No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
Sparsely Connected Low Complexity CNN for Unmanned Vehicles Detection - Sensing RF Signal
Abstract
Unmanned aerial systems, namely drones, have greatly improved and expanded drastically over the years. Due to their efficiency and ease of use, drones have been utilized in a wide range of applications. Despite various potential uses, drones are also being utilized for illegal operations and exposing security threats to citizens. It is vital to install an effective anti-drone system to identify and defend against intruding malevolent drones to protect national security. Although there has been tremendous advancement in the development of machine learning to deploy lightweight architectures in the sensor industry, no such drone detection approach has yet been described in the literature. Therefore, this paper proposed a lightweight convolution neural network (CNN), namely RFDNet, to investigate the problem of 17 types of drone RF fingerprint classification problems in the low SNR regime. The network is configured with two principle modules, which are leveraged by the grouped and depth-wise convolution layers, incorporating accuracy improvement while keeping the complexity low. Notably, most existing networks fail to outstandingly detect drones at low SNR levels because the RF signal envelope is distorted and the transient information is lost in the noise. To solve this issue, we collected an open-source RF dataset that stores 17 types of drone RF signals at 30 dB SNR. To investigate the RF dataset at various SNR levels, we regenerate the dataset at different SNRs (i.e.,
dB to 30 dB SNR with the 5 dB interval) and analyze the performance of the proposed network. The empirical results show that RFDNet performed outstandingly compared to the existing deep learning-based drone detection methods and achieved an overall 99.07% accuracy at 15 dB signal-to-noise ratio (SNR).
... Additionally, the angle of azimuth and elevation were computed from channel state information (CSI) to position drones [28]. Utilizing timefrequency images (TFI) can also achieve drone RID [29], [30], [31], signal detection [32], and denoising [33], [34] modules can also be introduced integrated into TFI-based drones RID frameworks. ...
We propose an algorithm based on Zadoff-Chu (ZC) sequences and time-frequency images (TFI) to achieve drone remote identification (RID). Specifically, by analyzing the modulation parameters and frame structures of drone ratio-frequency (RF) signals in the DroneRFa dataset, we extract prior information about ZC sequences with surprising correlation properties and robustness. Cross-correlation is performed between locally generated ZC sequences and drone signals to derive ZC sequence-based features. Then, these ZF sequence features are fused with TFI features containing communication protocol information to achieve drone RID. To reduce computational costs, data reduction of the cross-correlation features is performed by analyzing the frame structures and modulation parameters, ensuring that the feature performance remained unaffected. Three feature fusion methods, namely probability-weighted addition, feature vector addition, and feature vector concatenation, are analyzed. Simulation results demonstrate that the proposed algorithm improves the average accuracy by at least 2.5\% compared to existing methods, which also indicate robust RID performance under burst interference and background noise. For RF sampling signals at varying flight distances, the proposed algorithm achieves a maximum accuracy of 99.11\%.
Identifying the physical aspect of the earth’s surface (Land cover) and also how we exploit the land (Land use) is a challenging problem in environment monitoring and much of other subdomains. One of the most efficient ways to do this is through Remote Sensing (analyzing satellite images). For such classification using satellite images, there exist many algorithms and methods, but they have several problems associated with them, such as improper feature extraction, poor efficiency, etc. Problems associated with established land-use classification methods can be solved by using various optimization techniques with the Convolutional neural networks(CNN). The structure of the Convolutional neural network model is modified to improve the classification performance, and the overfitting phenomenon that may occur during training is avoided by optimizing the training algorithm. This work mainly focuses on classifying land types such as forest lands, bare lands, residential buildings, Rivers, Highways, cultivated lands, etc. The outcome of this work can be further processed for monitoring in various domains.
The use of unmanned aerial vehicles (UAVs) is growing rapidly across many civil application domains, including real-time monitoring, providing wireless coverage, remote sensing, search and rescue, delivery of goods, security and surveillance, precision agriculture, and civil infrastructure inspection. Smart UAVs are the next big revolution in the UAV technology promising to provide new opportunities in different applications, especially in civil infrastructure in terms of reduced risks and lower cost. Civil infrastructure is expected to dominate more than $45 Billion market value of UAV usage. In this paper, we present UAV civil applications and their challenges. We also discuss the current research trends and provide future insights for potential UAV uses. Furthermore, we present the key challenges for UAV civil applications, including charging challenges, collision avoidance and swarming challenges, and networking and security-related
challenges. Based on our review of the recent literature, we discuss open research challenges and draw high-level insights on how these challenges might be approached.
