Bees play a crucial role in the global food chain,pollinating over 75% of food and producing valuable products such as bee pollen,propolis,and royal jelly.However,theAsian hornet poses a serious threat to bee populati...Bees play a crucial role in the global food chain,pollinating over 75% of food and producing valuable products such as bee pollen,propolis,and royal jelly.However,theAsian hornet poses a serious threat to bee populations by preying on them and disrupting agricultural ecosystems.To address this issue,this study developed a modified YOLOv7tiny(You Only Look Once)model for efficient hornet detection.The model incorporated space-to-depth(SPD)and squeeze-and-excitation(SE)attention mechanisms and involved detailed annotation of the hornet’s head and full body,significantly enhancing the detection of small objects.The Taguchi method was also used to optimize the training parameters,resulting in optimal performance.Data for this study were collected from the Roboflow platformusing a 640×640 resolution dataset.The YOLOv7tinymodel was trained on this dataset.After optimizing the training parameters using the Taguchi method,significant improvements were observed in accuracy,precision,recall,F1 score,andmean average precision(mAP)for hornet detection.Without the hornet head label,incorporating the SPD attentionmechanism resulted in a peakmAP of 98.7%,representing an 8.58%increase over the original YOLOv7tiny.By including the hornet head label and applying the SPD attention mechanism and Soft-CIOU loss function,themAP was further enhanced to 97.3%,a 7.04% increase over the original YOLOv7tiny.Furthermore,the Soft-CIOU Loss function contributed to additional performance enhancements during the validation phase.展开更多
Edge Machine Learning(EdgeML)and Tiny Machine Learning(TinyML)are fast-growing fields that bring machine learning to resource-constrained devices,allowing real-time data processing and decision-making at the network’...Edge Machine Learning(EdgeML)and Tiny Machine Learning(TinyML)are fast-growing fields that bring machine learning to resource-constrained devices,allowing real-time data processing and decision-making at the network’s edge.However,the complexity of model conversion techniques,diverse inference mechanisms,and varied learning strategies make designing and deploying these models challenging.Additionally,deploying TinyML models on resource-constrained hardware with specific software frameworks has broadened EdgeML’s applications across various sectors.These factors underscore the necessity for a comprehensive literature review,as current reviews do not systematically encompass the most recent findings on these topics.Consequently,it provides a comprehensive overview of state-of-the-art techniques in model conversion,inference mechanisms,learning strategies within EdgeML,and deploying these models on resource-constrained edge devices using TinyML.It identifies 90 research articles published between 2018 and 2025,categorizing them into two main areas:(1)model conversion,inference,and learning strategies in EdgeML and(2)deploying TinyML models on resource-constrained hardware using specific software frameworks.In the first category,the synthesis of selected research articles compares and critically reviews various model conversion techniques,inference mechanisms,and learning strategies.In the second category,the synthesis identifies and elaborates on major development boards,software frameworks,sensors,and algorithms used in various applications across six major sectors.As a result,this article provides valuable insights for researchers,practitioners,and developers.It assists them in choosing suitable model conversion techniques,inference mechanisms,learning strategies,hardware development boards,software frameworks,sensors,and algorithms tailored to their specific needs and applications across various sectors.展开更多
Deep learning-based intelligent recognition algorithms are increasingly recognized for their potential to address the labor-intensive challenge of manual pest detection.However,their deployment on mobile devices has b...Deep learning-based intelligent recognition algorithms are increasingly recognized for their potential to address the labor-intensive challenge of manual pest detection.However,their deployment on mobile devices has been constrained by high computational demands.Here,we developed GBiDC-PEST,a mobile application that incorporates an improved,lightweight detection algorithm based on the You Only Look Once(YOLO)series singlestage architecture,for real-time detection of four tiny pests(wheat mites,sugarcane aphids,wheat aphids,and rice planthoppers).GBiDC-PEST incorporates several innovative modules,including GhostNet for lightweight feature extraction and architecture optimization by reconstructing the backbone,the bi-directional feature pyramid network(BiFPN)for enhanced multiscale feature fusion,depthwise convolution(DWConv)layers to reduce computational load,and the convolutional block attention module(CBAM)to enable precise feature focus.The newly developed GBiDC-PEST was trained and validated using a multitarget agricultural tiny pest dataset(Tpest-3960)that covered various field environments.GBiDC-PEST(2.8 MB)significantly reduced the model size to only 20%of the original model size,offering a smaller size than the YOLO series(v5-v10),higher detection accuracy than YOLOv10n and v10s,and faster detection speed than v8s,v9c,v10m and v10b.In Android deployment experiments,GBiDCPEST demonstrated enhanced performance in detecting pests against complex backgrounds,and the accuracy for wheat mites and rice planthoppers was improved by 4.5-7.5%compared with the original model.The GBiDC-PEST optimization algorithm and its mobile deployment proposed in this study offer a robust technical framework for the rapid,onsite identification and localization of tiny pests.This advancement provides valuable insights for effective pest monitoring,counting,and control in various agricultural settings.展开更多
