Pub Date : 2024-07-12DOI: 10.1109/TAFE.2024.3421238
Bernardita Štitić;Luca Urbinati;Giuseppe Di Guglielmo;Luca P. Carloni;Mario R. Casu
Combining data-driven machine learning (ML) with microwave sensing (MWS) makes it possible to analyze packaged food in real time without any contact and spot low-density contaminants (e.g., plastics or glass splinters) that current industrial food safety methods cannot detect. This is achieved by training ML classifiers on the scattered signal reflected by the target food product exposed to MWs. In this article, we present an enhanced ML flow to boost foreign body detection accuracy of ML classifiers in MWS systems. Innovations include assessing the performance of a multiclass classifier, training it with MW frequency pairs as features, data augmentation, a new preprocessing scaler suitable for the feature distributions in the datasets, quantization, and pruning. The final test results, obtained using our previously designed MWS system and collected dataset of contaminated hazelnut-cocoa spread jars, show a multiclass accuracy for the floating-point model of 96.5%, which corresponds to a binary-equivalent accuracy of 97.3%. This is an improvement of +3.3% against the binary classifier of the previous work. The quantized and pruned model, instead, reached a multiclass accuracy of 94.2%, or a binary accuracy of 95.4%, thus still improving the previous work by +1.4%. Also, we achieved a latency of 26 $mu$ s on an AMD/Xilinx Kria K26 field programmable gate array (FPGA), a result which is ideal for high-throughput food production lines. Furthermore, we expand our latest work with supplementary details and experiments to further validate the proposed ML flow, including a comparative analysis against our prior results. Lastly, we share our datasets publicly on OpenML.
将数据驱动的机器学习（ML）与微波传感（MWS）相结合，可以对包装食品进行无接触实时分析，并发现当前工业食品安全方法无法检测的低密度污染物（如塑料或玻璃碎片）。这是通过对暴露在微波中的目标食品反射的散射信号训练 ML 分类器来实现的。在本文中，我们介绍了一种增强型 ML 流程，以提高 MWS 系统中 ML 分类器的异物检测精度。创新之处包括：评估多类分类器的性能、将微波频率对作为特征对其进行训练、数据增强、适合数据集特征分布的新预处理扩展器、量化和剪枝。最终测试结果显示，浮点模型的多类准确率为 96.5%，相当于二元等效准确率 97.3%。与之前的二进制分类器相比，提高了 3.3%。而经过量化和剪枝处理的模型的多分类准确率为 94.2%，二进制准确率为 95.4%，因此仍比之前的工作提高了 +1.4%。此外，我们还在 AMD/Xilinx Kria K26 现场可编程门阵列 (FPGA) 上实现了 26 美元/分钟的延迟，这一结果非常适合高通量食品生产线。此外，我们还通过补充细节和实验扩展了我们的最新工作，以进一步验证所提出的 ML 流程，包括与我们之前的结果进行比较分析。最后，我们在 OpenML 上公开分享我们的数据集。
{"title":"Enhanced Machine-Learning Flow for Microwave-Sensing Systems for Contaminant Detection in Food","authors":"Bernardita Štitić;Luca Urbinati;Giuseppe Di Guglielmo;Luca P. Carloni;Mario R. Casu","doi":"10.1109/TAFE.2024.3421238","DOIUrl":"https://doi.org/10.1109/TAFE.2024.3421238","url":null,"abstract":"Combining data-driven machine learning (ML) with microwave sensing (MWS) makes it possible to analyze packaged food in real time without any contact and spot low-density contaminants (e.g., plastics or glass splinters) that current industrial food safety methods cannot detect. This is achieved by training ML classifiers on the scattered signal reflected by the target food product exposed to MWs. In this article, we present an enhanced ML flow to boost foreign body detection accuracy of ML classifiers in MWS systems. Innovations include assessing the performance of a multiclass classifier, training it with MW frequency pairs as features, data augmentation, a new preprocessing scaler suitable for the feature distributions in the datasets, quantization, and pruning. The final test results, obtained using our previously designed MWS system and collected dataset of contaminated hazelnut-cocoa spread jars, show a multiclass accuracy for the floating-point model of 96.5%, which corresponds to a binary-equivalent accuracy of 97.3%. This is an improvement of +3.3% against the binary classifier of the previous work. The quantized and pruned model, instead, reached a multiclass accuracy of 94.2%, or a binary accuracy of 95.4%, thus still improving the previous work by +1.4%. Also, we achieved a latency of 26 \u0000<inline-formula><tex-math>$mu$</tex-math></inline-formula>\u0000s on an AMD/Xilinx Kria K26 field programmable gate array (FPGA), a result which is ideal for high-throughput food production lines. Furthermore, we expand our latest work with supplementary details and experiments to further validate the proposed ML flow, including a comparative analysis against our prior results. Lastly, we share our datasets publicly on OpenML.","PeriodicalId":100637,"journal":{"name":"IEEE Transactions on AgriFood Electronics","volume":"2 2","pages":"181-189"},"PeriodicalIF":0.0,"publicationDate":"2024-07-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142430797","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-07-12DOI: 10.1109/TAFE.2024.3420028
Bhabesh Deka;Debarun Chakraborty
