT
applications which often rely on a 3-layer architecture (sensor-
cloud-terminal), a Fog-assisted IoT application has extra layers
between the sensor nodes and the cloud. Depending on the
application and type of acquired data, a different number of
Edge/Fog layers can be deployed [7]–[9].
Even though Edge and Fog computing can provide many
advanced services, Fog-based systems still cannot work prop-
erly in remote areas where the Internet is not stable or
covered, as they often rely on high-speed local networks
for real-time processing and latency-critical applications. A
solution in these scenarios is to deploy low-power wide-area
network technologies, such as LoRa, which enable long-range
transmissions with the drawback of reduced data rates. LoRa
is one of the most popular LPWAN protocols for the physical
layer [10], offering low-power and long-range communication
up to 10 or 20 km in open and line-of-sight transmissions [11].
However, LoRa cannot be used to send data with high data
rate due to local regulations and limitations to the transmission
duty cycle in most regions of the world of 0.1%, 1% or
10%. Therefore, LoRa alone cannot help to solve the existing
problems of IoT applications in remote areas.
In this paper, we propose an advanced Edge/Fog assisted
LoRa-based system for remote farming. The system targets
remote areas in developing countries where the need is ev-
ident. We propose to leverage Edge and Fog computing,
together with LoRa communication to back the limitations
of each other while still providing smart services and more
efficient computation distribution when compared to cloud-
centric solutions. In addition, the integration of Edge and Fog
computing with LoRa into IoT systems can help to achieve a
high level of energy efficiency for sensor nodes. When some
abnormalities occur, Fog-assisted IoT systems can provide
support for latency-critical applications.
As LoRa has limited transmission speed, and there are strict
regulations controlling the duty cycle, we propose a novel
data compression technique integrating the spatial distribution
of sensor nodes, which can be densely deployed in smart
farms, with image compression methods. We have simulated
the output of thousands of sensor nodes and applied different
image compression techniques. We compare our proposed
method with standard JPEG compression and show how can
we achieve up to 67% reduction of data size while keeping
decompression error under 5%. This is a threshold often found
in many inexpensive sensors. For instance, it is equivalent to
having a 1 degree accuracy in a temperature sensor measuring
around 20oC. The main contributions of the paper are:
Advanced architecture for monitoring and controlling
remote farms where Internet connectivity is not reliable.
Integration of image compression techniques with convo-
lutional neural networks to compress data from multiple
sensor nodes at once exploiting their spatial distribution.
Enhancement of edge gateway for pre-processing data
and considerably reducing the amount of data to be
transmitted over the LoRa link.
The remainder of this paper is structured as follows: Section
II presents related work in remote monitoring and management
systems for smart farming. Section III describes the architec-
ture of a Fog-assisted IoT system with the integration of Lora.
Section IV includes the system implementation, edge layer
analytics and discusses experimental results, while section VI
concludes the work.
II. REL ATED WORK
Many efforts have been devoted to proposing smart and re-
mote farm monitoring. Maia et al. [12] present a remote mon-
itoring system for precision agriculture. The system consists
of monitoring nodes, central nodes, and node in cloud. Data
collected from sensors (e.g., soil temperature and humidity)
integrated into a monitoring node is sent to a central node via
ZigBee. The central node then forwards the received data to
nodes in cloud via the Internet.
Ragai et al. [13] present are a remote control and monitoring
system for fish farms by using wireless sensor networks.
The wireless sensor node collects different data such as
temperature, pH, dissolved Oxygen and sends the collected
data via ZigBee to gateways forwarding to cloud servers via
the Internet. When cloud servers detect abnormalities such as
too low or too high value of temperature and pH, they send
an alarm to farmers.
Saraf et al. [14] propose an IoT-based system for smart
irrigation monitoring and controlling. The system consists of
wireless sensor nodes, a center node connected to cloud and
an Android application. The sensor node acquires temperature,
humidity, soil moisture and water level of the tank and
transmits the data to a center node via Zigbee. The center
node then forwards the collected data to cloud which runs data
processing algorithms and do actions e.g., sending commands
to actuator nodes to control water.
Shaout et al. [15] present an embedded system for remote
agricultural areas monitoring. The system consists of sensor
nodes which are based on Arduino board, Bluetooth and
sensors. The sensor node can collect temperature, humidity,
soil moisture and temperature, and transmit to Data Droid
modem for further processing.
Yang et al. [16] introduce a remote farm monitoring system
which consists of client block, server block and end-user
client. Client block is made from an LPC11C14 microcon-
troller, fan, light sensor, temperature and humidity sensor,
buzzer, LED and OLED display screen. The data collected
from a client block is sent to server block via ZigBee. A
server block comprises of a camera, Wi-Fi, GPRS, and ZigBee
module. The acquired data is forwarded by the server block
to an end-user client via Wi-Fi.
Ezhilazhahi et al. [17] propose an IoT-based system for
plant soil moisture monitoring. Sensor nodes of the system
collect soil moisture data and transmit the data via ZigBee
to a gateway for data aggregation and transmission. The data
then is sent to a remote server. End-users can use their mobile
phone to access the server to achieve real-time soil moisture.
Similarly, James et al. [18] present a remote monitoring system
for plant growth. The system built from Rasberry Pi collects
relative humidity, atmospheric temperature and soil moisture
and sends the data via Bluetooth to a mobile application of a
farmer.
