Low-cost Edge Research Articles

Real-time Performance Comparison of Face Detection Algorithms using Raspberry Pi

Abstract This study reviewed state-of-the-art face-detection techniques like Haar cascade, Dlib HOG, MTCNN, and MediaPipe; implemented and tested them on Raspberry Pi, and evaluated their accuracy, speed, and frames per second. Overall, the research underscores the practical challenges of face detection, including varying lighting, facial expressions, occlusions, poses, scale of face, and accessories, and provides valuable insights for developers and researchers working on edge AI applications on low-cost edge devices. The study found that the MediaPipe face detection algorithm demonstrated robust performance, even with low-quality images, and showed good efficiency and resource management on the Raspberry Pi. The research emphasized the importance of considering factors like accuracy, speed, and resource efficiency in face detection on edge devices. The findings suggest that MediaPipe is a strong candidate for applications requiring efficient face detection, affordable and versatile platforms like Raspberry Pi. By focusing on the Raspberry Pi, the study offers a unique perspective on the performance of state-of-the-art face detection algorithms in real-world, resource-constrained environments, making it a significant contribution to the field of face recognition and edge computing.

High-precision and lightweight small-target detection algorithm for low-cost edge intelligence

The proliferation of edge devices driven by advancements in Internet of Things (IoT) technology has intensified the challenge of achieving high-precision small target detection, as it demands extensive computational resources. This amplifies the conflict between the need for precise detection and the requirement for cost-efficiency across numerous edge devices. To solve this problem, this paper introduces an enhanced target detection algorithm, MSGD-YOLO, built upon YOLOv8. The Faster Implementation of CSP Bottleneck with 2 convolutions (C2f) module is enhanced through the integration of the Ghost module and dynamic convolution, resulting in a more lightweight architecture while enhancing feature generation. Additionally, Spatial Pyramid Pooling with Enhanced Local Attention Network (SPPELAN) replaces Spatial Pyramid Pooling Fast (SPPF) to expand the receptive field, optimizing multi-level feature aggregation for improved performance. Furthermore, a novel Multi-Scale Ghost Convolution (MSGConv) and Multi-Scale Generalized Feature Pyramid Network (MSGPFN) are introduced to enhance feature fusion and integrate multi-scale information. Finally, four optimized dynamic convolutional detection heads are employed to capture target features more accurately and improve small target detection precision. Evaluation on the VisDrone2019 dataset shows that compared with YOLOv8-n, MSGD-YOLO improves mAP@50 and mAP@50–95 by 14.1% and 11.2%, respectively. In addition, the model not only achieves a 16.1% reduction in parameters but also attains a processing speed of 24.6 Frames Per Second (FPS) on embedded devices, thereby fulfilling real-time detection requirements.

Fault detection and diagnosis of grid-connected photovoltaic systems using energy valley optimizer based lightweight CNN and wavelet transform

Early fault detection and diagnosis of grid-connected photovoltaic systems (GCPS) is imperative to improve their performance and reliability. Low-cost edge devices have emerged as innovative solutions for real-time monitoring, reducing latency, and improving response times. In this work, a lightweight Convolutional Neural Network (CNN) is designed and fine-tuned using Energy Valley Optimizer (EVO) for fault diagnosis. The CNN input consists of two-dimensional scalograms generated using Continuous Wavelet Transform (CWT). The proposed diagnosis technique demonstrated superior performance compared to benchmark architectures, namely MobileNet, NASNetMobile, and InceptionV3, achieving higher test accuracies and lower losses on binary and multi-fault classification tasks on balanced, unbalanced, and noisy datasets. Further, a quantitative comparison is conducted with similar recent studies. The obtained results indicate good performance and high reliability of the proposed fault diagnosis method.