Traditional methods of manual litchi fruit counting are labor-intensive, time-consuming, and prone to errors. Moreover, due to its complex growth structures, such as occlusion with leaves and branches, overlapping, and uneven color, it becomes more challenging for the current baseline detection and instance segmentation models to accurately identify the litchi fruits. The advancement of deep learning architecture and modern sensing technology, such as unmanned aerial vehicle (UAV), had shown great potential for improving fruit counting accuracy and efficiency. In this article, we propose a modified Mask-region-based convolutional neural network-based instance segmentation model using channel attention to detect and count litchis in complex natural environments using UAV. In addition, we build a UAV-Litchi dataset consisting of 1000 images with 31 000 litchi annotations, collected by the DJI Phantom 4 with RGB sensor and labeled with a LabelImg annotation tool. Experimental results show that the proposed model with the squeeze-and-excitation block improves the detection accuracy of litchi fruits, achieving a mean average precision, recall, and F1 score of 81.47%, 92.81%, and 88.40%, respectively, with an average inference time of 7.72 s. The high accuracy and efficiency of the proposed model demonstrate its potential for precise and accurate litchi detection in precision agriculture.
{"title":"UAV Sensing-Based Litchi Segmentation Using Modified Mask-RCNN for Precision Agriculture","authors":"Bhabesh Deka;Debarun Chakraborty","doi":"10.1109/TAFE.2024.3420028","DOIUrl":"https://doi.org/10.1109/TAFE.2024.3420028","url":null,"abstract":"Traditional methods of manual litchi fruit counting are labor-intensive, time-consuming, and prone to errors. Moreover, due to its complex growth structures, such as occlusion with leaves and branches, overlapping, and uneven color, it becomes more challenging for the current baseline detection and instance segmentation models to accurately identify the litchi fruits. The advancement of deep learning architecture and modern sensing technology, such as unmanned aerial vehicle (UAV), had shown great potential for improving fruit counting accuracy and efficiency. In this article, we propose a modified Mask-region-based convolutional neural network-based instance segmentation model using channel attention to detect and count litchis in complex natural environments using UAV. In addition, we build a UAV-Litchi dataset consisting of 1000 images with 31 000 litchi annotations, collected by the DJI Phantom 4 with RGB sensor and labeled with a LabelImg annotation tool. Experimental results show that the proposed model with the squeeze-and-excitation block improves the detection accuracy of litchi fruits, achieving a mean average precision, recall, and F1 score of 81.47%, 92.81%, and 88.40%, respectively, with an average inference time of 7.72 s. The high accuracy and efficiency of the proposed model demonstrate its potential for precise and accurate litchi detection in precision agriculture.","PeriodicalId":100637,"journal":{"name":"IEEE Transactions on AgriFood Electronics","volume":"2 2","pages":"509-517"},"PeriodicalIF":0.0,"publicationDate":"2024-07-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142408887","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
In the realm of agricultural automation, the precise identification of crop stress holds immense significance for enhancing crop productivity. Existing methods primarily focus on controlled environments, which may not accurately reflect field conditions. Field-based leaf image analysis poses challenges due to varying image quality and sunlight intensity. Moreover, the complexity of crop stress images, with their random lesion distribution, diverse symptoms, and complex backgrounds, further complicates the analysis. To overcome these limitations, a lightweight hybrid convolutional neural network has been developed. This system integrates the powerful three-deep blocks model with an autoencoder running in parallel to highlight regions of crop stress effectively. To support this approach, we have introduced the Indian Rice Disease Dataset (IRDD) with labeled images. The proposed system reports an average true positive rate (TPR) of 0.8766 and an average positive predicted value of 0.8720 on IRDD, which are higher than other state-of-the-art crop disease detection models. The system is validated on benchmark datasets, yielding significant results: TPR of 0.9870 (rice), 0.9985 (tomato), and 0.8559 (corn). Furthermore, the proposed model outperforms recent state-of-the-art works on the benchmark PlantDoc dataset, showing its effectiveness in generalizing plant disease identification tasks. Finally, an ablation study has been carried out to explore the importance of the two parallel branches. Overall, this study acts as a bridge between advanced science and practical application, showcasing how interdisciplinary automation could revolutionize crop disease identification, improve agricultural efficiency, and reshape broader industrial practices.