Unmanned Aerial Vehicles (UAVs), or drones, which can be considered as a coverage extender for Internet of Everything (IoE), have drawn high attention recently. The proliferation of drones will raise privacy and security concerns in public. This paper investigates the problem of classification of drones from Radio Frequency (RF) fingerprints at the low Signal-to-Noise Ratio (SNR) regime. We use Convolutional Neural Networks (CNNs) trained with both RF time-series images and the spectrograms of 15 different off-the-shelf drone controller RF signals. When using time-series signal images, the CNN extracts features from the signal transient and envelope. As the SNR decreases, this approach fails dramatically because the information in the transient is lost in the noise, and the envelope is distorted heavily. In contrast to time-series representation of the RF signals, with spectrograms, it is possible to focus only on the desired frequency interval, i.e., 2.4 GHz ISM band, and filter out any other signal component outside of this band. These advantages provide a notable performance improvement over the time-series signals-based methods. To further increase the classification accuracy of the spectrogram-based CNN, we denoise the spectrogram images by truncating them to a limited spectral density interval. Creating a single model using spectrogram images of noisy signals and tuning the CNN model parameters, we achieve a classification accuracy varying from 92% to 100% for an SNR range from -10 dB to 30 dB, which significantly outperforms the existing approaches to our best knowledge.
Solar irradiance prediction is an indispensable area of the photovoltaic (PV) power management system. However, PV management may be subject to severe penalties due to the unsteadiness pattern of PV output power that depends on solar radiation. A high-precision long short-term memory (LSTM)-based neural network model named SIPNet to predict solar irradiance in a short time interval is proposed to overcome this problem. Solar radiation depends on the environmental sensing of meteorological information such as temperature, pressure, humidity, wind speed, and direction, which are different dimensions in measurement. LSTM neural network can concurrently learn the spatiotemporal of multivariate input features via various logistic gates. Moreover, SIPNet can estimate the future solar irradiance given the historical observation of the meteorological information and the radiation data. The SIPNet model is simulated and compared with the actual and predicted data series and evaluated by the mean absolute error (MAE), mean square error (MSE), and root MSE. The empirical results show that the value of MAE, MSE, and root mean square error of SIPNet is 0.0413, 0.0033, and 0.057, respectively, which demonstrate the effectiveness of SIPNet and outperforms other existing models.
ABSTRACT
This dataset contains RF signals from drone remote controllers (RCs) of different makes and models. The RF signals transmitted by the drone RCs to communicate with the drones are intercepted and recorded by a passive RF surveillance system, which consists of a high-frequency oscilloscope, directional grid antenna, and low-noise power amplifier. The drones were idle during the data capture process. All the drone RCs transmit signals in the 2.4 GHz band. There are 17 drone RCs from eight different manufacturers and ~1000 RF signals per drone RC, each spanning a duration of 0.25 ms.
Instructions:
The dataset contains ~1000 RF signals in .mat format from the remote controllers (RCs) of the following drones:
DJI (5): Inspire 1 Pro, Matrice 100, Matrice 600*, Phantom 4 Pro*, Phantom 3
Spektrum (4): DX5e, DX6e, DX6i, JR X9303
Futaba (1): T8FG
Graupner (1): MC32
HobbyKing (1): HK-T6A
FlySky (1): FS-T6
Turnigy (1): 9X
Jeti Duplex (1): DC-16.
In the dataset, there are two pairs of RCs for the drones indicated by an asterisk above, making a total of 17 drone RCs. Each RF signal contains 5 million samples and spans a time period of 0.25 ms.
The scripts provided with the dataset defines a class to create drone RC objects and creates a database of objects as well as a database in table format with all the available information, such as make, model, raw RF signal, sampling frequency, etc. The scripts also include functions to visualize data and extract a few example features from the raw RF signal (e.g., transient signal start point). Instructions for using the scripts are included at the top of each script and can also be viewed by typing help scriptName in MATLAB command window.
The drone RC RF dataset was used in the following papers:
M. Ezuma, F. Erden, C. Kumar, O. Ozdemir, and I. Guvenc, "Micro-UAV detection and classification from RF fingerprints using machine learning techniques," in Proc. IEEE Aerosp. Conf., Big Sky, MT, Mar. 2019, pp. 1-13.
M. Ezuma, F. Erden, C. K. Anjinappa, O. Ozdemir, and I. Guvenc, "Detection and classification of UAVs using RF fingerprints in the presence of Wi-Fi and Bluetooth interference," IEEE Open J. Commun. Soc., vol. 1, no. 1, pp. 60-79, Nov. 2019.
E. Ozturk, F. Erden, and I. Guvenc, "RF-based low-SNR classification of UAVs using convolutional neural networks." arXiv preprint arXiv:2009.05519, Sept. 2020.
Other details regarding the dataset and data collection and processing can be found in the above papers and attached documentation.
############
Author Contributions:
Experiment design: O. Ozdemir and M. Ezuma
Data collection: M. Ezuma
Scripts: F. Erden and C. K. Anjinappa
Documentation: F. Erden
Supervision, revision, and funding: I. Guvenc
############
Acknowledgment
This work was supported in part by NASA through the Federal Award under Grant NNX17AJ94A, and in part by NSF under CNS-1939334 (AERPAW, one of NSF's Platforms for Advanced Wireless Research (PAWR) projects).
