Article
To read the full-text of this research, you can request a copy directly from the authors.

Abstract

Early diagnosis of nutrient deficiencies can play a major role in avoiding significant agricultural losses and increasing the final yield while preserving the environment through efficient fertilizer usage. In this work, we study how well nutrient deficiency symptoms can be recognized in RGB images by using deep neural networks and transfer learning. Two different datasets, presenting real-world conditions, were used for this purpose. The first one was the Deep Nutrient Deficiency for Sugar Beet (DND-SB) dataset, which contains 5648 images of sugar beets presenting nitrogen (N), phosphorous (P), and potassium (K) deficiencies, the omission of liming (Ca) and full fertilization. The second one, collected on the field for this research and currently publicly available, was a dataset combining different orange tree images with iron (Fe), potasssium (K), magnesium (Mg), and manganese (Mn) deficiencies. Image classification via fine-tuning with EfficientNetB4, whose original weights came from a noisy student training on ImageNet, obtained the best performances on both datasets with 98.65% and 98.52% Top-1 accuracies. Additionally, the Grad-CAM++ analysis showed that the models were performing an accurate analysis of the most relevant part inside the images. Finally, the use of agricultural transfer learning did not report improvement in the performances.

No full-text available

Request Full-text Paper PDF

To read the full-text of this research,
you can request a copy directly from the authors.

Article
Full-text available
The classification of plug seedling quality plays an active role in enhancing the quality of seedlings. The EfficientNet-B7-CBAM model, an improved convolutional neural network (CNN) model, was proposed to improve classification efficiency and reduce high cost. To ensure that the EfficientNet-B7 model simultaneously learns crucial channel and spatial location information, the convolutional block attention module (CBAM) has been incorporated. To improve the model’s ability to generalize, a transfer learning strategy and Adam optimization algorithm were introduced. A system for image acquisition collected 8,109 images of pepper plug seedlings, and data augmentation techniques improved the resulting data set. The proposed EfficientNet-B7-CBAM model achieved an average accuracy of 97.99% on the test set, 7.32% higher than before the improvement. Under the same experimental conditions, the classification accuracy increased by 8.88–20.05% to classical network models such as AlexNet, VGG16, InceptionV3, ResNet50, and DenseNet121. The proposed method had high accuracy in the plug seedling quality classification task. It was well-adapted to numerous types of plug seedlings, providing a reference for developing a fast and accurate algorithm for plug seedling quality classification.
Article
Full-text available
In order to enable timely actions to prevent major losses of crops caused by lack of nutrients and, hence, increase the potential yield throughout the growing season while at the same time prevent excess fertilization with detrimental environmental consequences, early, non-invasive, and on-site detection of nutrient deficiency is required. Current non-invasive methods for assessing the nutrient status of crops deal in most cases with nitrogen (N) deficiency only and optical sensors to diagnose N deficiency, such as chlorophyll meters or canopy reflectance sensors, do not monitor N, but instead measure changes in leaf spectral properties that may or may not be caused by N deficiency. In this work, we study how well nutrient deficiency symptoms can be recognized in RGB images of sugar beets. To this end, we collected the Deep Nutrient Deficiency for Sugar Beet (DND-SB) dataset, which contains 5648 images of sugar beets growing on a long-term fertilizer experiment with nutrient deficiency plots comprising N, phosphorous (P), and potassium (K) deficiency, as well as the omission of liming (Ca), full fertilization, and no fertilization at all. We use the dataset to analyse the performance of five convolutional neural networks for recognizing nutrient deficiency symptoms and discuss their limitations.
Article
Full-text available
Symptoms of nutrient deficiencies in rice plants often appear on the leaves. The leaf color and shape, therefore, can be used to diagnose nutrient deficiencies in rice. Image classification is an efficient and fast approach for this diagnosis task. Deep convolutional neural networks (DCNNs) have been proven to be effective in image classification, but their use to identify nutrient deficiencies in rice has received little attention. In the present study, we explore the accuracy of different DCNNs for diagnosis of nutrient deficiencies in rice. A total of 1818 photographs of plant leaves were obtained via hydroponic experiments to cover full nutrition and 10 classes of nutrient deficiencies. The photographs were divided into training, validation, and test sets in a 3 : 1 : 1 ratio. Fine-tuning was performed to evaluate four state-of-the-art DCNNs: Inception-v3, ResNet with 50 layers, NasNet-Large, and DenseNet with 121 layers. All the DCNNs obtained validation and test accuracies of over 90%, with DenseNet121 performing best (validation accuracy = 98.62 ± 0.57%; test accuracy = 97.44 ± 0.57%). The performance of the DCNNs was validated by comparison to color feature with support vector machine and histogram of oriented gradient with support vector machine. This study demonstrates that DCNNs provide an effective approach to diagnose nutrient deficiencies in rice.
Article
Full-text available
In light of growing challenges in agriculture with ever growing food demand across the world, efficient crop management techniques are necessary to increase crop yield. Precision agriculture techniques allow the stakeholders to make effective and customized crop management decisions based on data gathered from monitoring crop environments. Plant phenotyping techniques play a major role in accurate crop monitoring. Advancements in deep learning have made previously difficult phenotyping tasks possible. This survey aims to introduce the reader to the state of the art research in deep plant phenotyping.