{"title":"Robust Deep Convolutional Solutions for Identifying Biotic Crop Stress in Wild Environments","authors":"Chiranjit Pal;Imon Mukherjee;Sanjay Chatterji;Sanjoy Pratihar;Pabitra Mitra;Partha Pratim Chakrabarti","doi":"10.1109/TAFE.2024.3422187","DOIUrl":"https://doi.org/10.1109/TAFE.2024.3422187","url":null,"abstract":"In the realm of agricultural automation, the precise identification of crop stress holds immense significance for enhancing crop productivity. Existing methods primarily focus on controlled environments, which may not accurately reflect field conditions. Field-based leaf image analysis poses challenges due to varying image quality and sunlight intensity. Moreover, the complexity of crop stress images, with their random lesion distribution, diverse symptoms, and complex backgrounds, further complicates the analysis. To overcome these limitations, a lightweight hybrid convolutional neural network has been developed. This system integrates the powerful three-deep blocks model with an autoencoder running in parallel to highlight regions of crop stress effectively. To support this approach, we have introduced the Indian Rice Disease Dataset (IRDD) with labeled images. The proposed system reports an average true positive rate (TPR) of 0.8766 and an average positive predicted value of 0.8720 on IRDD, which are higher than other state-of-the-art crop disease detection models. The system is validated on benchmark datasets, yielding significant results: TPR of 0.9870 (rice), 0.9985 (tomato), and 0.8559 (corn). Furthermore, the proposed model outperforms recent state-of-the-art works on the benchmark PlantDoc dataset, showing its effectiveness in generalizing plant disease identification tasks. Finally, an ablation study has been carried out to explore the importance of the two parallel branches. Overall, this study acts as a bridge between advanced science and practical application, showcasing how interdisciplinary automation could revolutionize crop disease identification, improve agricultural efficiency, and reshape broader industrial practices.","PeriodicalId":100637,"journal":{"name":"IEEE Transactions on AgriFood Electronics","volume":"2 2","pages":"497-508"},"PeriodicalIF":0.0,"publicationDate":"2024-07-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142408885","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Wireless sensor networks can be a low-cost and efficient solution for monitoring environmental parameters in agriculture. In this work, we analyze the potentialities of using a network based on a general-purpose Internet of Things protocol such as OpenThread and on low-cost general-purpose physical and chemical sensors. This article aims to test and verify the platform functionalities when monitoring environmental parameters such as temperature, relative humidity, and visible and infrared irradiance in a real environment. We designed wireless sensor nodes and a data collection and visualization dashboard. We tested the sensors' response under controlled settings, conducted connectivity and network topology tests, validated the system functionality via in-field measurements, and discussed the main issues and potential capabilities of these sensors and network architecture.