TinyML for smart agriculture: Comparative analysis of TinyML platforms and practical deployment for maize leaf disease identification

Maize, a vital global food crop, is susceptible to various diseases impacting agricultural yield and quality, necessitating effective monitoring strategies. This study presents a comprehensive approach utilizing advanced Convolutional Neural Network (CNN) architecture for efficient maize leaf disease detection and classification. Leveraging datasets from Kaggle, Harvard Dataverse, and Mendeley, the proposed CNN maize disease detection (MDD) system integrates optimization techniques, data augmentation, batch size adjustments, and fine-tuning strategies. Evaluation through data splitting reveals significant correlations among maize leaf classifications and the influence of data augmentation on image quality. Following training, the CNN-MDD model is deployed on a Microcontroller Unit (MCU) simulation, achieving promising optimization parameters: latency of 7.60 ms, RAM size of 726.60 KB, classifier flash size of 344.70 KB, and an accuracy of 94.60 %. These optimized parameters align well with MCU specifications, highlighting the potential for practical deployment on low-cost edge devices like Arduino Ble 33 Sense. Additionally, this research contributes to the creation of a Ghanaian maize crop pest and disease dataset and the development of an optimal and cost-effective CNN-MDD MCU edge device, aiming to provide local farmers in Ghana and Sub-Saharan Africa with a swift, offline solution for maize disease detection, thus fostering sustainable agricultural practices. We conduct a comparative analysis of Edge Impulse and TensorFlow, two prominent machine-learning platforms for TinyML model development and deployment in various applications, including smart crop disease detection. Utilizing a dataset comprising 12,344 raw images and 7,454 augmented images of healthy and diseased maize leaves, we evaluate the workflow, features, performance, and usability of each platform. Our findings indicate that Edge Impulse holds a slight advantage over TensorFlow in terms of ease of use, reliability, and memory footprint, whereas TensorFlow offers greater flexibility, customization options, and higher accuracy. We conclude that while both platforms are suitable for crop disease detection, a combined approach allows for leveraging TensorFlow's robust training capabilities for complex neural networks alongside the simplicity of Edge Impulse for deploying models on low-powered edge devices. This synergistic approach enhances the effectiveness and efficiency of crop disease detection systems, ultimately benefiting agricultural practices and food security.

Hardhat-wearing detection based on YOLOv5 in Internet-of-Things

<p>Worker safety is paramount in many industries. An essential component of industrial safety protocols involves the proper use of hardhats. However, due to lax safety awareness, many workers neglect to wear hardhats correctly, leading to frequent on-site accidents in China. Traditional detection methods, such as manual inspection and video surveillance, are inefficient and costly. Real-time monitoring of hardhat use is vital to boost compliance with hardhat usage and decrease accident rates. Recently, the advancement of the Internet of Things (IoT) and edge computing has provided an opportunity to improve these methods. In this study, two detection models based on You Only Look Once (YOLO) v5, hardhat-YOLOv5s and hardhat-YOLOv5n, were designed, validated, and implemented, tailored for hardhat detection. First, a public hardhat dataset was enriched to bolster the detection model’s robustness. Then, hardhat detection models were trained using the YOLOv5s and YOLOv5n, each catering to edge computing terminals with varying performance capacities. Finally, the models were validated using image and video data. The experimental results indicated that both models provided high detection precision and satisfied practical application needs. On the augmented public dataset, the hardhat-YOLOv5s and hardhat-YOLOv5n models have a Mean Average Precision (mAP) of 87.9% and 85.5%, respectively, for all six classes. Compared with the hardhat-YOLOv5s model, Parameters and Giga Floating-point Operations (GFLOPs) of the hardhat-YOLOv5n model decrease by 74.8% and 73.4%, respectively, and Frame per Second (FPS) increases by 30.5% on the validation dataset, which is more suitable for low-cost edge computing terminals with less computational power.</p>

TinyML Olive Fruit Variety Classification by Means of Convolutional Neural Networks on IoT Edge Devices