Gayatri et al. [19] introduce a remote monitoring system for
agriculture by using a combination of IoT and cloud comput-
ing. The system relies on Wi-Fi/4G networks for making the
interconnection between sensors, gateways and cloud servers.
End-users such as farmers can use a farmer console to control
or give an instruction to sensor devices in their farms remotely.
Channe et al. [20] propose a smart system for remote
monitoring agriculture. The system is based on IoT, cloud
computing, mobile computing and bid-data analysis. The
system senses soil and environmental properties, and then
stores in cloud servers. In addition, the system processes and
analyzes the data to extra useful information such as fertilizer
requirements and best crop sequences. End-users such as
farmers can access the raw data and extracted information
stored in cloud servers remotely.
Although the mentioned works show benefits of remote
monitoring, they still have many disadvantages (e.g., energy
inefficiency) and cannot be suitable for remote monitoring
farms in remote areas where the Internet connectivity is
not covered or stable. Therefore, we propose an enhanced
architecture utilizing Edge and Fog computing, IoT, and LoRa
to overcome the mentioned problems and offer many advan-
tages services such as data processing at the edge and push
notification.
III. SYS TEM ARCHITECTURE
The system architecture is illustrated in Fig. 1. The proposed
architecture consists of 5 layers, namely sensor layer, Edge
layer, Fog layer, cloud layer, and terminal layer. The sensor
and edge layers are connected via nRF in our proposed
system design, but other wireless solutions such as Wi-Fi or
Bluetooth 5 can be utilized if they provide enough bandwidth
and range. The Edge and Fog layer are connected via LoRa,
which in open areas can enable transmissions of over 10
or even 20 km with a low data rate. The Fog gateways
are in turn connected to cloud servers via wired Ethernet
or wireless solutions with high throughput such as Wi-Fi or
mobile 4G/5G. The final layer consists of the front-end user
application and interfaces.
A. Sensor layer
The sensor layer consists of several groups of sensor nodes.
Sensor nodes and actuator nodes are part of this layer, and can
be deployed to different areas of the farm. The main difference
between these nodes is that the sensing node mainly collects
and sends the data to Edge gateways in Edge layer while
the actuating node primarily receives commands from Edge
gateway to control actuators such as turning water system. A
prototype sensor node is shown in Fig. 2, equipped with a
micro-controller, nRF wireless module, and different sensors.
Depending on the sensor node location, some of the sensor
nodes are equipped with solar panels. Most of the hardware
components of a sensor node are connected via an SPI wire
communication protocol as SPI supports high data rate with
low energy consumption.
B. Edge layer
The Edge layer is formed by Edge gateways which are
responsible for receiving data from sensor nodes via nRF,
processing data and sending the processed and compressed
data to Fog gateways via LoRa. This intermediate step helps
to significantly reduce the amount of data transmitted over
the LoRa links, and fulfill the strict requirements of LoRa’s
duty cycle. An Edge gateway mainly consists of an embedded
board with nRF and LoRa wireless modules. Some gateways
can be equipped with sensors or cameras. These gateways are
equipped with Linux because Linux offers many advantages
such as free to use, lightweight, enhanced security, customiza-
tion, reliability and high performance while Linux does not
require powerful hardware.
Edge gateway offers many advantages services such as
push notification, channel categorization and security. For
instance, when an edge gateway detects abnormalities such
as hardware failure in sensor nodes, the gateway can send
an instant message to end-users such as farmers or system
administrators to inform about the problem. The possibilities
of data processing in the Edge layer range from multi-robot
mapping [9], to bio-signal analysis [2].
An nRF protocol supports more than 100 channels in which
each channel has its own corresponding frequency (e.g., ).
This creates a premise to utilize these channels in the most
optimized way to reduce error rate or bandwidth limitations.
Some channels are more interfered by noise surrounding or
are busier due to a large amount of data transmitted by a vast
number of sensor nodes. It is recommended to use a suitable
channel or a group of channels in a specific moment. This
service can be implemented in Edge gateways.
In order to provide some levels of security, Edge gateway
can implement some algorithms including encryption algo-
rithms such as AES-128, AES-256, or cryptography algo-
rithms such as ECDSA. Although running these algorithms can
increase latency and an Edge gateway’s energy consumption,
the overhead is not significant because an Edge gateway is
often equipped with powerful hardware and uses power-line
from a power socket.
Edge gateway can run data compression or data processing
methods to reduce a large amount of data transmitted over
a LoRa network. Depending on data or the applications,
lossless or lossy data compression can be applied at Edge
gateways. Lossy compression can provide a high compression
rate e.g., 50:1 while lossless compression often offers a low
compression rate e.g., 10:1 For example, temperature data can
be applied with lossy compression as temperature is often
collected by several sensors in an area and the changing rate of
temperature is not fast in terms of minutes. Therefore, losing or
missing a sample or several samples cannot reduce a system’s
reliability. Besides the mentioned services, Edge gateway can
offer other advanced services such as interoperability, data
fusion and mobility support.