Recently, Unmanned Aerial Vehicles (UAV)'s, also known as drone, are becoming rapidly popular due to the advancement of their technology and the significant decrease in their cost. Although, commercial drones have proven their effectiveness in many applications such as cinematography, agriculture monitoring, search and rescue, building 3D inspection, and many other day to day applications, they have also been alarmingly used in malicious activities that are targeting to harm individuals and societies which raises great privacy, safety and security concerns. Hence, in this research, we propose a new solution drone detection solution based on the radio frequency emitted during the live communication session between the drone and its controller using the convolutional neural network (CNN). The results of the study have proven the effectiveness of using CNN for drone detection with accuracy and F1 score of over 99.7% and drone identification with accuracy and F1 score of 88.4%. Moreover, the results yielded from this experiment have outperformed those reported in the literature for Radio Frequency (RF) based drone detection using Deep Neural Networks.
This paper investigates the problem of detection and classification of unmanned aerial vehicles (UAVs) in the presence of wireless interference signals using a passive radio frequency (RF) surveillance system. The system uses a multistage detector to distinguish signals transmitted by a UAV controller from the background noise and interference signals. First, RF signals from any source are detected using a Markov models-based naïve Bayes decision mechanism. When the receiver operates at a signal-to-noise ratio (SNR) of 10 dB, and the threshold, which defines the states of the models, is set at a level 3.5 times the standard deviation of the preprocessed noise data, a detection accuracy of 99.8% with a false alarm rate of 2.8% is achieved. Second, signals from Wi-Fi and Bluetooth emitters, if present, are detected based on the bandwidth and modulation features of the detected RF signal. Once the input signal is identified as a UAV controller signal, it is classified using machine learning (ML) techniques. Fifteen statistical features extracted from the energy transients of the UAV controller signals are fed to neighborhood component analysis (NCA), and the three most significant features are selected. The performance of the NCA and five different ML classifiers are studied for 15 different types of UAV controllers. A classification accuracy of 98.13% is achieved by k-nearest neighbor classifier at 25 dB SNR. Classification performance is also investigated at different SNR levels and for a set of 17 UAV controllers which includes two pairs from the same UAV controller models.
This paper presents a comprehensive review of current literature on drone detection and classification using machine learning with different modalities. This research area has emerged in the last few years due to the rapid development of commercial and recreational drones and the associated risk to airspace safety. Addressed technologies encompass radar, visual, acoustic, and radio-frequency sensing systems. The general finding of this study demonstrates that machine learning-based classification of drones seems to be promising with many successful individual contributions. However, most of the performed research is experimental and the outcomes from different papers can hardly be compared. A general requirement-driven specification for the problem of drone detection and classification is still missing as well as reference datasets which would help in evaluating different solutions.
The Internet of things (IoT) has significant importance in the beyond fifth generation (B5G) communication systems. However, the IoT is vulnerable to disasters because the network is mains powered and the devices are delicate. In this study, an unmanned aerial vehicle (UAV) is utilized to assist with emergency communications in a heterogeneous IoT (Het-IoT) and a distributed non-orthogonal multiple access (NOMA) scheme is proposed without the requirement of successive interference cancellation (SIC). In order to accommodate the communications of the surviving users and IoT devices efficiently, a multi-objective resource allocation (MORA) scheme is proposed for the UAV-assisted Het-IoT. At first, the original MORA problem is formulated and decoupled with the user power initialization. Then, based on a reweighted message-passing algorithm (ReMPA), the subchannels are assigned to the devices and the users in a stepwise manner. Finally, the transmitting power of the users and the devices is jointly fine-tuned using an iterative access control scheme. The simulation results confirm that the distributed SIC-free NOMA (DSF-NOMA) based on the MORA scheme provides satisfactory communication performances for the users and the devices with a tradeoff between the two subsystems. Compared with the traditional orthogonal frequency division multiple access (OFDMA) scheme and the MORA scheme with random subchannel assignment, the proposed DSF-NOMA-based MORA scheme yields better performances in terms of the sum rate of the users and the access ratio of the devices.
The residual unit and its variations are wildly used in building very deep neural networks for alleviating optimization difficulty. In this work, we revisit the standard residual function as well as its several successful variants and propose a unified framework based on tensor Block Term Decomposition (BTD) to explain these apparently different residual functions from the tensor decomposition view. With the BTD framework, we further propose a novel basic network architecture, named the Collective Residual Unit (CRU). CRU further enhances parameter efficiency of deep residual neural networks by sharing core factors derived from collective tensor factorization over the involved residual units. It enables efficient knowledge sharing across multiple residual units, reduces the number of model parameters, lowers the risk of over-fitting, and provides better generalization ability. Extensive experimental results show that our proposed CRU network brings outstanding parameter efficiency -- it achieves comparable classification performance with ResNet-200 while using a model size as small as ResNet-50 on the ImageNet-1k and Places365-Standard benchmark datasets.
The rapid uptake of mobile devices and the rising popularity of mobile applications and services pose unprecedented demands on mobile and wireless networking infrastructure. Upcoming 5G systems are evolving to support exploding mobile traffic volumes, agile management of network resource to maximize user experience, and extraction of fine-grained real-time analytics. Fulfilling these tasks is challenging, as mobile environments are increasingly complex, heterogeneous, and evolving. One potential solution is to resort to advanced machine learning techniques to help managing the rise in data volumes and algorithm-driven applications. The recent success of deep learning underpins new and powerful tools that tackle problems in this space. In this paper we bridge the gap between deep learning and mobile and wireless networking research, by presenting a comprehensive survey of the crossovers between the two areas. We first briefly introduce essential background and state-of-the-art in deep learning techniques with potential applications to networking. We then discuss several techniques and platforms that facilitate the efficient deployment of deep learning onto mobile systems. Subsequently, we provide an encyclopedic review of mobile and wireless networking research based on deep learning, which we categorize by different domains. Drawing from our experience, we discuss how to tailor deep learning to mobile environments. We complete this survey by pinpointing current challenges and open future directions for research.