{"title":"Network Connectivity and Sensor Response of an OpenThread WSN Platform for Crop Monitoring","authors":"Alessandro Checco;Vasco Fabiani;Maurizio Palmisano;Davide Polese","doi":"10.1109/TAFE.2024.3420648","DOIUrl":"https://doi.org/10.1109/TAFE.2024.3420648","url":null,"abstract":"Wireless sensor networks can be a low-cost and efficient solution for monitoring environmental parameters in agriculture. In this work, we analyze the potentialities of using a network based on a general-purpose Internet of Things protocol such as OpenThread and on low-cost general-purpose physical and chemical sensors. This article aims to test and verify the platform functionalities when monitoring environmental parameters such as temperature, relative humidity, and visible and infrared irradiance in a real environment. We designed wireless sensor nodes and a data collection and visualization dashboard. We tested the sensors' response under controlled settings, conducted connectivity and network topology tests, validated the system functionality via \u0000<italic>in-field</i>\u0000 measurements, and discussed the main issues and potential capabilities of these sensors and network architecture.","PeriodicalId":100637,"journal":{"name":"IEEE Transactions on AgriFood Electronics","volume":"2 2","pages":"218-225"},"PeriodicalIF":0.0,"publicationDate":"2024-07-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10592646","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142430801","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-07-09DOI: 10.1109/TAFE.2024.3416414
V. Cabrera;A. Delbuggio;H. Cardoso;D. Fraga;A. Gómez;M. Pedemonte;R. Ungerfeld;J. Oreggioni
Sheep production in extensive conditions faces several challenges. These challenges could be addressed with behavior monitoring systems, contributing to animal well-being, enhancing animal research, and improving productivity. This article presents the design, manufacture, and test of an online sheep behavior monitoring system for extensive conditions. It comprises a wearable electronic collar device and a cloud server (deployed with Amazon Web Services) for storing data and providing a web user interface. The collar has an Icarus Internet of Things (IoT) Board, allowing motion data collection with a three-axis accelerometer, global navigation satellite system (GNSS) location data acquisition, and narrowband IoT communication. The device has solar panels and a battery. Our application acquires accelerometer data at 25 Hz, location data every 10–30 s, and battery level and cellular signal strength every 50 s. We encoded accelerometer samples to reduce the transmitted data. We manufactured 30 collars that collect and transmit data to the cloud server. Our system facilitates data processing, both collar and server side. We introduce a preliminary Random Forest algorithm for behavior classification on the device that identifies “still,” “walking,” and “running” with a 78% general accuracy. The device's autonomy exceeds ten days in continuous operation (streaming raw and processed data) while if the device transmits only processed data and GNSS data every 4 h, autonomy rises to 100 days. This allows us to glimpse the application of this system in long-term research experiments and farming production.
{"title":"Harnessing Technology for Livestock Research: An Online Sheep Behavior Monitoring System","authors":"V. Cabrera;A. Delbuggio;H. Cardoso;D. Fraga;A. Gómez;M. Pedemonte;R. Ungerfeld;J. Oreggioni","doi":"10.1109/TAFE.2024.3416414","DOIUrl":"https://doi.org/10.1109/TAFE.2024.3416414","url":null,"abstract":"Sheep production in extensive conditions faces several challenges. These challenges could be addressed with behavior monitoring systems, contributing to animal well-being, enhancing animal research, and improving productivity. This article presents the design, manufacture, and test of an online sheep behavior monitoring system for extensive conditions. It comprises a wearable electronic collar device and a cloud server (deployed with Amazon Web Services) for storing data and providing a web user interface. The collar has an Icarus Internet of Things (IoT) Board, allowing motion data collection with a three-axis accelerometer, global navigation satellite system (GNSS) location data acquisition, and narrowband IoT communication. The device has solar panels and a battery. Our application acquires accelerometer data at 25 Hz, location data every 10–30 s, and battery level and cellular signal strength every 50 s. We encoded accelerometer samples to reduce the transmitted data. We manufactured 30 collars that collect and transmit data to the cloud server. Our system facilitates data processing, both collar and server side. We introduce a preliminary Random Forest algorithm for behavior classification on the device that identifies “still,” “walking,” and “running” with a 78% general accuracy. The device's autonomy exceeds ten days in continuous operation (streaming raw and processed data) while if the device transmits only processed data and GNSS data every 4 h, autonomy rises to 100 days. This allows us to glimpse the application of this system in long-term research experiments and farming production.","PeriodicalId":100637,"journal":{"name":"IEEE Transactions on AgriFood Electronics","volume":"2 2","pages":"306-313"},"PeriodicalIF":0.0,"publicationDate":"2024-07-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142430692","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-07-08DOI: 10.1109/TAFE.2024.3416415
Ngoc-Bao-Van Le;Jun-Ho Huh
The development of Internet of Things (IoTs) technology in agriculture, notably aquaculture, has increased over the years due to empowering real-time monitoring and improved environmental sustainability. To ensure the development and survival of aquatic life, farm employees must constantly check and take prompt action to protect the sustainable habitat in ponds. It is also crucial for providing technical assistance to farmers during the growing season. To address these issues, we present the building of a smart aquaculture assistant system integrated with IoT and Big Data. The main components of this system include the IoT layer assistant layer. First, tracking the pond environment functions through sensor systems set up in the pond, such as turbidity, pH, and temperature sensors, which are built for the IoT layer. Our assistant can make suggestions for farmers by visualizing farm environment conditions and outdoor weather analysis. The assistant layer uses a fine-tuning NLP model GPT 3.5 for our aquaculture farming 500 frequently asked questions and crawled knowledge dataset. The chatbot can provide naturally logical replies and knowledge related to farming activities. The system is implemented as mobile and desktop applications using React Native and Python to monitor or manipulate administrative tasks. Thus, thanks to our smart assistant, prompt preventive action can be taken to reduce losses, increase productivity, and enhance farmers’ knowledge related to farming.