In this paper, we propose a novel integration of image
compression techniques with accumulated data from multiple
sensor nodes. Rather than using standard JPEG or other lossy
compression techniques, we have implemented a compres-
sion solution based on deep learning. Convolutional neural
networks have been increasingly used in recent years for
SENSOR LAYER
TERMINAL
nRF
nRF
EDGE LAYER
CLOUD LAYER
LoRa
FOG LAYER
LoRa Repeaters
END-USER APPLICATION
nRF
LoRa
Smart
Edge
Gateways
Actuators
Crops
Livestock
LoRa
Application Servers
Databases
Da
Data compression
Sensor node status
Da
Cloud analytics / Big Data
Cloud storage
Da
Data visualization
Sensor node management
Da
Data acquisition
Power-efficient design
Da
LoRa repeaters
Load balancers
2019_06_africon_diagram(3).drawio https://www.draw.io/
1 of 1 6/7/19, 7:34 PM
Fig. 1. Proposed 5-layer Sensor-Edge-Fog-Cloud-Terminal system architecture with LoRa wireless link used as the Edge-Fog bridge
lossy compression of images [21]–[23]. Convolutional neural
networks provide robust methods for compression with less
error after decompression when compared to traditional meth-
ods such as JPEG. We utilize a semantic perceptual image
compression method proposed by Prakash et al. [24]. Their
method is able to highlight semantically-salient regions, which
are in turn encoded with higher accuracy. We have selected this
method for IoT sensor data as it can increase the compression
rate when the data is very homogeneous, while preserving
regions of interest with proper accuracy. In order to use this
method, edge gateways first gather data from multiple sensor
nodes, and this data is encoded in the pixels of an image. For
instance, the three channels of an RGB image can be used to
encode three different variables, and the position of pixels can
represent the spatial distribution of sensor nodes.
C. Fog layer
There are two tiers in the Fog layer. The first tier consists of
an optional group of repeaters which mainly receive the data
transmitted from Edge gateways and forward the data to Fog
gateways via LoRa. The use of this tier and the specific number
of repeaters depend on the distance between the deployment
location, such as a remote farm, and the nearest Fog gateway.
For instance, according to a work from Pet¨
aj¨
aj¨
arvi et al.,
about 20 repeaters are required to maintain a stable LoRa
communication link over a distance of 300 km, in near line-
of-sign transmission [25]. In the same work, the authors warn
that even though LoRa can provide long-range transmissions
(e.g. 25 to 30 km), packet loss data increases dramatically
after the 10-15 km threshold.
The second tier in the Fog layer consists of a series of
interconnected Fog gateways which are communicated with
each other and share information about sensor nodes, whether
they are in a near location or not. If fog gateways are not
connected through a local area network, then VPNs or other
virtual network solutions can be utilized. Fog gateways are
located in areas where there exist stale and reliable Internet
connectivity. Fog gateways are connected to cloud servers
via Wi-Fi, Ethernet or 4G. Similar to Edge gateways, Fog
Fig. 2. Sensor Node Prototype
gateways can offer advanced services such as push notifica-
tion, distributed data storage, security, data fusion, and data
processing. More detailed information of Fog services and are
discussed in our previous articles [7], [8], [26]–[28].
D. Cloud layer and end-user terminal layer
Cloud layer consists of cloud servers and its services such
as global data storage, big data analysis, push notification,
data processing with complex algorithms. Depending on the
application, specific cloud services can be implemented.
End-user terminal layer consists of mobile applications and
web browser which are used to access real-time data and input
commands to control actuators in a remote farm.
IV. IMPLEMENTATION AND RES ULTS
In this paper, a complete system for remote monitoring
and controlling a farm in a remote area is implemented. In
particular, a group of sensor nodes including sensing and
actuating nodes, an Edge-gateway, a repeater, a Fog gateway,
cloud servers and end-user terminal are implemented.
A sensing node is implemented with an nRF module
nRF24L01, a microcontroller AVR ATmega8, a group of sen-
sors (i.e., DHT111 temperature and humidity sensor and soil
sensor). Some of the sensor nodes can use the power socket
while some can be equipped with solar panel and battery.
Depending on the placement of the sensing nodes, one of the
mentioned methods is applied for providing power for sensing
nodes. The sensing node mainly senses and transmits the
collected data to an Edge gateway. Therefore, it is programmed
to sleep in most of the time to save power except for a moment
when it senses and transmit data. When it is working, the radio
receiving part is turned off to save power.
An actuating node is also implemented with an nRF module
nRF24L01, a microcontroller AVR ATmega8, a relay or a
controlling circuit to turn on/off or control the actuator such
as a water system. Different from a sensing node, an actuating
node is powered by a power socket and its radio receiving part
is turned on always to wait for the command from a farmer
or a system administrator.
The Edge gateway is implemented with a Rasberry Pi
v3B, which has quad-core 1.4 GHz CPU and 1G RAM.
The Rasberry Pi is connected with nRF24L01+PA+LNA for
receiving data from sensor nodes via nRF. Comparing with
an nRF24L01 module having a printed onboard antenna, an
nRF24L01+ module has a long antenna. This long antenna
helps to increase receiving power at the receiver end although
it also increases energy consumption of the Edge gateway.
The high energy consumption of an edge gateway is not a
problematic issue as edge gateways are often powered by a
power socket. In addition, the Rasberry Pi is connected with
a LoRa chip (i.e, Dragino hat including LoRa chip and GPS)
for sending the processed data to repeaters.
In this paper, a repeater is a simple LoRa gateway which
receives data from several Edge-gateways and forwards the
data to Fog gateways via LoRa. In future work, a repeater can
be more investigated to perform some other services such as
data compression, and data processing.
Fog gateway is implemented with a LoRa gateway and an
Intel UP gateway. Fog gateway is equipped with Ethernet,
Wi-Fi and 4G module for connecting to the Internet and
interconnecting between each other. In nominal operation,
Ethernet is preferable when it is available. Otherwise, Wi-Fi
and 4G are used. Fog gateway is powered by a power socket.