Article
Full-text available
Nitrogen (N) concentration is a significant parameter to check the status of health in rice crop. Nitrogen (N) plays an essential role in the growth and productivity of rice plant. This paper proposes a convolutional neural network (CNN) based approach for prediction of rice nitrogen deficiency. The pre-trained CNN architecture is modified to improve the classification accuracy with the inclusion of pre-eminent classifier like support vector machine (SVM) by replacing the last output layer of CNN. Here, six leading deep learning architectures such as ResNet-18, ResNet-50, GoogleNet, AlexNet, VGG-16 and VGG-19 with SVM are used for prediction of nitrogen deficiency with 5790 number image samples. The performance of each classifier is measured and compared in terms of accuracy, sensitivity, specificity, false positive rate (FPR) and F1 score. Again, the statistical analysis is performed to choose the better classification model considering the results of 100 independent simulations. The statistical analysis confirmed the superiority of ResNet-50+SVM than the other five CNN-based classification models with an accuracy of 99.84%. Besides, the accuracy score of CNN classification models is compared with other traditional image classification models such as bag-of-feature, colour feature + SVM, local binary patterns (LBP) + SVM, histogram of oriented gradients (HOG)+SVM and Gray Level Co-occurrence Matrix (GLCM)+SVM.
Article
Full-text available
Frequently, the vineyards in the Douro Region present multiple grape varieties per parcel and even per row. An automatic algorithm for grape variety identification as an integrated software component was proposed that can be applied, for example, to a robotic harvesting system. However, some issues and constraints in its development were highlighted, namely, the images captured in natural environment, low volume of images, high similarity of the images among different grape varieties, leaf senescence, and significant changes on the grapevine leaf and bunch images in the harvest seasons, mainly due to adverse climatic conditions, diseases, and the presence of pesticides. In this paper, the performance of the transfer learning and fine-tuning techniques based on AlexNet architecture were evaluated when applied to the identification of grape varieties. Two natural vineyard image datasets were captured in different geographical locations and harvest seasons. To generate different datasets for training and classification, some image processing methods, including a proposed four-corners-in-one image warping algorithm, were used. The experimental results, obtained from the application of an AlexNet-based transfer learning scheme and trained on the image dataset pre-processed through the four-corners-in-one method, achieved a test accuracy score of 77.30%. Applying this classifier model, an accuracy of 89.75% on the popular Flavia leaf dataset was reached. The results obtained by the proposed approach are promising and encouraging in helping Douro wine growers in the automatic task of identifying grape varieties.
Article
Full-text available
Plant diseases cause great damage in agriculture, resulting in significant yield losses. The recent expansion of deep learning methods has found its application in plant disease detection, offering a robust tool with highly accurate results. The current limitations and shortcomings of existing plant disease detection models are presented and discussed in this paper. Furthermore, a new dataset containing 79,265 images was introduced with the aim to become the largest dataset containing leaf images. Images were taken in various weather conditions, at different angles, and daylight hours with an inconsistent background mimicking practical situations. Two approaches were used to augment the number of images in the dataset: traditional augmentation methods and state-of-the-art style generative adversarial networks. Several experiments were conducted to test the impact of training in a controlled environment and usage in real-life situations to accurately identify plant diseases in a complex background and in various conditions including the detection of multiple diseases in a single leaf. Finally, a novel two-stage architecture of a neural network was proposed for plant disease classification focused on a real environment. The trained model achieved an accuracy of 93.67%.
Article
Full-text available
Abstract Deep convolutional neural networks have performed remarkably well on many Computer Vision tasks. However, these networks are heavily reliant on big data to avoid overfitting. Overfitting refers to the phenomenon when a network learns a function with very high variance such as to perfectly model the training data. Unfortunately, many application domains do not have access to big data, such as medical image analysis. This survey focuses on Data Augmentation, a data-space solution to the problem of limited data. Data Augmentation encompasses a suite of techniques that enhance the size and quality of training datasets such that better Deep Learning models can be built using them. The image augmentation algorithms discussed in this survey include geometric transformations, color space augmentations, kernel filters, mixing images, random erasing, feature space augmentation, adversarial training, generative adversarial networks, neural style transfer, and meta-learning. The application of augmentation methods based on GANs are heavily covered in this survey. In addition to augmentation techniques, this paper will briefly discuss other characteristics of Data Augmentation such as test-time augmentation, resolution impact, final dataset size, and curriculum learning. This survey will present existing methods for Data Augmentation, promising developments, and meta-level decisions for implementing Data Augmentation. Readers will understand how Data Augmentation can improve the performance of their models and expand limited datasets to take advantage of the capabilities of big data.