Machine learning (ML) within the edge internet of things (IoT) is instrumental in making significant shifts in various industrial domains, including smart farming. To increase the efficiency of farming operations and ensure ML accessibility for both small and large-scale farming, the need for a low-cost ML-enabled framework is more pressing. In this paper, we present an end-to-end solution that utilizes tiny ML (TinyML) for the low-cost adoption of ML in classification tasks with a focus on the post-harvest process of olive fruits. We performed dataset collection to build a dataset that consists of several varieties of olive fruits, with the aim of automating the classification and sorting of these fruits. We employed simple image segmentation techniques by means of morphological segmentation to create a dataset that consists of more than 16,500 individually labeled fruits. Then, a convolutional neural network (CNN) was trained on this dataset to classify the quality and category of the fruits, thereby enhancing the efficiency of the olive post-harvesting process. The goal of this study is to show the feasibility of compressing ML models into low-cost edge devices with computationally constrained settings for tasks like olive fruit classification. The trained CNN was efficiently compressed to fit into a low-cost edge controller, maintaining a small model size suitable for edge computing. The performance of this CNN model on the edge device, focusing on metrics like inference time and memory requirements, demonstrated its feasibility with an accuracy of classification of more than 97.0% and minimal edge inference delays ranging from 6 to 55 inferences per second. In summary, the results of this study present a framework that is feasible and efficient for compressing CNN models on edge devices, which can be utilized and expanded in many agricultural applications and also show the practical insights for implementing the used CNN architectures into edge IoT devices and show the trade-offs for employing them using TinyML.

Real-time stored product insect detection and identification using deep learning: System integration and extensibility to mobile platforms

Existing stored product insect monitoring methods are time-consuming, costly, and often require specialized equipment or training. This study proposed an integrated insect monitoring system that employs a simple RGB camera and data-driven deep-learning models to detect and identify stored product insect species in warehouses, food facilities, and retail environments. Top-down images of six common insect species were acquired with a setup simulating a conceptualized probe-type monitor under varying lighting and background conditions. These images were preprocessed and manually annotated, resulting in an insect dataset of 2630 images with 14,509 labeled insects. A state-of-the-art computer vision model from YOLO family was selected, and six YOLO variants (YOLOv5s/m/l and YOLOv5s/m/l) were trained and evaluated on the insect dataset. All trained YOLO models delivered an impressive performance in terms of high detection accuracy (above 76% for mAP@[0.50:0.95]) and fast inference time (12–36 ms range). Subsequently, the best-performing YOLOv8l model was integrated and deployed on a mobile device, achieving a good detection performance with average detection speeds of 16 and 29 fps on a desktop computer and smartphone, respectively. The study provided an end-to-end framework for automatic and real-time insect detection and identification in a stored product environment. The lightweight variant of YOLO can be deployed on low-cost edge hardware, mobile devices, or cloud computing. The proposed system is relatively fast, accurate, and inexpensive for insect monitoring and may prove an alternate solution to existing methods. The system would serve as a decision-making tool for stored product facilities managers and can be easily scaled and adapted in a variety of stored product environments.

SmartLite: A DBMS-Based Serving System for DNN Inference in Resource-Constrained Environments

Many IoT applications require the use of multiple deep neural networks (DNNs) to perform various tasks on low-cost edge devices with limited computation resources. However, existing DNN model serving platforms, such as TensorFlow Serving and TorchServe, are resource-intensive and require high-performance GPUs that are often not available on low-cost edge devices. In this paper, we propose SmartLite, a lightweight DBMS that addresses these challenges by storing the parameters and structural information of neural networks as database tables and implementing neural network operators inside the DBMS engine. SmartLite quantizes model parameters as binarized values, applies neural pruning techniques to compress the models, and transforms tensor manipulations into value lookup operations of the DBMS to reduce computation overhead. Experimental results show that SmartLite requires 98% less memory while achieving about a 134% performance speedup compared to Torch-Serve. Our proposed solution addresses the challenges of running multiple DNN models on low-cost edge devices and provides a significant contribution to the field of IoT applications.