Therefore, power consumption is not an issue. In this paper,
we reuse and adapt fog services, cloud applications, and end-
user terminals which have been implemented in our previous
papers [7], [8], [26]–[28]. The user interface has been modified
for data visualization.
A. Sensor Node
The implemented sensor node collects and transmits data
(i.e., temperature, humidity, or soil moisture) with a data rate
of 1 sample per minute to Edge gateways which are about
200 meters. This data rate is sufficient for farm applications
as temperature, humidity, soil moisture and other similar data
do not change frequently in terms of minutes. The sensor
node is supplied with 3.3V and its energy consumption is
measured with a power monitor tool MonSoon [29]. The
energy consumption of the sensor node per minute is 138.92
mJ. In addition, we also test the transmission rate of the sensor
node by sending more than 1000 samples with line-of-sight.
The result shows that the success rate is larger than 99% and
less than 5 samples are collapsed. If we use a 2000 mAh
battery (5.6mm x 49.2mm x 68.8mm), a sensor node can work
up to 863 hours. When a small solar panel (145mm x 145mm
x 2mm) is used at noon with Finland sunlight, we can harvest
around 1650 mJ per second. This number should be much
larger if the solar panel is placed at Africa where sunlight is
much more powerful at noon. With the solar panel and an
average of 4 hours per day in Africa, a sensor node’s battery
can fully be charged. Lithium batteries often allow for 400 to
500 charging cycles. Therefore, a sensor node with the solar
panel can be used for the same number of days.
For verifying the functionality of the proposed system, we
use a Web browser to access data including temperature,
humidity and soil moisture. The results show that the system
works properly. For example, temperature and relative humid-
ity of the experimented land is shown (i.e., 20 degrees Celsius
and 57%, respectively) in real-time. In addition, we also send
some commands via a Web interface to control a water system
of the experimented land.
B. Data Compression
In order to validate the proposed compression method, we
have simulated the generation of 16384 samples, each with
3 8-bit channels. These channels can be used, for instance,
for temperature, humidity and air quality data. The data is
then saved as a 128x128 image, which is compressed using
JPEG format. Data can be structured such that neighbor
pixels in the image correspond to neighbor sensor nodes
when taking into account their spatial distribution. Instead
of sending information about individual sensor nodes, edge
gateways gather data from all of them and then merge the
data into an image where spatial information is encoded. The
correspondence between sensor node identifiers and image
pixels must be done either beforehand or through an initial
message when the edge gateways are configured.
Figure 3 shows an example of image compression using the
semantic perceptual image compression for simulated random
data with equal mean and variance through the image. A
different method has been used to generate simulated data in
Figure 4, where the image has been divided in four quadrants
of 64x64 pixels each and different mean and variance have
been applied. In a real scenario, the first approach simulates a
field with only one kind of crops and homogeneous conditions,
where the second approach simulates a field with multiple
crops or different soil conditions. In both cases, sub-figure
(a) represents the random data, while (b) illustrates the map
generated by the method from [24], and (c) shows the overlay
of the original image and the map.
The compression error and reduction of data batch size is
shown in Tables I and II, where we can see that for 57%
compression rate the error amounts to only 2%, a reduction
of over 40% when compared to standard JPEG compression.
We can also see that error rates decrease when there is more
information added to the image. This is particularly beneficial
for detecting abnormalities with higher precision due to the
semantic compression method utilized.
(a) (b) (c)
Fig. 3. Illustration of simulated data with constant distribution.
TABLE I
COMPRESSION COMPARISON IN RANDOM NOISE
Compression Error Size reduction
JPEG-50% 3.73% 57%
JPEG-30% 6.17% 67%
CNN+JPEG-50% 2.53% 58%
CNN+JPEG-30% 4.55% 67%
V. CONCLUSION AND FUTURE WORK
We have presented a hybrid 5-layer architecture for IoT
systems deployed in smart farms. The proposed architecture
consists of sensor, Edge, Fog, cloud and terminal layers.
This architecture can be used to multiply the options of IoT
deployments with LoRa or other LPWAN technologies that
provide low-power and long-range transmission but limited
data rate. We propose different methods for compressing data
and reducing the LPWAN link load, minimizing the amount of
data transmitted and stored in cloud servers while maintaining
a similar user experience. In particular, we propose a novel
integration of image compression algorithms to be applied in
IoT data from the agricultural and farming industries, such as
temperature, humidity, air quality or soil properties. In future
work, we will focus on deploying multiple sensor nodes in
a real scenario and provide a wider set of compression and
analytics algorithms for the edge layer.
REFERENCES
[1] S. Tibken. In farm country, forget broadband. you might not have
internet at all. Accessed: Jun, 2019, Updated: Oct, 2018, Available:
https://cnet.co/2AmEWgL/.
[2] J. Pe˜
na Queralta et al. EdgeAI in LoRabased healthcare monitoring:
A case study on fall detection system with LSTM Recurrent Neural
Networks. In IEEE 42nd TSP, 2019.
[3] T. N. Gia et al. Low-cost fog-assisted health-care iot system with energy-
efficient sensor nodes. In 2017 IWCMC, pages 1765–1770. IEEE, 2017.
[4] I. Tcarenko et al. Energy-efficient iot-enabled fall detection system with
messenger-based notification. In International Conference on Wireless
Mobile Communication and Healthcare, pages 19–26. Springer, 2016.