Article
Full-text available
During the process of plant growth, such as during the flowering stages and fruit development, the plants need to be provided with the various minerals and nutrients to grow. Nutrient deficiency is the cause of serious diseases in plant growth, affecting crop yield. In this article, we employed artificial neural network models to recognize, classify, and predict the nutritional deficiencies occurring in tomato plants (Solanum lycopersicum L.). To classify and predict the different macronutrient deficiencies in the cropping process, this paper handles the captured images of the macronutrient deficiency. This deficiency during the fruiting and leafing phases of tomato plant are based on a deep convolutional neural network (CNN). A total of 571 images were captured with tomato leaves and fruits containing the crop species at the growth stage. Among all images, 80% (461 captured images) were used for the training dataset and 20% (110 captured images) were applied for the validation dataset. In this study, we provide an analysis of two different model architectures based on convolutional neural network for classifying and predicting the nutrient deficiency symptoms. For instance, Inception-ResNet v2 and Autoencoder with the captured images of tomato plant growth under the greenhouse conditions. Moreover, a major type of statistical structure, namely Ensemble Averaging, was applied with two aforementioned predictive models to increase the accuracy of predictive validation. Three mineral nutrients, i.e., Calcium/Ca2+, Potassium/K+, and Nitrogen/N, are considered for use in evaluating the nutrient status in the development of tomato plant with these models. The aim of this study is to predict the nutrient deficiency accurately in order to increase crop production and prevent the emergence of tomato pathology caused by lack of nutrients. The predictive performance of the three models in this paper are validated, with the accuracy rates of 87.273% and 79.091% for Inception-ResNet v2 and Autoencoder, respectively, and with 91% validity using Ensemble Averaging.
Article
Full-text available
Robotic weed control has seen increased research of late with its potential for boosting productivity in agriculture. Majority of works focus on developing robotics for croplands, ignoring the weed management problems facing rangeland stock farmers. Perhaps the greatest obstacle to widespread uptake of robotic weed control is the robust classification of weed species in their natural environment. The unparalleled successes of deep learning make it an ideal candidate for recognising various weed species in the complex rangeland environment. This work contributes the first large, public, multiclass image dataset of weed species from the Australian rangelands; allowing for the development of robust classification methods to make robotic weed control viable. The DeepWeeds dataset consists of 17,509 labelled images of eight nationally significant weed species native to eight locations across northern Australia. This paper presents a baseline for classification performance on the dataset using the benchmark deep learning models, Inception-v3 and ResNet-50. These models achieved an average classification accuracy of 95.1% and 95.7%, respectively. We also demonstrate real time performance of the ResNet-50 architecture, with an average inference time of 53.4 ms per image. These strong results bode well for future field implementation of robotic weed control methods in the Australian rangelands.
Article
Full-text available
Deep neural nets Transfer learning Image database Disease classification Deep learning is quickly becoming one of the most important tools for image classification. This technology is now beginning to be applied to the tasks of plant disease classification and recognition. The positive results that are being obtained using this approach hide some issues that are seldom taken into account in the respective experiments. This article presents an investigation into the main factors that affect the design and effectiveness of deep neural nets applied to plant pathology. An in-depth analysis of the subject, in which advantages and shortcomings are highlighted, should lead to more realistic conclusions on the subject. The arguments used throughout the text are built upon both studies found in the literature and experiments carried out using an image database carefully built to reflect and reproduce many of the conditions expected to be found in practice. This database, which contains almost 50,000 images, is being made freely available for academic purposes.
Article
Full-text available
The essentiality of 14 mineral elements so far have been reported in plant nutrition. Eight of these elements were known as micronutrients due to their lower concentrations in plants (usually ≤100 mg/kg/dw). However, it is still challenging to mention an exact number of plant micronutrients since some elements have not been strictly proposed yet either as essential or beneficial. Micronutrients participate in very diverse metabolic processes, including from the primary and secondary metabolism to the cell defense, and from the signal transduction to the gene regulation, energy metabolism, and hormone perception. Thus, the attempt to understand the molecular mechanism(s) behind their transport has great importance in terms of basic and applied plant sciences. Moreover, their deficiency or toxicity also caused serious disease symptoms in plants, even plant destruction if not treated, and many people around the world suffer from the plant-based dietary deficiencies or metal toxicities. In this sense, shedding some light on this issue, the 13 mineral elements (Fe, B, Cu, Mn, Mo, Si, Zn, Ni, Cl, Se, Na, Al, and Co), required by plants at trace amounts, has been reviewed with the primary focus on the transport proteins (transporters/channels) in plant roots. So, providing the compiled but extensive information about the structural and functional roles of micronutrient transport genes/proteins in plant roots.
Conference Paper
Full-text available
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), which connects each layer 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 - our network has L(L+1)/2 direct connections. For each layer, the feature-maps of all preceding layers are used as inputs, and its own feature-maps are used as inputs into all subsequent layers. DenseNets have several compelling advantages: they alleviate the vanishing-gradient problem, strengthen feature propagation, encourage feature reuse, and substantially reduce the number of parameters. We evaluate our proposed architecture on four highly competitive object recognition benchmark tasks (CIFAR-10, CIFAR-100, SVHN, and ImageNet). DenseNets obtain significant improvements over the state-of-the-art on most of them, whilst requiring less memory and computation to achieve high performance. Code and models are available at https://github.com/liuzhuang13/DenseNet.
Article
Full-text available
The latest generation of convolutional neural networks (CNNs) has achieved impressive results in the field of image classification. This paper is concerned with a new approach to the development of plant disease recognition model, based on leaf image classification, by the use of deep convolutional networks. Novel way of training and the methodology used facilitate a quick and easy system implementation in practice. The developed model is able to recognize 13 different types of plant diseases out of healthy leaves, with the ability to distinguish plant leaves from their surroundings. According to our knowledge, this method for plant disease recognition has been proposed for the first time. All essential steps required for implementing this disease recognition model are fully described throughout the paper, starting from gathering images in order to create a database, assessed by agricultural experts. Caffe, a deep learning framework developed by Berkley Vision and Learning Centre, was used to perform the deep CNN training. The experimental results on the developed model achieved precision between 91% and 98%, for separate class tests, on average 96.3%.