Recent research on deep neural networks has focused primarily on improving accuracy. For a given accuracy level, is typically possible to identify multiple DNN architectures that achieve that accuracy level. With equivalent accuracy, smaller DNN architectures offer at least three advantages: (1) Smaller DNNs require less communication across servers during distributed training. (2) Smaller DNNs require less bandwidth to export a new model from the cloud to an autonomous car. (3) Smaller DNNs are more feasible to deploy on FPGAs and other hardware with limited memory. To provide all of these advantages, we propose a small DNN architecture called SqueezeNet. SqueezeNet achieves AlexNet-level accuracy on ImageNet with 50x fewer parameters. Additionally, with model compression techniques we are able to compress SqueezeNet to less than 1MB (461x smaller than AlexNet). The SqueezeNet architecture is available for download here: https://github.com/DeepScale/SqueezeNet
Recent work has shown that convolutional networks can be substantially deeper, more accurate and efficient to train if they contain shorter connections between layers close to the input and those close to the output. In this paper we embrace this observation and introduce the Dense Convolutional Network (DenseNet), where each layer is directly connected to every other layer in a feed-forward fashion. Whereas traditional convolutional networks with L layers have L connections, one between each layer and its subsequent layer (treating the input as layer 0), our network has L(L+1)/2 direct connections. For each layer, the feature maps of all preceding layers are treated as separate inputs whereas its own feature maps are passed on as inputs to all subsequent layers. Our proposed connectivity pattern has several compelling advantages: it alleviates the vanishing gradient problem and strengthens feature propagation; despite the increase in connections, it encourages feature reuse and leads to a substantial reduction of parameters; its models tend to generalize surprisingly well. We evaluate our proposed architecture on five highly competitive object recognition benchmark tasks. The DenseNet obtains significant improvements over the state-of-the-art on all five of them (e.g., yielding 3.74% test error on CIFAR-10, 19.25% on CIFAR-100 and 1.59% on SVHN).
In this paper, we introduce a new channel pruning method to accelerate very deep convolutional neural networks.Given a trained CNN model, we propose an iterative two-step algorithm to effectively prune each layer, by a LASSO regression based channel selection and least square reconstruction. We further generalize this algorithm to multi-layer and multi-branch cases. Our method reduces the accumulated error and enhance the compatibility with various architectures. Our pruned VGG-16 achieves the state-of-the-art results by 5x speed-up along with only 0.3% increase of error. More importantly, our method is able to accelerate modern networks like ResNet, Xception and suffers only 1.4%, 1.0% accuracy loss under 2x speed-up respectively, which is significant.
Unmanned aerial vehicles (UAV) are among the major growing technologies that have many beneficial applications, yet they can also pose a significant threat. Recently, several incidents occurred with UAVs violating privacy of the public and security of sensitive facilities, including several nuclear power plants in France. The threat of UAVs to the security of nuclear facilities is of great importance and is the focus of this work. This paper presents an overview of UAV technology and classification, as well as its applications and potential threats. We show several examples of recent security incidents involving UAVs in France, USA, and United Arab Emirates. Further, the potential threats to nuclear facilities and measures to prevent them are evaluated. The importance of measures for detection, delay, and response (neutralization) of UAVs at nuclear facilities are discussed. An overview of existing technologies along with their strength and weaknesses are shown. Finally, the results of a gap analysis in existing approaches and technologies is presented in the form of potential technological and procedural areas for research and development. Based on this analysis, directions for future work in the field can be devised and prioritized.
We present a class of efficient models called MobileNets for mobile and embedded vision applications. MobileNets are based on a streamlined architecture that uses depth-wise separable convolutions to build light weight deep neural networks. We introduce two simple global hyper-parameters that efficiently trade off between latency and accuracy. These hyper-parameters allow the model builder to choose the right sized model for their application based on the constraints of the problem. We present extensive experiments on resource and accuracy tradeoffs and show strong performance compared to other popular models on ImageNet classification. We then demonstrate the effectiveness of MobileNets across a wide range of applications and use cases including object detection, finegrain classification, face attributes and large scale geo-localization.