[5] M. Jiang et al. Iot-based remote facial expression monitoring system
with semg signal. In 2016 IEEE SAS. IEEE, 2016.
[6] T. N. Gia et al. Iot-based continuous glucose monitoring system: A
feasibility study. Procedia Computer Science, 109:327–334, 2017.
[7] T. N. Gia and M. Jiang. Exploiting fog computing in health monitoring.
Fog and Edge Computing: Principles and Paradigms, 2019.
[8] A. M. Rahmani et al. Exploiting smart e-health gateways at the edge
of healthcare internet-of-things: A fog computing approach. Future
Generation Computer Systems, 78:641–658, 2018.
[9] V. K. Sarker et al. Offloading slam for indoor mobile robots with edge-
fog-cloud computing. In ICASERT, 2019.
[10] J. Pe ˜
na Queralta et al. Comparative study of LPWAN technologies on
unlicensed bands for M2M communication in the IoT: beyond LoRa
and LoRaWAN. In 14th FNC, 2019.
(a) (b) (c)
Fig. 4. Simulated data with four different distributions.
TABLE II
COMPRESSION COMPARISON IN 4-QUADRA NT NOI SE
Compression Error Size reduction
JPEG-50% 3.41% 56%
JPEG-30% 5.45% 67%
CNN+JPEG-50% 2.08% 57%
CNN+JPEG-30% 4.37% 67%
[11] V. K. Sarker et al. A survey on lora for iot: Integrating edge computing.
In SLICE, 2019.
[12] R. F. Maia et al. Precision agriculture using remote monitoring systems
in brazil. In IEEE GHTC. IEEE, 2017.
[13] H. F. Ragai et al. Remote control and monitoring of fish farms using
wireless sensor networks. In 12th ICCES. IEEE, 2017.
[14] S. B. Saraf and D. H. Gawali. Iot based smart irrigation monitoring and
controlling system. In 2017 RTEICT, pages 815–819. IEEE, 2017.
[15] A. Shaout et al. An embedded system for agricultural monitoring of
remote areas. In 11th ICENCO, pages 58–67. IEEE, 2015.
[16] S. Yang, S. Tong, and L. Liang. Remote farm environment monitoring
system based on embedded system and zigbee technology. In 2015
ICSPCC, pages 1–5. IEEE, 2015.
[17] A. M. Ezhilazhahi, and P. T. V. Bhuvaneswari. Iot enabled plant soil
moisture monitoring using wireless sensor networks. In 2017 ICSSS,
pages 345–349. IEEE, 2017.
[18] J. James et al. Plant growth monitoring system, with dynamic user-
interface. In 2016 IEEE Region 10 Humanitarian Technology Conference
(R10-HTC), pages 1–5. IEEE, 2016.
[19] M. K. Gayatri et al. Providing smart agricultural solutions to farmers
for better yielding using iot. In 2015 IEEE Technological Innovation in
ICT for Agriculture and Rural Development (TIAR). IEEE, 2015.
[20] H. Channe, S. Kothari, and D. Kadam. Multidisciplinary model
for smart agriculture using internet-of-things (iot), sensors, cloud-
computing, mobile-computing & big-data analysis. Int. J. Computer
Technology & Applications, 6(3):374–382, 2015.
[21] M. Li et al. Learning convolutional networks for content-weighted image
compression. In Proceedings of the IEEE Conference on Computer
Vision and Pattern Recognition, pages 3214–3223, 2018.
[22] B. Bayar, and M. C. Stamm. On the robustness of constrained convolu-
tional neural networks to jpeg post-compression for image resampling
detection. In 2017 IEEE International Conference on Acoustics, Speech
and Signal Processing (ICASSP), pages 2152–2156. IEEE, 2017.
[23] L. Cavigelli et al. Cas-cnn: A deep convolutional neural network for
image compression artifact suppression. In 2017 International Joint
Conference on Neural Networks (IJCNN), pages 752–759. IEEE, 2017.
[24] A. Prakash et al. Semantic perceptual image compression using deep
convolution networks. pages 250–259, 04 2017.
[25] J. Petajajarvi et al. On the coverage of lpwans: range evaluation and
channel attenuation model for lora technology. In 2015 ITST, pages
55–59. IEEE, 2015.
[26] B. Negash et al. Leveraging fog computing for healthcare iot. In Fog
Computing in the Internet of Things, pages 145–169. Springer, 2018.
[27] T. N. Gia et al. Fog computing in body sensor networks: An energy
efficient approach. In Proc. IEEE BSN, 2015.
[28] T. N. Gia et al. Fog computing approach for mobility support in internet-
of-things systems. IEEE Access, 6:36064–36082, 2018.
[29] Monsson Solutions. High voltage power monitor. Accessed: June,2019,
Available: https://www.msoon.com/online-store.
... As can be seen in Table 10, the raspberry pi microprocessor is the most used with 29%, of a total of scientific articles after fine granulation. Out of the 94 articles surveyed, 58 (61.05%) do not mention the type of controller to be used Raspberry PI [19,23,24,40,52,57,60,61,63,64,66,71,72,75,84,92,98,101,106,107,119] Arduino [19, 20, 47, 50, 56, 57, 60-64, 67, 72, 75, 93-95, 98, 100, 106, 119] ESP [20,47,50,55,58,59,62,65,93,95,100,111] for deploying AI on IoT devices. This indicates a lack of emphasis on the importance of choosing the appropriate hardware controller for the intended use case. ...