Article
Full-text available
Crop diseases are a major threat to food security, but their rapid identification remains difficult in many parts of the world due to the lack of the necessary infrastructure. The combination of increasing global smartphone penetration and recent advances in computer vision made possible by deep learning has paved the way for smartphone-assisted disease diagnosis. Using a public dataset of 54,306 images of diseased and healthy plant leaves collected under controlled conditions, we train a deep convolutional neural network to identify 14 crop species and 26 diseases (or absence thereof). The trained model achieves an accuracy of 99.35% on a held-out test set, demonstrating the feasibility of this approach. When testing the model on a set of images collected from trusted online sources - i.e. taken under conditions different from the images used for training - the model still achieves an accuracy of 31.4%. While this accuracy is much higher than the one based on random selection (2.6%), a more diverse set of training data is needed to improve the general accuracy. Overall, the approach of training deep learning models on increasingly large and publicly available image datasets presents a clear path towards smartphone-assisted crop disease diagnosis on a massive global scale.
Conference Paper
Full-text available
Layer-sequential unit-variance (LSUV) initialization -- a simple method for weight initialization for deep net learning -- is proposed. The method consists of the two steps. First, pre-initialize weights of each convolution or inner-product layer with orthonormal matrices. Second, proceed from the first to the final layer, normalizing the variance of the output of each layer to be equal to one. Experiment with different activation functions (maxout, ReLU-family, tanh) show that the proposed initialization leads to learning of very deep nets that (i) produces networks with test accuracy better or equal to standard methods and (ii) is at least as fast as the complex schemes proposed specifically for very deep nets such (Romero et al. (2015)) and Highway (Srivastava et al. (2015)). Performance is evaluated on GoogLeNet, CaffeNet, FitNets and Residual nets and the state-of-the-art, or very close to it, is achieved on the MNIST, CIFAR-10/100 and ImageNet datasets.
Article
Full-text available
Whereas before 2006 it appears that deep multilayer neural networks were not successfully trained, since then several algorithms have been shown to successfully train them, with experimental results showing the superiority of deeper vs less deep architectures. All these experimental results were obtained with new initialization or training mechanisms. Our objective here is to understand better why standard gradient descent from random initialization is doing so poorly with deep neural networks, to better understand these recent relative successes and help design better algorithms in the future. We first observe the influence of the non-linear activations functions. We find that the logistic sigmoid activation is unsuited for deep networks with random initialization because of its mean value, which can drive especially the top hidden layer into saturation. Surprisingly, we find that saturated units can move out of saturation by themselves, albeit slowly, and explaining the plateaus sometimes seen when training neural networks. We find that a new non-linearity that saturates less can often be beneficial. Finally, we study how activations and gradients vary across layers and during training, with the idea that training may be more difficult when the singular values of the Jacobian associated with each layer are far from 1. Based on these considerations, we propose a new initialization scheme that brings substantially faster convergence.
Article
Full-text available
Multilayer neural networks trained with the back-propagation algorithm constitute the best example of a successful gradient based learning technique. Given an appropriate network architecture, gradient-based learning algorithms can be used to synthesize a complex decision surface that can classify high-dimensional patterns, such as handwritten characters, with minimal preprocessing. This paper reviews various methods applied to handwritten character recognition and compares them on a standard handwritten digit recognition task. Convolutional neural networks, which are specifically designed to deal with the variability of 2D shapes, are shown to outperform all other techniques. Real-life document recognition systems are composed of multiple modules including field extraction, segmentation recognition, and language modeling. A new learning paradigm, called graph transformer networks (GTN), allows such multimodule systems to be trained globally using gradient-based methods so as to minimize an overall performance measure. Two systems for online handwriting recognition are described. Experiments demonstrate the advantage of global training, and the flexibility of graph transformer networks. A graph transformer network for reading a bank cheque is also described. It uses convolutional neural network character recognizers combined with global training techniques to provide record accuracy on business and personal cheques. It is deployed commercially and reads several million cheques per day
Technical Report
The management of nutrients in agricultural production is a critical process that can affect the environmental impact of agricultural enterprises, in terms of emissions to water and air, and their economic viability. In particular, the use of nitrogen fertilisers can result in losses of nitrate to surface and groundwater and emissions of ammonia and nitrous oxide to the atmosphere. The issue of nutrient pollution from agriculture has been an ongoing challenge. Many of the processes are well understood, but their management in the context of agricultural production is often one of trying to strike a balance between competing objectives, which can increase the complexity of decision making processes at both the farm and policy levels. In the European Union (EU), several policies have been developed and implemented to address nutrient pollution from agriculture including Council Directive 91/676/EEC (the Nitrates Directive). Article 5 requires each Member State (MS) to develop nitrate action programme (NAP) measures. These can vary from one MS (or region) to another, thus there is scope for differences across the EU. There have been numerous previous studies on the implementation of the Nitrates Directive, which have either tended to focus on individual MSs or on specific measures or issues. However, to date there have been few detailed studies that have aimed to holistically synthesise and compare NAP measures across the EU to identify differences in implementation and potential impact. To address this, during 2019-20 the European Commission contracted the Agriculture and Environment Research Unit at the University of Hertfordshire in the UK to undertake a project titled: "Providing support in relation to the identification of approaches and measures in action programmes Directive 91/676/EEC" (Ref. ENV.D.1/SER/2018/0017). This document is the final report for this project. The project involved the design and development of an inventory of NAP measures, associated inventory database and software, and the Nitrate Action Programme Information (NAPINFO) web application. The inventory was populated by