{"title":"AgTech: Building Smart Aquaculture Assistant System Integrated IoT and Big Data Analysis","authors":"Ngoc-Bao-Van Le;Jun-Ho Huh","doi":"10.1109/TAFE.2024.3416415","DOIUrl":"https://doi.org/10.1109/TAFE.2024.3416415","url":null,"abstract":"The development of Internet of Things (IoTs) technology in agriculture, notably aquaculture, has increased over the years due to empowering real-time monitoring and improved environmental sustainability. To ensure the development and survival of aquatic life, farm employees must constantly check and take prompt action to protect the sustainable habitat in ponds. It is also crucial for providing technical assistance to farmers during the growing season. To address these issues, we present the building of a smart aquaculture assistant system integrated with IoT and Big Data. The main components of this system include the IoT layer assistant layer. First, tracking the pond environment functions through sensor systems set up in the pond, such as turbidity, pH, and temperature sensors, which are built for the IoT layer. Our assistant can make suggestions for farmers by visualizing farm environment conditions and outdoor weather analysis. The assistant layer uses a fine-tuning NLP model GPT 3.5 for our aquaculture farming 500 frequently asked questions and crawled knowledge dataset. The chatbot can provide naturally logical replies and knowledge related to farming activities. The system is implemented as mobile and desktop applications using React Native and Python to monitor or manipulate administrative tasks. Thus, thanks to our smart assistant, prompt preventive action can be taken to reduce losses, increase productivity, and enhance farmers’ knowledge related to farming.","PeriodicalId":100637,"journal":{"name":"IEEE Transactions on AgriFood Electronics","volume":"2 2","pages":"471-482"},"PeriodicalIF":0.0,"publicationDate":"2024-07-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142408403","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Yield estimation plays a crucial role in agricultural planning and decision-making processes. It helps farmers predict the number of fruits they can harvest, which impacts factors such as resource allocation, market forecasting, and financial planning. However, it is a time-consuming task and prone to human errors. In recent years, computer vision and machine learning techniques have demonstrated their effectiveness in supporting farming activities. Automatic fruit detection, localization, and counting algorithms are emerging as valuable tools for facilitating farming activities. However, despite the promising results, the effort required to collect and process reliable datasets often limits the widespread diffusion of such algorithms. For this reason, weakly supervised methodologies have become popular since they do not rely on large amounts of manually labeled data. Achieving results comparable to those obtained by supervised approaches is still one of the main challenges. In line with these considerations, we developed a weakly supervised framework for counting and locating fruits on trees whose training relies solely on image-level presence/absence (fruit) binary annotations. Our network incorporates the state-of-the-art learn-to-pay-attention modules, which generate binary predictions for network optimization in the loss computation. The experimental results, conducted on a publicly available mango orchard dataset, showed significant enhancements in robustness and accuracy compared to other weakly supervised state-of-the-art approaches.