This paper presents incremental network quantization (INQ), a novel method, targeting to efficiently convert any pre-trained full-precision convolutional neural network (CNN) model into a low-precision version whose weights are constrained to be either powers of two or zero. Unlike existing methods which are struggled in noticeable accuracy loss, our INQ has the potential to resolve this issue, as benefiting from two innovations. On one hand, we introduce three interdependent operations, namely weight partition, group-wise quantization and re-training. A well-proven measure is employed to divide the weights in each layer of a pre-trained CNN model into two disjoint groups. The weights in the first group are responsible to form a low-precision base, thus they are quantized by a variable-length encoding method. The weights in the other group are responsible to compensate for the accuracy loss from the quantization, thus they are the ones to be re-trained. On the other hand, these three operations are repeated on the latest re-trained group in an iterative manner until all the weights are converted into low-precision ones, acting as an incremental network quantization and accuracy enhancement procedure. Extensive experiments on the ImageNet classification task using almost all known deep CNN architectures including AlexNet, VGG-16, GoogleNet and ResNets well testify the efficacy of the proposed method. Specifically, at 5-bit quantization, our models have improved accuracy than the 32-bit floating-point references. Taking ResNet-18 as an example, we further show that our quantized models with 4-bit, 3-bit and 2-bit ternary weights have improved or very similar accuracy against its 32-bit floating-point baseline. Besides, impressive results with the combination of network pruning and INQ are also reported. The code will be made publicly available.
The paper introduces an innovative way of exploiting aerial robotics or Unmanned Aerial Vehicles (UAVs) in the communication field which adds up a wider angle of applications to the already vast use of this new technology. Since IEEE 802 Wi-Fi technology is popular and widely used nowadays (in smartphones, laptops, access points, smart TVs etc.), performance reducing interference caused by having any two or more close-by 2.4 GHz networks on the same channel can occur especially in highly dense populated urban areas. We design an aerial robot (UAV) with a powerful Wi-Fi receiver along with high gain antennas and some other equipment enabling it to operate as a so called Passive sniffer, listening to all Wi-Fi traffic in range and extracting network information along with GPS coordinates and illustrate the results on Geographical maps using Wi-Fi Triangulation Techniques. Having such capabilities on an autonomous aerial robot enables geographical packet analysis and the observation of widespread traffic, plus the ability to analyse and graph the areas with coverage and locate the highly crowded areas with higher interference. Therefore finding solutions to mitigate this undesirable performance degrading interference.
Flying Ad Hoc Networks (FANETs) are essential for a wide range of military and civil applications as they significantly reduce mission duration and improve coverage compared to using a single Unmanned Aerial Vehicle (UAV). Nevertheless, the dynamic and mobile nature of FANETs, coupled with fluctuating data traffic patterns, pose significant challenges for adaptive routing and efficient packet delivery. At present, most of the existing routing protocols for FANETs are designed based on mobility or topology information without taking into account data traffic and the associated joint decision-making approach. This narrow focus can result in poor network performance, characterized by longer delivery delay and lower delivery ratio, which can significantly impair the overall Quality of Service (QoS) in FANETs. To address these challenges, this paper proposes the Joint Prediction and Entropy weight-based (JPE) routing protocol by considering both current and future network conditions. The protocol uses a Long Short-Term Memory (LSTM) model to predict and obtain the mobility, buffer available size, and Link Expiration Time (LET) of each Neighbouring UAV (NU) in order to avoid high-mobility, high-traffic, and weak-link UAVs and establish an appropriate path. The routing decision problem is then formulated as an optimization problem and solved using the proposed Entropy Weight-based Multi-Metric (EWMM) approach to make fast and joint routing decisions. The integrated prediction and decision process considers both current and future multi-metric factors that can lead to packet loss or delay. Simulation results demonstrate the effectiveness of the LSTM-based Joint Prediction (JP) model and show that the JPE protocol outperforms the PAP and SPA protocols, improving Packet Delivery Ratio (PDR) and delay performance by 30.13% and 31.24% respectively.
Software-Defined Networking (SDN)-based Industrial Internet of Things (IIoT) networks have a centralized controller that is a single attractive target for unauthorized users to attack. Cybersecurity in IIoT networks is becoming the most significant challenge, especially from increasingly sophisticated Distributed Denial-of-Service (DDoS) attacks. This situation necessitates efficient approaches to mitigate recent attacks following the incompetence of existing techniques that focus more on DDoS detection. Most existing DDoS detection capabilities are computationally complex and are no longer efficient enough to protect against DDoS attacks. Thus, the need for a low-cost approach for DDoS attack classification. This study presents a competent feature selection method Extreme Gradient Boosting (XGBoost) for determining the most relevant data features with a hybrid Convolutional Neural Network and Long Short-Term Memory (CNN-LSTM) for DDoS attack classification. The proposed model evaluated the CICDDoS2019 dataset with improved accuracy and low-complexity capability for low latency IIoT requirements. Performance results show that the proposed model achieves a high accuracy of 99.50% with a time cost of 0.179 ms.
In recent years, the availability of commercial unmanned air vehicles (UAVs) has increased enormously because of device miniaturization and low cost. However, the abuse of UAVs needs to be investigated to prevent serious security threats for civilians. Therefore, this paper presents a convolutional neural network-based surveillance system for drone detection and its type identification, namely CNN-SSDI. The network architecture is cleverly designed based on deep convolution layers to successfully learn all intrinsic feature maps of radio-frequency signals that are collected from three different types of drones. Further, a detailed comparative analysis of various kernel impairments of the convolution layer structure was investigated under various performance metrics evaluation and higher accuracy in drone surveillance systems. According to the empirical results, CNN-SSDI can detect a UAV with 99.8% accuracy and recognize drone types with an accuracy of 94.5%, which outperforms other existing drone detection and identification techniques.