undertaking an extensive literature review to collate information on the implementation of NAP measures in each MS. Eighty NAPs were reviewed (21 MS NAPs and 59 regional NAPs within 7 MSs); 73 NAP measures were defined (with 35 defined as core measures) containing ~13600 measure variants. These were stored in the inventory database using a range of quantitative and qualitative fields. The NAPINFO web application was populated using the inventory database and was used to facilitate a consultation with MS experts to check, amend and refine the data in the inventory. Around two-thirds (66%) of MSs responded to the consultation covering 46% of the NAPs. The information collated on NAPs came from a range of sources (written in different languages) that are subject to change over time, including during the lifetime of this project. Consequently, the inventory should be considered a 'first edition' and is essentially a snapshot of the situation within each MS/region in 2019. The inventory software was used to interrogate the database to compare the implementation of NAP measures within each of the 80 NAPs. Twelve of the core measures are implemented in over 75% of the NAPs and majority of the measures in Annex III of the Nitrates Directive are implemented in over 90% of the NAPs. A more detailed comparison of the implementation of the measures is provided in Annex A, which also includes the findings of an extensive literature review on key processes and elements covered by each NAP measure with respect to nitrate loss and other environmental impacts (which was used to support the NAP characterisation). The analysis revealed that there is significant variability between NAPs (in terms of the measures within each NAP and how they have been implemented), but this does not necessary correlate with variation in the risk of nitrate loss, as some of the variation will be due to tailoring to regional circumstances. However, there is scope for MSs/regions to learn from each other, especially where regional pedoclimatic and agricultural circumstances are similar and where there are differences in the measures implemented. Each NAP has been characterised with respect to its potential to reduce the risk of nitrate loss and other environmental impacts including: nitrous oxide; methane; carbon dioxide; ammonia; phosphorus; pathogens; substances with a high biochemical oxygen demand; soil erosion; biodiversity; water use; and pesticide loss. A bespoke risk-based method (drawing upon the source-pathway-receptor concept) was developed for this project that accounts for the key risk elements associated with each environmental impact and which is tailored using spatial data and statistics that account for regional variabilities including climate, geology, slope, flooding and irrigation. The characterisation process results in a Risk Mitigation Potential (RMP) class (low to high) for each NAP and each measure within the NAP. The measures are ranked based on their contribution to the overall RMP of the NAP and they are also assessed individually to identify potential for enhancement. Those measures within the inventory but not in the NAP for an MS/region have also been characterised to identify their potential as possible new measures to be implemented. The full outputs for each NAP are provided in a separate report for each MS and an Excel workbook (Annexes C1-28). The RMPs for all NAPs are also summed and presented in Annex D. Detailed analysis and insights can be found in Section 4.4 and Annexes C and D. However, as examples, the characterisation revealed that across all the NAPs measures related to fertiliser application are addressing both run-off and leaching, but have a slightly higher mitigation potential for losses of via run-off; closed periods (M30) is ranked the highest most often across all the NAPs and tends to have slightly higher mitigation potential for losses of via leaching (compared to run-off); and closed periods also contribute to reducing the risk of nitrous oxide emissions, with over half of NAPs having moderate to high mitigation potential. The extensive literature review also collated data with respect to agricultural land use, nitrates in surface and groundwater bodies and any transboundary issues within each MS. The main purpose of this is to provide context for the outputs from the other tasks. The main outputs with regard to water quality and transboundary impacts are presented in the reports for each MS (Annexes C1-28). Summarising this information on a broad continental scale is challenging, as the aggregation processes tend to result in the loss of important detail, especially with respect to sampling points, frequencies, identification of hotspots and temporal boundaries. This is partly due fragmented data reporting, different data sources and availability. However, the data that has been summarised does provide a snapshot of the most recent concentrations monitored in each MS, the general trend and the scope for transboundary impacts. Finally, the NAP measures have been compared to measures implemented under Directive 2016/2284, the National Emission Ceilings Directive (NECD), to identify any potential synergies and conflicts with respect impacts on emissions of nitrate and ammonia. During 2019, National Air Pollution Control Programmes were being submitted by MSs; twenty of these were available to review within the lifetime of this project, from which 31 common NECD measures were identified and added to the NAPINFO database. Synergies and conflicts were identified firstly by comparing NECD measures to NAP measures and, secondly, by examining the effect of NAP measures ammonia emissions (derived from the risk characterisation). The general conclusion is that there appears to be synergy between the NECD and the Nitrates Directive and for the few instances where a potential conflict was identified there is scope for mitigation. It is important to note that this study did not aim to assess whether the measures are 'sufficient' with respect to meeting the requirements of the ND or for improving water quality (the RMP class does not assess this). This is because it is difficult to integrate the findings and correlate outputs from the implementation analysis; the risk characterisation process; and trends in the nitrate concentration of surface and groundwater (and any transboundary issues). The broad nature of the analyses and the data collated do not facilitate a credible or robust assessment. Correlating land use practices and interventions with changes in nitrates in surface and groundwater bodies (and any transboundary issues) needs to be done on a more detailed and localised site by site basis. However, the information presented in this project, does provide a holistic picture which could be used as a basis to identify areas where potential impacts could be explored more specifically (e.g. specific measures/interventions in specific regions or catchments). This has been an ambitious and challenging project that has aimed to provide a broad and current picture of the implementation of NAP measures under the Nitrates Directive. The inventory of NAP measures coupled with risk characterisation; data on water quality and transboundary issues; and the analysis of synergies and conflicts with the NECD, provides a holistic overview that can be used to support the full implementation of existing legislation, policy development and to identify areas for future research.