... Other areas of research in Edge AI for IoT include integrating image compression algorithms in IoT data from agriculture and agricultural industries, modeling collaborative and distributed heterogeneous operating systems, and deploying AI on devices in a smart home context [63,87,88]. There is also an interest in performing experiments on real fish farms, improving classification accuracies and deploying the model on ESP32 while integrating connectivity media such as LoRaWan and SIM Card Router [47]. ...
Preprint
Full-text available
The Internet of Things (IoT) and Artificial Intelligence (AI) are two technologiesthat have long been applied to the development of smart systems. These systemscover various areas, such as smart cities, energy management, autonomous cars,etc. Intelligence, autonomy, and real-time monitoring are the fundamental ele-ments that characterize these application areas. The convergence of multi-agentsystems (MAS), IoT, and AI has opened up new possibilities for creating smarterbuildings and cities. This integration leverages the power of MAS to enable intelli-gent communication and collaboration among various entities, while IoT providesa vast network of interconnected sensors and devices that collect and transmitreal-time data. On the other hand, AI algorithms process and analyze this data toderive valuable insights and make informed decisions. Multi-agent systems tech-nology is one of the most used for the implementation of such intelligent systems.This article aims to focus on multi-agent systems, although it also covers otherrelated topics by carrying out a Systematic Literature Review (SLR) on relevantresearch works in the previous ten years. Eleven research questions are consideredat the beginning of the review, including typical research topics and applicationdomains. From the SLR results the research directions are: (i) Development ofa methodology that shows how to integrate the different applications indepen-dent of the scenarios that are deployed in. Additionally, elaboration of the toolsused in the integration process; (ii) Deployment of an agent in a microprocessor; (iii) How to implement and connect Multi-agent systems (MAS) technology andInternet of Things (IoT) devices (processors, controllers, sensors, and actuators).
... Regarding the distributed applications of LoRa, most of the existing literature (e.g., [13,14]) focuses on utilizing deep learning techniques for processing the collected data on LoRa gateways and end devices, with applications in areas such as fall detection and agriculture. However, there is relatively little research on the massive uplink end device access and multi-task downlink publication to end devices using LoRa. ...
Article
Full-text available
As a narrowband communication technology, long-range (LoRa) contributes to the long development of Internet of Things (IoT) applications. The LoRa gateway plays an important role in the IoT transport layer, and security and efficiency are the key issues of the current research. In the centralized working model of IoT systems built by traditional LoRa gateways, all the data generated and reported by end devices are processed and stored in cloud servers, which are susceptible to security issues such as data loss and data falsification. Edge computing (EC), as an innovative approach that brings data processing and storage closer to the endpoints, can create a decentralized security infrastructure for LoRa gateway systems, resulting in an EC-assisted IoT working model. Although this paradigm delivers unique features and an improved quality of service (QoS), installing IoT applications at LoRa gateways with limited computing and memory capabilities presents considerable obstacles. This article proposes the design and implementation of an “EC-assisted LoRa gateway” using edge computing. Our proposed latency-aware algorithm (LAA) can greatly improve the reliability of the network system by using a distributed edge computing network technology that can achieve maintenance operations, such as detection, repair, and replacement of failures of edge nodes in the network. Then, an EC-assisted LoRa gateway prototype was developed on an embedded hardware system. Finally, experiments were conducted to evaluate the performance of the proposed EC-assisted LoRa gateway. Compared with the conventional LoRa gateway, the proposed edge intelligent LoRa gateway had 41.1% lower bandwidth utilization and handled more end devices, ensuring system availability and IoT network reliability more effectively.
... New approaches to agriculture are needed to meet this growing demand for food sustainably. This has led many governments and international organizations to invest heavily in food, agriculture, and related technologies [3][4][5]. However, the development of these industries has also created a range of security challenges. ...
Preprint
Full-text available
p>This study describes several critical security conditions and requirements, as well as security threats, related to smart farming and precision agriculture environments. </p
... New approaches to agriculture are needed to meet this growing demand for food sustainably. This has led many governments and international organizations to invest heavily in food, agriculture, and related technologies [3][4][5]. However, the development of these industries has also created a range of security challenges. ...
Preprint
Full-text available
p>This study describes several critical security conditions and requirements, as well as security threats, related to smart farming and precision agriculture environments. </p
Chapter
In order to meet the unique requirements of latency-sensitive applications like augmented reality and industrial IoT, which generate enormous amounts of data that are prohibitive to transport to distant cloud data centers for processing, fog computing is increasingly integrated in the IoT and CPS systems. Simultaneously, the integration of fog computing in these systems also opened up potential possibilities for developing a larger range of applications, each of which places unique demands on developing fog computing platforms. This chapter aims at conducting a survey relating to these important applications that have been studied and proposed in the literature.