An Edge Transfer Learning Approach for Calibrating Soil Electrical Conductivity Sensors

Smart agriculture utilizes Internet of Things (IoT) technologies to enable low-cost electrical conductivity (EC) sensors to support farming intelligence. Due to aging and changes in weather and soil conditions, EC sensors are prone to long-term drift over years of operation. Therefore, regular recalibration is necessary to ensure data accuracy. In most existing solutions, an EC sensor is calibrated by using the standard sensor to build the calibration table. This paper proposes SensorTalk3, an ensemble approach of machine learning models including XGBOOST and Random Forest, which can be executed at an edge device (e.g., Raspberry Pi) without GPU acceleration. Our study indicates that the soil information (both temperature and moisture sensor data) plays an important role in SensorTalk3, which significantly outperforms the existing calibration approaches. The MAPE of SensorTalk3 can be as low as 1.738%, compared to the 7.792% error of the original sensor. Our study indicates that when the errors of uncalibrated moisture and temperature sensors are not larger than 8.3%, SensorTalk3 can accurately calibrate EC. SensorTalk3 can perform model training during data collection at the edge node. When all training data are collected, AI training is also finished at the edge node. Such an AI training approach has not been found in existing edge AI approaches. We also proposed the dual-sensor detection solution to determine when to conduct recalibration. The overhead of this solution is less than twice the optimal detection scenario (which cannot be achieved practically). If the two non-standard sensors are homogeneous and stable, then the optimal detection scenario can be approached. Conventional methods require training calibration AI models in the cloud. However, SensorTalk3 introduces a significant advancement by enabling on-site transfer learning in the edge node. Given the abundance of farming sensors deployed in the fields, performing local transfer learning using low-cost edge nodes proves to be a more cost-effective solution for farmers.

A Unified Parallel CORDIC-Based Hardware Architecture for LSTM Network Acceleration

Deep Neural Networks (DNNs) have recently become the standard tool for solving various practical problems in a wide range of applications with state-of-the-art performance. Recurrent Neural Networks (RNNs) such as Long Short-Term Memory (LSTM) are a subset of DNNs with fully connected single or multi-layer networks. The complex neurons and internal states of LSTM networks enable them to build a memory of events, making them ideal for time series applications. Despite the great potential of LSTM networks, their heterogeneous operations and computational resource requirements create a vast gap when it comes to the fast processing time required in real-time applications using low-power, low-cost edge devices. This work proposes a novel hardware architecture that combines serial-parallel computation with matrix algebra concepts and efficient low-power computer arithmetics for LSTM network acceleration. The hardware is based on a systolic ring of outer-product-based processing elements (PEs) and a reusable single activation function block (AFB). PEs and AFB are implemented using the coordinate rotation digital computer algorithm (CORDIC) in the linear and hyperbolic modes. Unlike most approaches, the proposed hardware can be configured to perform recurrent and non-recurrent fully connected layers (FC) computations, making it suitable for various low-power edge applications. The architecture is validated on the Xilinx PYNQ-Z1 development board using an open-source time series dataset. The implemented design achieves <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$114 \mu s$</tex-math></inline-formula> average latency and <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$1.8GOPS$</tex-math></inline-formula> throughput. The proposed design's low latency and <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$0.438W$</tex-math></inline-formula> power consumption makes it suitable for resource-constrained edge platforms.

PATCH

Recent advancements in deep learning have shown that multimodal inference can be particularly useful in tasks like autonomous driving, human health, and production line monitoring. However, deploying state-of-the-art multimodal models in distributed IoT systems poses unique challenges since the sensor data from low-cost edge devices can get corrupted, lost, or delayed before reaching the cloud. These problems are magnified in the presence of asymmetric data generation rates from different sensor modalities, wireless network dynamics, or unpredictable sensor behavior, leading to either increased latency or degradation in inference accuracy, which could affect the normal operation of the system with severe consequences like human injury or car accident. In this paper, we propose PATCH, a framework of speculative inference to adapt to these complex scenarios. PATCH serves as a plug-in module in the existing multimodal models, and it enables speculative inference of these off-the-shelf deep learning models. PATCH consists of 1) a Masked-AutoEncoder-based cross-modality imputation module to impute missing data using partially-available sensor data, 2) a lightweight feature pair ranking module that effectively limits the searching space for the optimal imputation configuration with low computation overhead, and 3) a data alignment module that aligns multimodal heterogeneous data streams without using accurate timestamp or external synchronization mechanisms. We implement PATCH in nine popular multimodal models using five public datasets and one self-collected dataset. The experimental results show that PATCH achieves up to 13% mean accuracy improvement over the state-of-art method while only using 10% of training data and reducing the training overhead by 73% compared to the original cost of retraining the model.