Conference Paper
Tomato crops are one of the most important agricultural products at economic level in the world. However, the quality of the tomato fruits is highly dependent to the growing conditions such as the nutrients. One of consequences of the latter during tomato harvesting is nutrient deficiency. Manually, it is possible to anticipate the lack of primary nutrients (i.e. nitrogen, phosphorus and potassium) by looking the appearance of the leaves in tomato plants. Thus, this paper presents a supervised vision-based monitoring system for detecting nutrients deficiencies in tomato crops by taking images from the leaves of the plants. It uses a Convolutional Neural Network (CNN) to recognize and classify the type of nutrient that is deficient in the plants. First, we created a data set of images of leaves of tomato plants showing different symptoms due to the nutrient deficiency. Then, we trained a suitable CNN-model with our images and other augmented data. Experimental results showed that our CNN-model can achieve 86.57% of accuracy. We anticipate the implementation of our proposal for future precision agriculture applications such as automated nutrient level monitoring and control in tomato crops.
Article
Nowadays, several studies in the field of deep learning in agriculture obtain high performances in weeds identification by fine-tuning neural networks, previously trained on general-purpose datasets containing images unrelated to agriculture. This work examines whether these achievements could be further improved by fine-tuning neural networks pre-trained on agricultural datasets instead of ImageNet. The experimental results showed that with the suggested method the overall performance can increase. Some architectures such as Xception and Inception-Resnet presented an improvement of 0.51% and 1.89% respectively, while reducing the number of epochs by 13.67%. It is then argued that an agricultural repository should be developed to engage research into making their pre-trained neural networks publicly available, for the benefit of research progress and efficiency.
Article
Reducing the use of pesticides through selective spraying is an important component towards a more sustainable computer-assisted agriculture. Weed identification at early growth stage contributes to reduced herbicide rates. However, while computer vision alongside deep learning have overcome the performance of approaches that use hand-crafted features, there are still some open challenges in the development of a reliable automatic plant identification system. These type of systems have to take into account different sources of variability, such as growth stages and soil conditions, with the added constraint of the limited size of usual datasets. This study proposes a novel crop/weed identification system that relies on a combination of fine-tuning pre-trained convolutional networks (Xception, Inception-Resnet, VGNets, Mobilenet and Densenet) with the “traditional” machine learning classifiers (Support Vector Machines, XGBoost and Logistic Regression) trained with the previously deep extracted features. The aim of this approach was to avoid overfitting and to obtain a robust and consistent performance. To evaluate this approach, an open access dataset of two crop [tomato (Solanum lycopersicum L.) and cotton (Gossypium hirsutum L.)] and two weed species [black nightshade (Solanum nigrum L.) and velvetleaf (Abutilon theophrasti Medik.)] was generated. The pictures were taken by different production sites across Greece under natural variable light conditions from RGB cameras. The results revealed that a combination of fine-tuned Densenet and Support Vector Machine achieved a micro F1 score of 99.29% with a very low performance difference between train and test sets. Other evaluated approaches also obtained repeatedly more than 95% F1 score. Additionally, our results analysis provides some heuristics for designing transfer-learning based systems to avoid overfitting without decreasing performance.
Article
Potato blackleg is a tuber-borne bacterial disease caused by species within the genera Dickeya and Pectobacterium that can cause decay of plant tissue and wilting through the action of cell wall degrading enzymes released by the pathogen. In case of serious infections, tubers may rot before emergence. Management is largely based on the use of pathogen-free seed potato tubers. For this, fields are visually monitored both for certification and also to take out diseased plants to avoid spread to neighboring plants. Imaging potentially offers a quick and non-destructive way to inspect the health of potato plants in a field. Early detection of blackleg diseased plants with modern vision techniques can significantly reduce costs. In this paper, we studied the use of deep learning for detecting blackleg diseased potato plants. Two deep convolutional neural networks were trained on RGB images with healthy and diseased plants. One of these networks (ResNet18) was experimentally found to produce a precision of 95 % and recall of 91 % for the disease class. These results show that convolutional neural networks can be used to detect blackleg diseased potato plants.