{"title":"Refined Weakly Supervised Yield Estimation Through the MiniLabelNet-Count Framework","authors":"Alessandro Rocco Denarda;Francesco Crocetti;Raffaele Brilli;Gabriele Costante;Paolo Valigi;Mario Luca Fravolini","doi":"10.1109/TAFE.2024.3419709","DOIUrl":"https://doi.org/10.1109/TAFE.2024.3419709","url":null,"abstract":"Yield estimation plays a crucial role in agricultural planning and decision-making processes. It helps farmers predict the number of fruits they can harvest, which impacts factors such as resource allocation, market forecasting, and financial planning. However, it is a time-consuming task and prone to human errors. In recent years, computer vision and machine learning techniques have demonstrated their effectiveness in supporting farming activities. Automatic fruit detection, localization, and counting algorithms are emerging as valuable tools for facilitating farming activities. However, despite the promising results, the effort required to collect and process reliable datasets often limits the widespread diffusion of such algorithms. For this reason, weakly supervised methodologies have become popular since they do not rely on large amounts of manually labeled data. Achieving results comparable to those obtained by supervised approaches is still one of the main challenges. In line with these considerations, we developed a weakly supervised framework for counting and locating fruits on trees whose training relies solely on image-level presence/absence (fruit) binary annotations. Our network incorporates the state-of-the-art learn-to-pay-attention modules, which generate binary predictions for network optimization in the loss computation. The experimental results, conducted on a publicly available mango orchard dataset, showed significant enhancements in robustness and accuracy compared to other weakly supervised state-of-the-art approaches.","PeriodicalId":100637,"journal":{"name":"IEEE Transactions on AgriFood Electronics","volume":"3 1","pages":"34-42"},"PeriodicalIF":0.0,"publicationDate":"2024-07-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143821627","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Effective crop health monitoring is essential for farmers to make informed decisions about managing their crops. In canola crop management, the rapid proliferation of flea beetle (FB) populations is a major concern, as these pests can cause significant crop damage. Traditional manual field monitoring for FBs is time consuming and error-prone due to its reliance on visual assessments of FB damage to small seedlings, making conducting frequent and accurate surveys difficult. One of the key pieces of information in assessing if control of FBs is required is the presence of live FBs in the canola crop. This article proposes a novel insect-monitoring framework that uses a solar-powered, intelligent trap called the smart insect trap (SIT), equipped with a high-resolution camera and a deep-learning-based object detection network. Using this SIT, coupled with a kairomonal lure, the FB population can be monitored hourly, and population increases can be identified quickly. The SIT processes images at the edge and sends results to the cloud every 40 min for FB monitoring and analysis. It uses a modified you look only once version 8 small (YOLOv8s) object detection network, FB-YOLO, to improve its ability to detect small FBs. The modification is implemented in the network's neck, which aggregates features from the deep and early pyramids of the backbone in the neck. Improved attention to small objects is achieved by incorporating spatially aware features from early pyramids. In addition, the network is integrated with an advanced box selection algorithm called confluence nonmax suppression (NMS-C) to prevent duplicate detections in highly overlapped clusters of FBs. The FB-YOLO achieved an average precision ( $text{mAP}@0.5$ ) of 89.97%, a 1.215% improvement over the YOLOv8s network with only 0.324 million additional parameters. Integrating NMS-C further improved the $text{mAP}@0.5$ by 0.19%, leading to an overall $text{mAP}@0.5$ of 90.16%.