AIoT-Driven Edge Computing for Rural Small-Scale Poultry Farming: Smart Environmental Monitoring and Anomaly Detection for Enhanced Productivity

The growing demand for chicken production has emphasized the importance of maintaining optimal conditions to improve quality and productivity.The integration of Artificial Intelligence (AI) and the Internet of Things(IoT) is recommended for the efficient management of the farm's environment. A potential solution is presented in this paper, utilizing IoT-based sensor nodes with ARM Cortex M3 - LPC 1769 and LORA technology to monitor chicken farms across diverse regions.The proposed solution incorporates a low-cost edge computing server-Jetson Nano device equipped with a machine learning model to categorize and monitor live environmental conditions in poultry farms. Real-time data from various branches is collected and analyzed using machine learning classification techniques including logistic regression, K nearest neighbors, and support vector machines.The performance of these algorithms is compared to identify the most effective approach. Upon evaluation, the K nearest neighbors emerges as the superior performer, achieving an impressive accuracy of 99.72% and an execution duration of 0.087 seconds on the Jetson Nano edge computing device. This cost-effective technology is tailored for small businesses in regions where farmers can gain valuable insights from data-driven decisions and closely monitor their operations. By incorporating AIoT into farm management, the challenges faced by small-scale poultry farming can be addressed, empowering farmers with enlightened techniques to improve overall productivity and quality.

A deep-learning framework running on edge devices for handgun and knife detection from indoor video-surveillance cameras

The early detection of handguns and knives from surveillance videos is crucial to enhance people’s safety. Despite the increasing development of Deep Learning (DL) methods for general object detection, weapon detection from surveillance videos still presents open challenges. Among these, the most significant are: (i) the very small size of the weapons with respect to the camera field of view and (ii) the need of a real-time feedback, even when using low-cost edge devices for computation. Complex and recently-developed DL architectures could mitigate the former challenge but do not satisfy the latter one. To tackle such limitation, the proposed work addresses the weapon-detection task from an edge perspective. A double-step DL approach was developed and evaluated against other state-of-the-art methods on a custom indoor surveillance dataset. The approach is based on a first Convolutional Neural Network (CNN) for people detection which guides a second CNN to identify handguns and knives. To evaluate the performance in a real-world indoor environment, the approach was deployed on a NVIDIA Jetson Nano edge device which was connected to an IP camera. The system achieved near real-time performance without relying on expensive hardware. The results in terms of both COCO Average Precision (AP = 79.30) and Frames per Second (FPS = 5.10) on the low-power NVIDIA Jetson Nano pointed out the goodness of the proposed approach compared with the others, encouraging the spread of automated video surveillance systems affordable to everyone.

Real-Time Intelligent Detection System for Illegal Wearing of On-Site Power Construction Worker Based on Edge-YOLO and Low-Cost Edge Devices

Ensuring personal safety and preventing accidents are critical aspects of power construction safety supervision. However, current monitoring methods are inefficient and unreliable as most of them rely on manual monitoring and transmission, which results in slow detection and delayed warnings regarding violations. To overcome these challenges, we propose an intelligent detection system that can accurately identify instances of illegal wearing of power construction workers in real-time. Firstly, we integrated the squeeze-and-excitation (SE) module into our convolutional neural network to enhance detection accuracy. This module effectively prioritizes informative features while suppressing less relevant ones, resulting in improved overall performance. Secondly, we present an embedded real-time detection system that utilizes Jetson Xavier NX and Edge-YOLO. This system promptly detects and alerts power construction workers of instances of illegal wearing behavior. To ensure a lightweight implementation, we design appropriate detection heads based on target size and distribution, reducing model parameters while enhancing detection speed and minimizing accuracy loss. Additionally, we employed data augmentation to enhance the system’s robustness. Our experimental results demonstrate that our improved Edge-YOLO model achieves high detection precision and recall rates of 0.964 and 0.966, respectively, with a frame rate of 35.36 frames per second when deployed on Jetson Xavier NX. Therefore, Edge-YOLO proves to be an ideal choice for intelligent real-time detection systems, providing superior accuracy and speed performance compared to the original YOLOv5s model and other models in the YOLO series for safety monitoring at construction sites.