Article
Classification of weeds amongst cash crops is a core procedure in automated weed control. Addressing volunteer potato control in sugar beets, in the EU Smartbot project the aim was to control more than 95% of volunteer potatoes and ensure less than 5% of undesired control of sugar beet plants. A promising way to meet these requirements is deep learning. Training an entire network from scratch, however, requires a large dataset and a substantial amount of time. In this situation, transfer learning can be a promising solution. This study first evaluates a transfer learning procedure with three different implementations of AlexNet and then assesses the performance difference amongst the six network architectures: AlexNet, VGG-19, GoogLeNet, ResNet-50, ResNet-101 and Inception-v3. All nets had been pre-trained on the ImageNet Dataset. These nets were used to classify sugar beet and volunteer potato images taken under ambient varying light conditions in agricultural environments. The highest classification accuracy for different implementations of AlexNet was 98.0%, obtained with an AlexNet architecture modified to generate binary output. Comparing different networks, the highest classification accuracy 98.7%, obtained with VGG-19 modified to generate binary output. Transfer learning proved to be effective and showed robust performance with plant images acquired in different periods of the various years on two types of soils. All scenarios and pre-trained networks were feasible for real-time applications (classification time < 0.1 s). Classification is only one step in weed detection, and a complete pipeline for weed detection may potentially reduce the overall performance.
Book
In 2007, the first edition of Handbook of Plant Nutrition presented a compendium of information on the mineral nutrition of plants available at that time-and became a bestseller and trusted resource. Updated to reflect recent advances in knowledge of plant nutrition, the second edition continues this tradition. With chapters written by a new team of experts, each element is covered in a different manner, providing a fresh look and new understanding of the material. The chapters extensively explore the relationship between plant genetics and the accumulation and use of nutrients by plants, adding to the coverage available in the first edition. The second edition features a chapter on lanthanides, which have gained importance in plant nutrition since the publication of the first edition, and contains chapters on the different mineral elements. It follows the general pattern of a description of the determination of essentiality or beneficial effects of the element, uptake and assimilation, physiological responses of plants to the element, genetics of its acquisition by plants, concentrations of the element and its derivatives and metabolites in plants, interaction of the element with uptake of other elements, diagnosis of concentrations of the element in plants, forms and concentrations of the element in soils and its availability to plants, soil tests and fertilizers used to supply the element. The book demonstrates how the appearance and composition of plants can be used to assess nutritional status and the value of soil tests for assessing nutrition status. It also includes recommendations of fertilizers that can be applied to remedy nutritional deficiencies. These features and more make Handbook of Plant Nutrition, Second Edition a practical, easy-to-use reference for determining, monitoring, and improving the nutritional profiles of plants worldwide.
Article
In this paper, convolutional neural network models were developed to perform plant disease detection and diagnosis using simple leaves images of healthy and diseased plants, through deep learning methodologies.Training of the models was performed with the use of an open database of 87,848 images, containing 25 different plants in a set of 58 distinct classes of [plant, disease] combinations, including healthy plants. Several model architectures were trained, with the best performance reaching a 99.53% success rate in identifying the corresponding [plant, disease] combination (or healthy plant). The significantly high success rate makes the model a very useful advisory or early warning tool, and an approach that could be further expanded to support an integrated plant disease identification system to operate in real cultivation conditions.
Article
Although deep learning has historical roots going back decades, neither the term "deep learning" nor the approach was popular just over five years ago, when the field was reignited by papers such as Krizhevsky, Sutskever and Hinton's now classic (2012) deep network model of Imagenet. What has the field discovered in the five subsequent years? Against a background of considerable progress in areas such as speech recognition, image recognition, and game playing, and considerable enthusiasm in the popular press, I present ten concerns for deep learning, and suggest that deep learning must be supplemented by other techniques if we are to reach artificial general intelligence.
Article
In this paper, we explore and compare multiple solutions to the problem of data augmentation in image classification. Previous work has demonstrated the effectiveness of data augmentation through simple techniques, such as cropping, rotating, and flipping input images. We artificially constrain our access to data to a small subset of the ImageNet dataset, and compare each data augmentation technique in turn. One of the more successful data augmentations strategies is the traditional transformations mentioned above. We also experiment with GANs to generate images of different styles. Finally, we propose a method to allow a neural net to learn augmentations that best improve the classifier, which we call neural augmentation. We discuss the successes and shortcomings of this method on various datasets.
Book
An understanding of the mineral nutrition of plants is of fundamental importance in both basic and applied plant sciences. The Second Edition of this book retains the aim of the first in presenting the principles of mineral nutrition in the light of current advances. This volume retains the structure of the first edition, being divided into two parts: Nutritional Physiology and Soil-Plant Relationships. In Part I, more emphasis has been placed on root-shoot interactions, stress physiology, water relations, and functions of micronutrients. In view of the worldwide increasing interest in plant-soil interactions, Part II has been considerably altered and extended, particularly on the effects of external and interal factors on root growth and chapter 15 on the root-soil interface. The second edition will be invaluable to both advanced students and researchers.