Article
Full-text available
High cost, long-range communication, and anomaly detection issues are associated with IoT systems in water quality monitoring. Therefore, this work proposes a prototype for a water quality monitoring system (IoT-WQMS) based on IoT technologies, which include in the system architecture a LoRa repeater and an anomaly detection algorithm. The system performs the data collection, data storage, anomaly detection, and alarm sending remotely and in real-time for the information to be captured by the multisensor node. The LoRa repeater allowed the spatial coverage of the LoRa communication to extend, making it possible to reach a place where originally there was no coverage with a single LoRa transmitter due to topography and line of sight. The prototype performed well in terms of packet loss rate, transmission time, and sensitivity, extending the long-range wireless communication distance. Indoor multinode testing validation for 29 days of the mean absolute error for average relative errors of water temperature, pH, turbidity, and total dissolved solids (TDS) were 0.65%, 0.30%, and 14.33%, respectively. The anomaly detector identified all erroneous data events due to node sensor recalibration and water recirculation pump failures. The IoT-WQMS increased the reliability of monitoring through the timely identification of any sensor malfunctions and extended the LoRa signal range, which are relevant features in the scope of in situ and real-time water quality monitoring.
Article
Artificial Intelligence (AI) at the edge is the utilization of AI in real-world devices. Edge AI refers to the practice of doing AI computations near the users at the network's edge, instead of centralised location like a cloud service provider's data centre. With the latest innovations in AI efficiency, the proliferation of Internet of Things (IoT) devices, and the rise of edge computing, the potential of edge AI has now been unlocked. This study provides a thorough analysis of AI approaches and capabilities as they pertain to edge computing, or Edge AI. Further, a detailed survey of edge computing and its paradigms including transition to Edge AI is presented to explore the background of each variant proposed for implementing Edge Computing. Furthermore, we discussed the Edge AI approach to deploying AI algorithms and models on edge devices, which are typically resource-constrained devices located at the edge of the network. We also presented the technology used in various modern IoT applications, including autonomous vehicles, smart homes, industrial automation, healthcare, and surveillance. Moreover, the discussion of leveraging machine learning algorithms optimized for resource-constrained environments is presented. Finally, important open challenges and potential research directions in the field of edge computing and edge AI have been identified and investigated. We hope that this article will serve as a common goal for a future blueprint that will unite important stakeholders and facilitates to accelerate development in the field of Edge AI.
Article
Full-text available
In recent times, the Internet of Things (IoT) applications, including smart transportation, smart healthcare, smart grid, smart city, etc. generate a large volume of real-time data for decision making. In the past decades, real-time sensory data have been offloaded to centralized cloud servers for data analysis through a reliable communication channel. However, due to the long communication distance between end-users and centralized cloud servers, the chances of increasing network congestion, data loss, latency, and energy consumption are getting significantly higher. To address the challenges mentioned above, fog computing emerges in a distributed environment that extends the computation and storage facilities at the edge of the network. Compared to centralized cloud infrastructure, a distributed fog framework can support delay-sensitive IoT applications with minimum latency and energy consumption while analyzing the data using a set of resource-constraint fog/edge devices. Thus our survey covers the layered IoT architecture, evaluation metrics, and applications aspects of fog computing and its progress in the last four years. Furthermore, the layered architecture of the standard fog framework and different state-of-the-art techniques for utilizing computing resources of fog networks have been covered in this study. Moreover, we included an IoT use case scenario to demonstrate the fog data offloading and resource provisioning example in heterogeneous vehicular fog networks. Finally, we examine various challenges and potential solutions to establish interoperable communication and computation for next-generation IoT applications in fog networks.
Article
Full-text available
Low power wide area networks (LPWAN) are widely used in IoT applications as they offer low power consumption and long-range communication. LoRaWAN and SigFox have taken the top positions in the unlicensed ISM bands, while LTE-M and NB-IoT have emerged within cellular networks. We focus on unlicensed bands operation because of their availability for both private and public use with one's own infrastructure. New technologies have since been developed to overcome limitations of LoRaWAN and SigFox, based on LoRa or other modulation techniques, and are finding their way mainly into the industrial IoT. These include Symphony Link or Ingenu RPMA. To the best of our knowledge, previous works have not been focused on comparing LPWAN technologies in-depth including alternatives to the link and network layers over LoRa other than LoRaWAN. This paper provides a detailed comparative study of these technologies and potential application scenarios. We defend that LoRaWAN is the most suitable for small-scale or public deployments, while Symphony Link provides a robust solution for industrial environments. SigFox is one the most widely deployed networks; and RPMA has the advantage of using the 2.4GHz band, equally regulated in most countries.
Conference Paper
Full-text available
Low power wide area networks (LPWAN) are widely used in IoT applications as they offer low power consumption and long-range communication. LoRaWAN and SigFox have taken the top positions in the unlicensed ISM bands, while LTE-M and NB-IoT have emerged within cellular networks. We focus on unlicensed bands operation because of their availability for both private and public use with one's own infrastructure. New technologies have since been developed to overcome limitations of LoRaWAN and SigFox, based on LoRa or other modulation techniques, and are finding their way mainly into the industrial IoT. These include Symphony Link or Ingenu RPMA. To the best of our knowledge, previous works have not been focused on comparing LPWAN technologies in-depth including alternatives to the link and network layers over LoRa other than LoRaWAN. This paper provides a detailed comparative study of these technologies and potential application scenarios. We defend that LoRaWAN is the most suitable for small-scale or public deployments, while Symphony Link provides a robust solution for industrial environments. SigFox is one the most widely deployed networks; and RPMA has the advantage of using the 2.4GHz band, equally regulated in most countries.