This paper presents a convolution neural network (CNN)-based direction of arrival (DOA) estimation method for radio frequency (RF) signals acquired by a nonuniform linear antenna array (NULA) in unmanned aerial vehicle (UAV) localization systems. The proposed deep CNN, namely RFDOA-Net, is designed with three primary processing modules, such as collective feature extraction, multi-scaling feature processing, and complexity-accuracy trade-off, to learn the multi-scale intrinsic characteristics for multi-class angle classification. In several specific modules, the regular convolutional and grouped convolutional layers are leveraged with different filter sizes to enrich diversified features and reduce network complexity besides adopting residual connection to prevent vanishing gradient. For performance evaluation, we generate a synthetic signal dataset for DOA estimation under the multipath propagation channel with the presence of additive noise, propagation attenuation and delay. In simulations, the effectiveness of RFDOA-Net is investigated comprehensively with various processing modules and antenna configurations. Compared with several state-of-the-art deep learning-based models, RFDOA-Net shows the superiority in terms of accuracy with over 94% accuracy at 5 dB signal-to-noise ratio (SNR) with cost-efficiency.
Despite several beneficial applications, unfortunately, drones are also being used for illicit activities such as drug trafficking, firearm smuggling or to impose threats to security-sensitive places like airports and nuclear power plants. The existing drone localization and neutralization technologies work on the assumption that the drone has already been detected and classified. Although we have observed a tremendous advancement in the sensor industry in this decade, there is no robust drone detection and classification method proposed in the literature yet. This paper focuses on radio frequency (RF) based drone detection and classification using the frequency signature of the transmitted signal. We have created a novel drone RF dataset using commercial drones and presented a detailed comparison between a two-stage and combined detection and classification framework. The detection and classification performance of both frameworks are presented for a single-signal and simultaneous multi-signal scenario. With detailed analysis, we show that You Only Look Once (YOLO) framework provides better detection performance compared to the Goodness-of-Fit (GoF) spectrum sensing for a simultaneous multi-signal scenario and good classification performance comparable to Deep Residual Neural Network (DRNN) framework.
Advancements in the Internet of Things (IoT) have led to the revolutionary potential to achieve intelligent environmental monitoring. However, extreme conditions across hard-to-reach areas, where public ground networks do not provide sufficient coverage, have resulted in the difficult backhaul of monitoring data from remote areas of interest. Because of the urgent demand for low-cost data collection in such areas, this paper proposes a novel drone-enabled IoT relay system to provide high-speed data collection to support remote environmental monitoring. The high-speed ability of 5 GHz communication technology was exploited to reduce the time required for data transmission between ground monitoring devices and the drone. Meanwhile, long-range (LoRa) technology, as a low-power and long-distance wireless communication technology, is adopted as a wake-up strategy for waking up the high-power 5 GHz module. Based on 5 GHz and LoRa technology, a drone-based on-board relay and ground intelligent terminal are designed. An application is used to demonstrate the feasibility of the designed system. Numerous real-world experiments validate the effectiveness of the designed drone-enabled IoT relay system and show its capability for high-speed data collection. The field experiments demonstrate that the system collects cached data with a stable 3.5 MB/s throughput at an altitude of 140 m. The breakthroughs in the drone-enabled IoT relay facilitate intelligent environmental monitoring in remote areas without public networks and provide a new perspective on data backhaul over areas of interest.
Recently wireless technologies are growing actively from all parts of globe. In context of wireless technology, 5G technology has become a most challenging and interesting topic in wireless research. This article provides an overview of internet of things in 5G wireless systems. IoT in 5G system will be a game changer in future generation. It will open a door for new wireless architecture and smart services. Recent cellular network LTE (4G) will not be sufficient and efficient to meet the demands of multiple device connectivity and high data rate, more bandwidth, low latency Quality of service (QoS) and low interference. To address these challenges, we consider 5G as the most promising technology. We make a depth overview on challenges and vision of various communication industries in 5G IoT systems. The different layers in 5G IoT systems are discussed in detail. This paper provides a comprehensive review on emerging and enabling technologies related to 5G system that enables internet of things. We consider the technology drivers for 5G wireless technology such 5G new radio (NR), MIMO antenna with beamformation technology, mm wave commutation technology, heterogeneous networks (HetNets), role of augmented reality in IoT are discussed in detail. We also make a review on low power wide area networks (LPWAN), security challenges and its control measure in 5G IoT scenario. This article introduces the role of augmented reality in 5G IoT scenario. The article also discusses the research gaps and future directions. The focus in this paper is also on application areas of IoT in 5G systems. We therefore outline some of the important research direction in 5G internet of things.