The causes of tiny spot defects on the surface of hot-dip galvanized automotive steel sheets were studied using scanning electron microscopy(SEM)and energy dispersive spectrometer(EDS),and effective control measures w...The causes of tiny spot defects on the surface of hot-dip galvanized automotive steel sheets were studied using scanning electron microscopy(SEM)and energy dispersive spectrometer(EDS),and effective control measures were introduced.The results show that rubbing against the top roller after galvanizing is easy due to the local thickness of tiny spot defect location coating;therefore,the surface morphology is different from the normal part.Three kinds of defects,namely zinc slag,small slivers,and pitting,are likely to cause local thickening of the coating after galvanizing,leading to the formation of tiny spots.Therefore,resolving the three types of defects can effectively control the generation of tiny spot defects.Among them,due to the hereditary nature of the small sliver defect,focusing on its control and supervision is necessary.展开更多
为实现田间环境下对玉米苗和杂草的高精度实时检测,本文提出一种融合带色彩恢复的多尺度视网膜(Multi-scale retinex with color restoration,MSRCR)增强算法的改进YOLOv4tiny模型。首先,针对田间环境的图像特点采用MSRCR算法进行图像...为实现田间环境下对玉米苗和杂草的高精度实时检测,本文提出一种融合带色彩恢复的多尺度视网膜(Multi-scale retinex with color restoration,MSRCR)增强算法的改进YOLOv4tiny模型。首先,针对田间环境的图像特点采用MSRCR算法进行图像特征增强预处理,提高图像的对比度和细节质量;然后使用Mosaic在线数据增强方式,丰富目标检测背景,提高训练效率和小目标的检测精度;最后对YOLOv4tiny模型使用K-means++聚类算法进行先验框聚类分析和通道剪枝处理。改进和简化后的模型总参数量降低了45.3%,模型占用内存减少了45.8%,平均精度均值(Mean average precision,mAP)提高了2.5个百分点,在Jetson Nano嵌入式平台上平均检测帧耗时减少了22.4%。本文提出的PruneYOLOv4tiny模型与Faster RCNN、YOLOv3tiny、YOLOv43种常用的目标检测模型进行比较,结果表明:PruneYOLOv4tiny的mAP为96.6%,分别比Faster RCNN和YOLOv3tiny高22.1个百分点和3.6个百分点,比YOLOv4低1.2个百分点;模型占用内存为12.2 MB,是Faster RCNN的3.4%,YOLOv3tiny的36.9%,YOLOv4的5%;在Jetson Nano嵌入式平台上平均检测帧耗时为131 ms,分别是YOLOv3tiny和YOLOv4模型的32.1%和7.6%。可知本文提出的优化方法在模型占用内存、检测耗时和检测精度等方面优于其他常用目标检测算法,能够为硬件资源有限的田间精准除草的系统提供可行的实时杂草识别方法。展开更多
文摘Bees play a crucial role in the global food chain,pollinating over 75% of food and producing valuable products such as bee pollen,propolis,and royal jelly.However,theAsian hornet poses a serious threat to bee populations by preying on them and disrupting agricultural ecosystems.To address this issue,this study developed a modified YOLOv7tiny(You Only Look Once)model for efficient hornet detection.The model incorporated space-to-depth(SPD)and squeeze-and-excitation(SE)attention mechanisms and involved detailed annotation of the hornet’s head and full body,significantly enhancing the detection of small objects.The Taguchi method was also used to optimize the training parameters,resulting in optimal performance.Data for this study were collected from the Roboflow platformusing a 640×640 resolution dataset.The YOLOv7tinymodel was trained on this dataset.After optimizing the training parameters using the Taguchi method,significant improvements were observed in accuracy,precision,recall,F1 score,andmean average precision(mAP)for hornet detection.Without the hornet head label,incorporating the SPD attentionmechanism resulted in a peakmAP of 98.7%,representing an 8.58%increase over the original YOLOv7tiny.By including the hornet head label and applying the SPD attention mechanism and Soft-CIOU loss function,themAP was further enhanced to 97.3%,a 7.04% increase over the original YOLOv7tiny.Furthermore,the Soft-CIOU Loss function contributed to additional performance enhancements during the validation phase.
文摘Edge Machine Learning(EdgeML)and Tiny Machine Learning(TinyML)are fast-growing fields that bring machine learning to resource-constrained devices,allowing real-time data processing and decision-making at the network’s edge.However,the complexity of model conversion techniques,diverse inference mechanisms,and varied learning strategies make designing and deploying these models challenging.Additionally,deploying TinyML models on resource-constrained hardware with specific software frameworks has broadened EdgeML’s applications across various sectors.These factors underscore the necessity for a comprehensive literature review,as current reviews do not systematically encompass the most recent findings on these topics.Consequently,it provides a comprehensive overview of state-of-the-art techniques in model conversion,inference mechanisms,learning strategies within EdgeML,and deploying these models on resource-constrained edge devices using TinyML.It identifies 90 research articles published between 2018 and 2025,categorizing them into two main areas:(1)model conversion,inference,and learning strategies in EdgeML and(2)deploying TinyML models on resource-constrained hardware using specific software frameworks.In the first category,the synthesis of selected research articles compares and critically reviews various model conversion techniques,inference mechanisms,and learning strategies.In the second category,the synthesis identifies and elaborates on major development boards,software frameworks,sensors,and algorithms used in various applications across six major sectors.As a result,this article provides valuable insights for researchers,practitioners,and developers.It assists them in choosing suitable model conversion techniques,inference mechanisms,learning strategies,hardware development boards,software frameworks,sensors,and algorithms tailored to their specific needs and applications across various sectors.