Conference Paper
Full-text available
Remote healthcare monitoring has exponentially grown over the past decade together with the increasing penetration of Internet of Things (IoT) platforms. IoT-based health systems help to improve the quality of healthcare services through real-time data acquisition and processing. However, traditional IoT architectures have some limitations. For instance, they cannot properly function in areas with poor or unstable Internet. Low power wide area network (LPWAN) technologies, including long-range communication protocols such as LoRa, are a potential candidate to overcome the lacking network infrastructure. Nevertheless, LPWANs have limited transmission bandwidth not suitable for high data rate applications such as fall detection systems or electrocardiography monitoring. Therefore, data processing and compression are required at the edge of the network. We propose a system architecture with integrated artificial intelligence that combines Edge and Fog computing, LPWAN technology, IoT and deep learning algorithms to perform health monitoring tasks. In particular, we demonstrate the feasibility and effectiveness of this architecture via a use case of fall detection using recurrent neural networks. We have implemented a fall detection system from the sensor node and Edge gateway to cloud services and end-user applications. The system uses inertial data as input and achieves an average precision of over 90\% and an average recall over 95\% in fall detection.
Conference Paper
Full-text available
Indoor mobile robots are widely used in industrial environments such as large logistic warehouses. They are often in charge of collecting or sorting products. For such robots, computation-intensive operations account for a significant percentage of the total energy consumption and consequently affect battery life. Besides, in order to keep both the power consumption and hardware complexity low, simple micro-controllers or single-board computers are used as onboard local control units. This limits the computational capabilities of robots and consequently their performance. Offloading heavy computation to Cloud servers has been a widely used approach to solve this problem for cases where large amounts of sensor data such as real-time video feeds need to be analyzed. More recently, Fog and Edge computing are being leveraged for offloading tasks such as image processing and complex navigation algorithms involving non-linear mathematical operations. In this paper, we present a system architecture for offloading computationally expensive localization and mapping tasks to smart Edge gateways which use Fog services. We show how Edge computing brings computational capabilities of the Cloud to the robot environment without compromising operational reliability due to connection issues. Furthermore, we analyze the power consumption of a prototype robot vehicle in different modes and show how battery life can be significantly improved by moving the processing of data to the Edge layer.
Conference Paper
Full-text available
Increased automation and intelligence in computer systems have revealed limitations of Cloud-based computing such as unpredicted latency in safety-critical and performance-sensitive applications. The amount of data generated from ubiquitous sensors has reached a degree where it becomes impractical to always store and process in the Cloud. Edge computing brings computation and storage to the Edge of the network near to where the data originates yielding reduced network load and better performance of services. In parallel, new wireless communication technologies have appeared to facilitate the expansion of Internet of Things (IoT). Instead of seeking higher data rates, low-power wide-area network aims at battery-powered sensor nodes and devices which require reliable communication for a prolonged period of time. Recently, Long Range (LoRa) has become a popular choice for IoT-based solutions. In this paper, we explore and analyze different application fields and related works which use LoRa and investigate potential improvement opportunities and considerations. Furthermore, we propose a generic architecture to integrate Edge computation capability in IoT-based applications for enhanced performance.
Article
Full-text available
Handover mechanism for mobility support in a remote real-time streaming IoT system was proposed in this paper. The handover mechanism serves to keep the connection between sensor nodes and a gateway with a low latency. The handover mechanism also attentively considers oscillating nodes which often occur in many streaming IoT systems. By leveraging the strategic position of smart gateways and Fog computing in a real-time streaming IoT system, sensor nodes’ loads were alleviated whereas advanced services, like push notification and local data storage, were provided. The paper discussed and analyzed metrics for the handover mechanism based on Wi-Fi. In addition, a complete remote real-time health monitoring IoT system was implemented for experiments. The results from evaluating our mobility handover mechanism for mobility support shows that the latency of switching from one gateway to another is 10% - 50% less than other state-of-the-art mobility support systems. The results show that the proposed handover mechanism is a very promising approach for mobility support in both Fog computing and IoT systems.
Research
Full-text available
Although precision agriculture has been adopted in few countries; the agriculture industry in India still needs to be modernized with the involvement of technologies for better production, distribution and cost control. In this paper we proposed a multidisciplinary model for smart agriculture based on the key technologies: Internet-of-Things (IoT), Sensors, Cloud-Computing, Mobile-Computing, Big-Data analysis. Farmers, Agro-Marketing agencies and Agro-Vendors need to be registered to the AgroCloud module through MobileApp module. AgroCloud storage is used to store the details of farmers, periodic soil properties of farmlands, agro-vendors and agro-marketing agencies, Agro e-governance schemes and current environmental conditions. Soil and environment properties are sensed and periodically sent to AgroCloud through IoT (Beagle Black Bone). Bigdata analysis on AgroCloud data is done for fertilizer requirements, best crop sequences analysis, total production, and current stock and market requirements. Proposed model is beneficial for increase in agricultural production and for cost control of Agro-products.
Chapter
This chapter exploits fog computing in health‐monitoring Internet‐of‐Things (IoT) systems for enhancing the quality of healthcare service. It shows an overview of the architecture of an IoT‐based system with fog computing. Fog computing services locating in a fog layer of smart gateways are diversified for serving IoT applications. The chapter discusses the fog computing services in smart e‐health gateways. The health‐monitoring IoT system consists of several wearable sensor nodes, smart gateways with fog services, cloud servers, and terminals. The chapter discusses detailed implementations of these components. It provides a case study, experimental results, and evaluation related to heart rate variability (HRV) analysis. The chapter presents the related applications in fog computing and discusses future research directions. Fog computing demonstrates that it is one of the most suitable candidates for augmenting IoT systems in healthcare and other domains.