{"title":"A Novel Computer Vision System for Efficient Flea Beetle Monitoring in Canola Crop","authors":"Muhib Ullah;Muhammad Shabbir Hasan;Abdul Bais;Tyler Wist;Shaun Sharpe","doi":"10.1109/TAFE.2024.3406329","DOIUrl":"https://doi.org/10.1109/TAFE.2024.3406329","url":null,"abstract":"Effective crop health monitoring is essential for farmers to make informed decisions about managing their crops. In canola crop management, the rapid proliferation of flea beetle (FB) populations is a major concern, as these pests can cause significant crop damage. Traditional manual field monitoring for FBs is time consuming and error-prone due to its reliance on visual assessments of FB damage to small seedlings, making conducting frequent and accurate surveys difficult. One of the key pieces of information in assessing if control of FBs is required is the presence of live FBs in the canola crop. This article proposes a novel insect-monitoring framework that uses a solar-powered, intelligent trap called the smart insect trap (SIT), equipped with a high-resolution camera and a deep-learning-based object detection network. Using this SIT, coupled with a kairomonal lure, the FB population can be monitored hourly, and population increases can be identified quickly. The SIT processes images at the edge and sends results to the cloud every 40 min for FB monitoring and analysis. It uses a modified you look only once version 8 small (YOLOv8s) object detection network, FB-YOLO, to improve its ability to detect small FBs. The modification is implemented in the network's neck, which aggregates features from the deep and early pyramids of the backbone in the neck. Improved attention to small objects is achieved by incorporating spatially aware features from early pyramids. In addition, the network is integrated with an advanced box selection algorithm called confluence nonmax suppression (NMS-C) to prevent duplicate detections in highly overlapped clusters of FBs. The FB-YOLO achieved an average precision (\u0000<inline-formula><tex-math>$text{mAP}@0.5$</tex-math></inline-formula>\u0000) of 89.97%, a 1.215% improvement over the YOLOv8s network with only 0.324 million additional parameters. Integrating NMS-C further improved the \u0000<inline-formula><tex-math>$text{mAP}@0.5$</tex-math></inline-formula>\u0000 by 0.19%, leading to an overall \u0000<inline-formula><tex-math>$text{mAP}@0.5$</tex-math></inline-formula>\u0000 of 90.16%.","PeriodicalId":100637,"journal":{"name":"IEEE Transactions on AgriFood Electronics","volume":"2 2","pages":"483-496"},"PeriodicalIF":0.0,"publicationDate":"2024-07-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142408683","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Agriculture is facing bigger challenges in the 21st century due to the scarcity of resources. Artificial intelligence is being integrated with agriculture to cater to people's needs, unlocking fresh avenues for sustainability and innovation. One of the crucial agricultural practices is plant growth monitoring to detect plant stress at an early stage. In the past, there have been preliminary attempts at plant growth monitoring using red–green–blue (RGB) and depth images. The major challenge of this approach is the unavailability of the depth camera at the farmers' end. In this work, we have developed a transformer-based plant age RGB-depth fusion knowledge distillation network (PA-RDFKNet), a multi-to-single modal teacher–student network, that exploits the combined knowledge of RGB-depth pairs at the training time to infer the growth using RGB images alone during test time. The model uses a distillation loss that combines response-based, feature-based, and relation-based knowledge distillation techniques in the offline scheme. The proposed knowledge distillation improves the mean squared error for RGB images from 2 to 0.14 weeks. The results are validated on three different datasets.
{"title":"PA-RDFKNet: Unifying Plant Age Estimation through RGB-Depth Fusion and Knowledge Distillation","authors":"Shreya Bansal;Malya Singh;Seema Barda;Neeraj Goel;Mukesh Saini","doi":"10.1109/TAFE.2024.3418818","DOIUrl":"https://doi.org/10.1109/TAFE.2024.3418818","url":null,"abstract":"Agriculture is facing bigger challenges in the 21st century due to the scarcity of resources. Artificial intelligence is being integrated with agriculture to cater to people's needs, unlocking fresh avenues for sustainability and innovation. One of the crucial agricultural practices is plant growth monitoring to detect plant stress at an early stage. In the past, there have been preliminary attempts at plant growth monitoring using red–green–blue (RGB) and depth images. The major challenge of this approach is the unavailability of the depth camera at the farmers' end. In this work, we have developed a transformer-based plant age RGB-depth fusion knowledge distillation network (PA-RDFKNet), a multi-to-single modal teacher–student network, that exploits the combined knowledge of RGB-depth pairs at the training time to infer the growth using RGB images alone during test time. The model uses a distillation loss that combines response-based, feature-based, and relation-based knowledge distillation techniques in the offline scheme. The proposed knowledge distillation improves the mean squared error for RGB images from 2 to 0.14 weeks. The results are validated on three different datasets.","PeriodicalId":100637,"journal":{"name":"IEEE Transactions on AgriFood Electronics","volume":"2 2","pages":"226-235"},"PeriodicalIF":0.0,"publicationDate":"2024-07-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142430796","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-07-02DOI: 10.1109/TAFE.2024.3417644
Maria Doglioni;Matteo Nardello;Davide Brunelli
Smart sensors used for intensive crop monitoring require minimal maintenance and should prioritize ecological sustainability. Consequently, battery-free energy harvesting represents a key aspect of sustainable development in smart agriculture. Plant microbial fuel cells (PMFCs) introduce a cutting-edge renewable energy source that scavenges energy from the symbiotic relationship between a plant and electron-generating bacteria in the soil, potentially supplying power as long as the plant lives. Characterizing PMFCs' power production is challenging, as it depends on many factors, such as soil impedance and plant condition. Electrochemical impedance spectroscopy (EIS) is often used in laboratory tests, but it is inefficient to deploy in off-grid contexts. This article introduces an ultralow power EIS biosensor architecture that utilizes PMFCs as an energy source and for the EIS measure. We prove that ultralow-power EIS is compatible with PMFCs' mW-level power production through an implementation that integrates an EIS analog frontend and PMFC-tailored harvesting electronics. The architecture also facilitates PMFC unloading periods, crucial for PMFC recovery and durability. Experimental results show that a full-range EIS sweep (21.3 mHz–21.8 kHz, 19 points) executed with the proposed architecture requires only 3.64 J. We highlight the potential of cost-effective, self-powered EIS in assisting PMFCs' development into reliable energy sources for battery-free nodes. We also demonstrate that plant state, as well as maximum power point could be monitored through ultralow power EIS measurements.