The omnipresence of unmanned aerial vehicles, or drones, among civilians can lead to technical, security, and public safety issues that need to be addressed, regulated and prevented. Security agencies are in continuous search for technologies and intelligent systems that are capable of detecting drones. Unfortunately, breakthroughs in relevant technologies are hindered by the lack of open source databases for drone’s Radio Frequency (RF) signals, which are remotely sensed and stored to enable developing the most effective way for detecting and identifying these drones. This paper presents a stepping stone initiative towards the goal of building a database for the RF signals of various drones under different flight modes. We systematically collect, analyze, and record raw RF signals of different drones under different flight modes such as: off, on and connected, hovering, flying, and video recording. In addition, we design intelligent algorithms to detect and identify intruding drones using the developed RF database. Three deep neural networks (DNN) are used to detect the presence of a drone, the presence of a drone and its type, and lastly, the presence of a drone, its type, and flight mode. Performance of each DNN is validated through a 10-fold cross-validation process and evaluated using various metrics. Classification results show a general decline in performance when increasing the number of classes. Averaged accuracy has decreased from 99.7% for the first DNN (2-classes), to 84.5% for the second DNN (4-classes), and lastly, to 46.8% for the third DNN (10-classes). Nevertheless, results of the designed methods confirm the feasibility of the developed drone RF database to be used for detection and identification. The developed drone RF database along with our implementations are made publicly available for students and researchers alike.
Neural networks are a subset of the field of artificial intelligence (AI). The predominant types of neural networks used for multidimensional signal processing are deep convolutional neural networks (CNNs). The term deep refers generically to networks having from a "few" to several dozen or more convolution layers, and deep learning refers to methodologies for training these systems to automatically learn their functional parameters using data representative of a specific problem domain of interest. CNNs are currently being used in a broad spectrum of application areas, all of which share the common objective of being able to automatically learn features from (typically massive) data bases and to generalize their responses to circumstances not encountered during the learning phase. Ultimately, the learned features can be used for tasks such as classifying the types of signals the CNN is expected to process. The purpose of this "Lecture Notes" article is twofold: 1) to introduce the fundamental architecture of CNNs and 2) to illustrate, via a computational example, how CNNs are trained and used in practice to solve a specific class of problems.
In this paper, we study the problem of designing efficient convolutional neural network architectures with the interest in eliminating the redundancy in convolution kernels. In addition to structured sparse kernels, low-rank kernels and the product of low-rank kernels, the product of structured sparse kernels, which is a framework for interpreting the recently-developed interleaved group convolutions (IGC) and its variants (e.g., Xception), has been attracting increasing interests. Motivated by the observation that the convolutions contained in a group convolution in IGC can be further decomposed in the same manner, we present a modularized building block, {IGCV2:} interleaved structured sparse convolutions. It generalizes interleaved group convolutions, which is composed of two structured sparse kernels, to the product of more structured sparse kernels, further eliminating the redundancy. We present the complementary condition and the balance condition to guide the design of structured sparse kernels, obtaining a balance among three aspects: model size, computation complexity and classification accuracy. Experimental results demonstrate the advantage on the balance among these three aspects compared to interleaved group convolutions and Xception, and competitive performance compared to other state-of-the-art architecture design methods.
Recently, unmanned aircraft systems (UASs) have attracted attention as a new avenue for commercial services. Using the flexible mobility of unmanned aircrafts (UAs), commercial services can be operated in wide areas. However, there is a problem in the wireless communication between the UA and its ground station. When several UASs are operated within the neighboring airspace, wireless-communication conflicts occur. One of the most effective solutions for this issue is to decide the communication schedule using a time-division multiple access (TDMA) scheme. Furthermore, by spatially reusing the time slot, numerous UAs can be operated within the neighboring airspace, in a limited frequency band. In this paper, we propose an efficient time-slot allocation for enhancing the frequency resource utilization. Our proposed scheme determines the time-slot allocation considering the time-slot spatial reuse, using a virtual cell-based space partitioning method. In addition, we consider the influence of the UA mobility on the network to decide the parameters for the proposed resource allocation system. The effectiveness of our proposed resource allocation is evaluated through computer-based simulation.
In this paper, we propose four new variants of the backpropagation algorithm to improve the generalization ability for feedforward neural networks. The basic idea of these methods stems from the Group Lasso concept which deals with the variable selection problem at the group level. There are two main drawbacks when the Group Lasso penalty has been directly employed during network training. They are numerical oscillations and theoretical challenges in computing the gradients at the origin. To overcome these obstacles, smoothing functions have then been introduced by approximating the Group Lasso penalty. Numerical experiments for classification and regression problems demonstrate that the proposed algorithms perform better than the other three classical penalization methods, Weight Decay, Weight Elimination, and Approximate Smoother, on both generalization and pruning efficiency. In addition, detailed simulations based on a specific data set have been performed to compare with some other common pruning strategies, which verify the advantages of the proposed algorithm. The pruning abilities of the proposed strategy have been investigated in detail for a relatively large data set, MNIST, in terms of various smoothing approximation cases.
In this work we investigate the effect of the convolutional network depth on its accuracy in the large-scale image recognition setting. Our main contribution is a thorough evaluation of networks of increasing depth, which shows that a significant improvement on the prior-art configurations can be achieved by pushing the depth to 16-19 weight layers. These findings were the basis of our ImageNet Challenge 2014 submission, where our team secured the first and the second places in the localisation and classification tracks respectively.