Neural Network Exploration for Keyword Spotting on Edge Devices

The introduction of artificial neural networks to speech recognition applications has sparked the rapid development and popularization of digital assistants. These digital assistants constantly monitor the audio captured by a microphone for a small set of keywords. Upon recognizing a keyword, a larger audio recording is saved and processed by a separate, more complex neural network. Deep neural networks have become an effective tool for keyword spotting. Their implementation in low-cost edge devices, however, is still challenging due to limited resources on board. This research demonstrates the process of implementing, modifying, and training neural network architectures for keyword spotting. The trained models are also subjected to post-training quantization to evaluate its effect on model performance. The models are evaluated using metrics relevant to deployment on resource-constrained systems, such as model size, memory consumption, and inference latency, in addition to the standard comparisons of accuracy and parameter count. The process of deploying the trained and quantized models is also explored through configuring the microcontroller or FPGA onboard the edge devices. By selecting multiple architectures, training a collection of models, and comparing the models using the techniques demonstrated in this research, a developer can find the best-performing neural network for keyword spotting given the constraints of a target embedded system.

An embedded system for remote monitoring and fault diagnosis of photovoltaic arrays using machine learning and the internet of things

In this paper a novel embedded system for remote monitoring and fault diagnosis of photovoltaic systems is introduced. The idea is to embed machine leaning algorithms into a low-cost edge device for real-time deployment. First, an artificial neural network is developed to detect faults. Then an effective stacking ensemble learning algorithm is developed to classify the nature of the fault. The method performance is evaluated through common error metrics such as RMSE, MAE, MAPE, r and confusion matrix. Additional algorithms are also embedded into the edge device in order to remotely control the photovoltaic array parameters. Users can be notified by email and SMS about the state of their photovoltaic array. The Blynk IoT platform is used to monitor remotely the photovoltaic array parameters. The experimental results demonstrate the ability of the proposed embedded system to diagnose and monitor the photovoltaic array with a good accuracy.

A low-cost edge AI-chip-based system for real-time algae species classification and HAB prediction

Harmful Algal Blooms (HAB) are damaging to ecosystem functions and pose challenges to environmental and fisheries management. The key to HAB management and understanding the complex algal growth dynamics is the development of robust systems for real-time monitoring of algae populations and species. Previous algae classification studies mainly rely on the combination of an in-situ imaging flow cytometer and an off-site lab-based algae classification model such as Random Forest (RF) for the analysis of high-throughput images. An on-site AI algae monitoring system on top of an edge AI chip embedded with the proposed Algal Morphology Deep Neural Network (AMDNN) model is developed to achieve real-time algae species classification and HAB prediction. Based on a detailed examination of real-world algae images, dataset augmentation is first performed: consisting of orientation, flipping, blurring, and Resizing with Aspect ratio Preserved (RAP). The dataset augmentation is shown to significantly improve classification performance which is superior to that of the competitive RF model. And the attention heatmaps show that for relatively regular-shaped algal species (e.g., Vicicitus), the model weights the color and texture information heavily; while the shape-related features are more important for complex-shaped algae (e.g., Chaetoceros). The AMDNN is tested on a dataset of 11,250 algae images containing the 25 most common HAB classes in Hong Kong subtropical waters with 99.87% test accuracy. Based on the fast and accurate algae classification, the AI-chip-based on-site system is applied to a one-month dataset in February 2020; the predicted trends of total cell counts and targeted HAB species counts are in good agreement with observations. The proposed edge AI algae monitoring system provides a platform for the development of practical HAB early warning systems that can effectively support environmental risk and fisheries management.