Conference Paper
In this paper we present a perception system for agriculture robotics that enables an unmanned ground vehicle (UGV) equipped with a multi spectral camera to automatically perform the crop/weed detection and classification tasks in real-time. Our approach exploits a pipeline that includes two different convolutional neural networks (CNNs) applied to the input RGB+near infra-red (NIR) images. A lightweight CNN is used to perform a fast and robust, pixel-wise, binary image segmentation, in order to extract the pixels that represent projections of 3D points that belong to green vegetation. A deeper CNN is then used to classify the extracted pixels between the crop and weed classes. A further important contribution of this work is a novel unsupervised dataset summarization algorithm that automatically selects from a large dataset the most informative subsets that better describe the original one. This enables to streamline and speed-up the manual dataset labeling process, otherwise extremely time consuming, while preserving good classification performance. Experiments performed on different datasets taken from a real farm robot confirm the effectiveness of our approach.
Conference Paper
Convolutional networks are at the core of most stateof-the-art computer vision solutions for a wide variety of tasks. Since 2014 very deep convolutional networks started to become mainstream, yielding substantial gains in various benchmarks. Although increased model size and computational cost tend to translate to immediate quality gains for most tasks (as long as enough labeled data is provided for training), computational efficiency and low parameter count are still enabling factors for various use cases such as mobile vision and big-data scenarios. Here we are exploring ways to scale up networks in ways that aim at utilizing the added computation as efficiently as possible by suitably factorized convolutions and aggressive regularization. We benchmark our methods on the ILSVRC 2012 classification challenge validation set demonstrate substantial gains over the state of the art: 21.2% top-1 and 5.6% top-5 error for single frame evaluation using a network with a computational cost of 5 billion multiply-adds per inference and with using less than 25 million parameters. With an ensemble of 4 models and multi-crop evaluation, we report 3.5% top-5 error and 17.3% top-1 error.
Article
Since the emergence of Deep Neural Networks (DNNs) as a prominent technique in the field of computer vision, the ImageNet classification challenge has played a major role in advancing the state-of-the-art. While accuracy figures have steadily increased, the resource utilization of winning models has not been properly taken into account. In this work, we present a comprehensive analysis of important metrics in practical applications: accuracy, memory footprint, parameters, operations count, inference time and power consumption. Key findings are: (1) fully connected layers are largely inefficient for smaller batches of images; (2) accuracy and inference time are in a hyperbolic relationship; (3) energy constraint are an upper bound on the maximum achievable accuracy and model complexity; (4) the number of operations is a reliable estimate of the inference time. We believe our analysis provides a compelling set of information that help design and engineer efficient DNNs.
Article
Very deep convolutional networks have been central to the largest advances in image recognition performance in recent years. One example is the Inception architecture that has been shown to achieve very good performance at relatively low computational cost. Recently, the introduction of residual connections in conjunction with a more traditional architecture has yielded state-of-the-art performance in the 2015 ILSVRC challenge; its performance was similar to the latest generation Inception-v3 network. This raises the question of whether there are any benefit in combining the Inception architecture with residual connections. Here we give clear empirical evidence that training with residual connections accelerates the training of Inception networks significantly. There is also some evidence of residual Inception networks outperforming similarly expensive Inception networks without residual connections by a thin margin. We also present several new streamlined architectures for both residual and non-residual Inception networks. These variations improve the single-frame recognition performance on the ILSVRC 2012 classification task significantly. We further demonstrate how proper activation scaling stabilizes the training of very wide residual Inception networks. With an ensemble of three residual and one Inception-v4, we achieve 3.08 percent top-5 error on the test set of the ImageNet classification (CLS) challenge
EfficientNet: rethinking model scaling for convolutional neural networks
  • M Tan
  • Q V Le
Fine-tuning deep convolutional networks for plant recognition
  • A K Reyes
  • J C Caicedo
  • J E Camargo
Impact of excessive nitrogen fertilizers on the environment and associated mitigation strategies
  • M Bashir
  • S Ali
  • M Ghauri
  • A Adris
  • R Harun
Bashir, M., Ali, S., Ghauri, M., Adris, A., & Harun, R. (2013). Impact Of Excessive Nitrogen Fertilizers On The Environment And Associated Mitigation Strategies.
ImageNet-trained CNNs are biased towards texture
  • R Geirhos
  • P Rubisch
  • C Michaelis
  • M Bethge
  • F Wichmann
  • W Brendel
Geirhos, R., Rubisch, P., Michaelis, C., Bethge, M., Wichmann, F., & Brendel, W. (2019). ImageNet-trained CNNs are biased towards texture; increasing shape bias improves accuracy and robustness. ArXiv, abs/1811.12231.
Human vs. supervised machine learning: who learns patterns faster? ArXiv
  • N Kühl
  • M Goutier
  • L Baier
  • C Wolff
  • D Martin
Kühl, N., Goutier, M., Baier, L., Wolff, C., & Martin, D. (2020). Human vs. supervised machine learning: Who learns patterns faster? ArXiv, abs/2012.03661.
EfficientNetV2: smaller models and faster training
  • Tan
  • A Canziani
  • A Paszke
  • E Culurciello
Canziani, A., Paszke, A., & Culurciello, E., 2016. An Analysis of Deep Neural Network Models for Practical Applications. arXiv:1605.07678.
Essential and Beneficial Trace Elements in Plants, and Their Transport in Roots: a
  • R Vatansever
  • I I Ozyigit
  • E Filiz
Vatansever, R., Ozyigit, I.I., & Filiz, E. (2016). Essential and Beneficial Trace Elements in Plants, and Their Transport in Roots: a Review. Applied Biochemistry and Biotechnology, 181, 464-482.