We trained a large, deep convolutional neural network to classify the 1.2 million high-resolution images in the ImageNet LSVRC-2010 contest into the 1000 dif- ferent classes. On the test data, we achieved top-1 and top-5 error rates of 37.5% and 17.0% which is considerably better than the previous state-of-the-art. The neural network, which has 60 million parameters and 650,000 neurons, consists of five convolutional layers, some of which are followed by max-pooling layers, and three fully-connected layers with a final 1000-way softmax. To make training faster, we used non-saturating neurons and a very efficient GPU implemen- tation of the convolution operation. To reduce overfitting in the fully-connected layers we employed a recently-developed regularization method called dropout that proved to be very effective. We also entered a variant of this model in the ILSVRC-2012 competition and achieved a winning top-5 test error rate of 15.3%, compared to 26.2% achieved by the second-best entry
We propose an efficient and unified framework, namely ThiNet, to simultaneously accelerate and compress CNN models in both training and inference stages. We focus on the filter level pruning, i.e., the whole filter would be discarded if it is less important. Our method does not change the original network structure, thus it can be perfectly supported by any off-the-shelf deep learning libraries. We formally establish filter pruning as an optimization problem, and reveal that we need to prune filters based on statistics information computed from its next layer, not the current layer, which differentiates ThiNet from existing methods. Experimental results demonstrate the effectiveness of this strategy, which has advanced the state-of-the-art. We also show the performance of ThiNet on ILSVRC-12 benchmark. ThiNet achieves 3.31 FLOPs reduction and 16.63 compression on VGG-16, with only 0.52 top-5 accuracy drop. Similar experiments with ResNet-50 reveal that even for a compact network, ThiNet can also reduce more than half of the parameters and FLOPs, at the cost of roughly 1 top-5 accuracy drop. Moreover, the original VGG-16 model can be further pruned into a very small model with only 5.05MB model size, preserving AlexNet level accuracy but showing much stronger generalization ability.
We introduce an extremely computation efficient CNN architecture named ShuffleNet, designed specially for mobile devices with very limited computing power (e.g., 10-150 MFLOPs). The new architecture utilizes two proposed operations, pointwise group convolution and channel shuffle, to greatly reduce computation cost while maintaining accuracy. Experiments on ImageNet classification and MS COCO object detection demonstrate the superior performance of ShuffleNet over other structures, e.g. lower top-1 error (absolute 6.7\%) than the recent MobileNet system on ImageNet classification under the computation budget of 40 MFLOPs. On an ARM-based mobile device, ShuffleNet achieves \textasciitilde 13 actual speedup over AlexNet while maintaining comparable accuracy.
This paper is concerned with path planning for unmanned aerial vehicles (UAVs) flying through low altitude urban environment. Although many different path planning algorithms have been proposed to find optimal or near-optimal collision-free paths for UAVs, most of them either do not consider dynamic obstacle avoidance or do not incorporate multiple objectives. In this paper, we propose a multi-objective path planning (MOPP) framework to explore a suitable path for a UAV operating in a dynamic urban environment, where safety level is considered in the proposed framework to guarantee the safety of UAV in addition to travel time. To this aim, two types of safety index maps (SIMs) are developed first to capture static obstacles in the geography map and unexpected obstacles that are unavailable in the geography map. Then an MOPP method is proposed by jointly using offline and online search, where the offline search is based on the static SIM and helps shorten the travel time and avoid static obstacles, while the online search is based on the dynamic SIM of unexpected obstacles and helps bypass unexpected obstacles quickly. Extensive experimental results verify the effectiveness of the proposed framework under the dynamic urban environment.
Residual units are wildly used for alleviating optimization difficulties when building deep neural networks. However, the performance gain does not well compensate the model size increase, indicating low parameter efficiency in these residual units. In this work, we first revisit the residual function in several variations of residual units and demonstrate that these residual functions can actually be explained with a unified framework based on generalized block term decomposition. Then, based on the new explanation, we propose a new architecture, Collective Residual Unit (CRU), which enhances the parameter efficiency of deep neural networks through collective tensor factorization. CRU enables knowledge sharing across different residual units using shared factors. Experimental results show that our proposed CRU Network demonstrates outstanding parameter efficiency, achieving comparable classification performance to ResNet-200 with the model size of ResNet-50. By building a deeper network using CRU, we can achieve state-of-the-art single model classification accuracy on ImageNet-1k and Places365-Standard benchmark datasets. (Code and trained models are available on GitHub)
We propose a new framework for pruning convolutional kernels in neural networks to enable efficient inference, focusing on transfer learning where large and potentially unwieldy pretrained networks are adapted to specialized tasks. We interleave greedy criteria-based pruning with fine-tuning by backpropagation - a computationally efficient procedure that maintains good generalization in the pruned network. We propose a new criterion based on an efficient first-order Taylor expansion to approximate the absolute change in training cost induced by pruning a network component. After normalization, the proposed criterion scales appropriately across all layers of a deep CNN, eliminating the need for per-layer sensitivity analysis. The proposed criterion demonstrates superior performance compared to other criteria, such as the norm of kernel weights or average feature map activation.