A Fine-Grained Attention Model for High Accuracy Operational Robot Guidance

Deep learning enhanced Internet of Things (IoT) is advancing the transformation toward smart manufacturing. Intelligent robot guidance is one of the most potential deep learning + IoT applications in the manufacturing industry. However, low costs, efficient computing, and extremely high localization accuracy are mandatory requirements for vision robot guidance, particularly in operational factories. Therefore, in this work, a low-cost edge computing-based IoT system is developed based on an innovative fine-grained attention model (FGAM). FGAM integrates a deep-learning-based attention model to detect the region of interest (ROI) and an optimized conventional computer vision model to perform fine-grained localization concentrating on the ROI. Trained with only 100 images collected from real production line, the proposed FGAM has shown superior performance over multiple benchmark models when validated using operational data. Eventually, the FGAM-based edge computing system has been deployed on a welding robot in a real-world factory for mass production. After the assembly of about 6000 products, the deployed system has achieved averaged overall process and transmission time down to 200 ms and overall localization accuracy up to 99.998%.

AD-YOLOv5s based UAV detection for low altitude security

UAV (Unmanned Aerial Vehicle) black flight at low altitude could cause serious safety risks. Consequently, it is crucial to detect and manage low altitude small UAVs. The existing methods of low altitude small UAV detection suffer from problems such as high false alarm rate, and poor real-time performance. In order to solve the above problems, we present a novel approach, named AD-YOLOv5s, to achieve low altitude small UAV detection with high precision and high real-time performance. Firstly, the feature enhancement method is used to expand the dataset. We optimize the model feature fusion, the prediction head structure, and the loss function. Based on the CBAM (Convolutional Block Attention Module) attention mechanism, feature enhancement is performed to improve the detection accuracy. Secondly, the ghost module and depthwise separable convolution are used to reduce the number of parameters of the model, and we propose the method of lightweight design of model to improve the detection speed. Compared with the YOLOv5s model, the experiment result shows that our proposed AD-YOLOv5s model improves the value of mAP by 2.2% and the value of Recall by 1.8%, reduces the value of GFLOPs by 29.9% and parameters by 38.8%, and achieves 27.6 FPS when the proposed model deploy on a low-cost edge computing device (jetson nano).

Aerial-Based Weed Detection Using Low-Cost and Lightweight Deep Learning Models on an Edge Platform

Highlights Lightweight deep learning models were trained on an edge device to identify weeds in aerial images. A customized configuration file was setup to train the models. These models were deployed to detect weeds in aerial images and videos (near real-time). CSPMobileNet-v2 and YOLOv4-lite are recommended models for weed detection using edge platform. Abstract. Deep learning (DL) techniques have proven to be a successful approach in detecting weeds for site-specific weed management (SSWM). In the past, most of the research work has trained and deployed pre-trained DL models on high-end systems coupled with expensive graphical processing units (GPUs). However, only a limited number of research studies have used DL models on an edge system for aerial-based weed detection. Therefore, while focusing on hardware cost minimization, eight DL models were trained and deployed on an edge device to detect weeds in aerial-image context and videos in this study. Four large models, namely CSPDarkNet-53, DarkNet-53, DenseNet-201, and ResNet-50, along with four lightweight models, CSPMobileNet-v2, YOLOv4-lite, EfficientNet-B0, and DarkNet-Ref, were considered for training a customized DL architecture. Along with trained model performance scores (average precision score, mean average precision (mAP), intersection over union, precision, and recall), other model metrics to assess edge system performance such as billion floating-point operations/s (BFLOPS), frame rates/s (FPS), and GPU memory usage were also estimated. The lightweight CSPMobileNet-v2 and YOLOv4-lite models outperformed others in detecting weeds in aerial image context. These models were able to achieve a mAP score of 83.2% and 82.2%, delivering an FPS of 60.9 and 61.1 during near real-time weed detection in aerial videos, respectively. The popular ResNet-50 model achieved a mAP of 79.6%, which was the highest amongst all the large models deployed for weed detection tasks. Based on the results, the two lightweight models, namely, CSPMobileNet-v2 and YOLOv4-lite, are recommended, and they can be used on a low-cost edge system to detect weeds in aerial image context with significant accuracy. Keywords: Aerial image, Deep learning, Edge device, Precision agriculture, Weed detection.