基金support of the Natural Science Foundation of Jiangsu Province,China(BK20240977)the China Scholarship Council(201606850024)+1 种基金the National High Technology Research and Development Program of China(2016YFD0701003)the Postgraduate Research&Practice Innovation Program of Jiangsu Province,China(SJCX23_1488)。
文摘Deep learning-based intelligent recognition algorithms are increasingly recognized for their potential to address the labor-intensive challenge of manual pest detection.However,their deployment on mobile devices has been constrained by high computational demands.Here,we developed GBiDC-PEST,a mobile application that incorporates an improved,lightweight detection algorithm based on the You Only Look Once(YOLO)series singlestage architecture,for real-time detection of four tiny pests(wheat mites,sugarcane aphids,wheat aphids,and rice planthoppers).GBiDC-PEST incorporates several innovative modules,including GhostNet for lightweight feature extraction and architecture optimization by reconstructing the backbone,the bi-directional feature pyramid network(BiFPN)for enhanced multiscale feature fusion,depthwise convolution(DWConv)layers to reduce computational load,and the convolutional block attention module(CBAM)to enable precise feature focus.The newly developed GBiDC-PEST was trained and validated using a multitarget agricultural tiny pest dataset(Tpest-3960)that covered various field environments.GBiDC-PEST(2.8 MB)significantly reduced the model size to only 20%of the original model size,offering a smaller size than the YOLO series(v5-v10),higher detection accuracy than YOLOv10n and v10s,and faster detection speed than v8s,v9c,v10m and v10b.In Android deployment experiments,GBiDCPEST demonstrated enhanced performance in detecting pests against complex backgrounds,and the accuracy for wheat mites and rice planthoppers was improved by 4.5-7.5%compared with the original model.The GBiDC-PEST optimization algorithm and its mobile deployment proposed in this study offer a robust technical framework for the rapid,onsite identification and localization of tiny pests.This advancement provides valuable insights for effective pest monitoring,counting,and control in various agricultural settings.
文摘The causes of tiny spot defects on the surface of hot-dip galvanized automotive steel sheets were studied using scanning electron microscopy(SEM)and energy dispersive spectrometer(EDS),and effective control measures were introduced.The results show that rubbing against the top roller after galvanizing is easy due to the local thickness of tiny spot defect location coating;therefore,the surface morphology is different from the normal part.Three kinds of defects,namely zinc slag,small slivers,and pitting,are likely to cause local thickening of the coating after galvanizing,leading to the formation of tiny spots.Therefore,resolving the three types of defects can effectively control the generation of tiny spot defects.Among them,due to the hereditary nature of the small sliver defect,focusing on its control and supervision is necessary.
文摘为实现田间环境下对玉米苗和杂草的高精度实时检测,本文提出一种融合带色彩恢复的多尺度视网膜(Multi-scale retinex with color restoration,MSRCR)增强算法的改进YOLOv4tiny模型。首先,针对田间环境的图像特点采用MSRCR算法进行图像特征增强预处理,提高图像的对比度和细节质量;然后使用Mosaic在线数据增强方式,丰富目标检测背景,提高训练效率和小目标的检测精度;最后对YOLOv4tiny模型使用K-means++聚类算法进行先验框聚类分析和通道剪枝处理。改进和简化后的模型总参数量降低了45.3%,模型占用内存减少了45.8%,平均精度均值(Mean average precision,mAP)提高了2.5个百分点,在Jetson Nano嵌入式平台上平均检测帧耗时减少了22.4%。本文提出的PruneYOLOv4tiny模型与Faster RCNN、YOLOv3tiny、YOLOv43种常用的目标检测模型进行比较,结果表明:PruneYOLOv4tiny的mAP为96.6%,分别比Faster RCNN和YOLOv3tiny高22.1个百分点和3.6个百分点,比YOLOv4低1.2个百分点;模型占用内存为12.2 MB,是Faster RCNN的3.4%,YOLOv3tiny的36.9%,YOLOv4的5%;在Jetson Nano嵌入式平台上平均检测帧耗时为131 ms,分别是YOLOv3tiny和YOLOv4模型的32.1%和7.6%。可知本文提出的优化方法在模型占用内存、检测耗时和检测精度等方面优于其他常用目标检测算法,能够为硬件资源有限的田间精准除草的系统提供可行的实时杂草识别方法。