用于密集作物监测的智能传感器需要最少的维护，并应优先考虑生态可持续性。因此，无电池能源采集是智能农业可持续发展的一个关键方面。植物微生物燃料电池（PMFCs）是一种先进的可再生能源，它能从植物与土壤中产生电子的细菌之间的共生关系中收集能量，只要植物存活，就能为其提供电能。表征 PMFC 的发电量具有挑战性，因为它取决于许多因素，如土壤阻抗和植物状况。电化学阻抗光谱法（EIS）通常用于实验室测试，但在离网环境中使用效率较低。本文介绍了一种超低功耗 EIS 生物传感器架构，该架构利用 PMFC 作为能源并用于 EIS 测量。我们通过集成 EIS 模拟前端和 PMFC 量身定制的采集电子设备的实现，证明超低功耗 EIS 可与 PMFC 的毫瓦级发电量兼容。该架构还有利于 PMFC 的卸载期，这对 PMFC 的恢复和耐用性至关重要。实验结果表明，使用所提议的架构执行一次全范围 EIS 扫频（21.3 mHz-21.8 kHz，19 个点）仅需 3.64 焦耳。我们强调了经济高效的自供电 EIS 在帮助 PMFC 发展成为无电池节点的可靠能源方面的潜力。我们还证明，可以通过超低功耗 EIS 测量来监测电站状态和最大功率点。
{"title":"Plant Microbial Fuel Cells: Energy Sources and Biosensors for battery-Free Smart Agriculture","authors":"Maria Doglioni;Matteo Nardello;Davide Brunelli","doi":"10.1109/TAFE.2024.3417644","DOIUrl":"https://doi.org/10.1109/TAFE.2024.3417644","url":null,"abstract":"Smart sensors used for intensive crop monitoring require minimal maintenance and should prioritize ecological sustainability. Consequently, battery-free energy harvesting represents a key aspect of sustainable development in smart agriculture. Plant microbial fuel cells (PMFCs) introduce a cutting-edge renewable energy source that scavenges energy from the symbiotic relationship between a plant and electron-generating bacteria in the soil, potentially supplying power as long as the plant lives. Characterizing PMFCs' power production is challenging, as it depends on many factors, such as soil impedance and plant condition. Electrochemical impedance spectroscopy (EIS) is often used in laboratory tests, but it is inefficient to deploy in off-grid contexts. This article introduces an ultralow power EIS biosensor architecture that utilizes PMFCs as an energy source and for the EIS measure. We prove that ultralow-power EIS is compatible with PMFCs' mW-level power production through an implementation that integrates an EIS analog frontend and PMFC-tailored harvesting electronics. The architecture also facilitates PMFC unloading periods, crucial for PMFC recovery and durability. Experimental results show that a full-range EIS sweep (21.3 mHz–21.8 kHz, 19 points) executed with the proposed architecture requires only 3.64 J. We highlight the potential of cost-effective, self-powered EIS in assisting PMFCs' development into reliable energy sources for battery-free nodes. We also demonstrate that plant state, as well as maximum power point could be monitored through ultralow power EIS measurements.","PeriodicalId":100637,"journal":{"name":"IEEE Transactions on AgriFood Electronics","volume":"2 2","pages":"460-470"},"PeriodicalIF":0.0,"publicationDate":"2024-07-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10582815","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142408787","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：