Printed Circuit Boards(PCBs)are materials used to connect components to one another to form a working circuit.PCBs play a crucial role in modern electronics by connecting various components.The trend of integrating mo...Printed Circuit Boards(PCBs)are materials used to connect components to one another to form a working circuit.PCBs play a crucial role in modern electronics by connecting various components.The trend of integrating more components onto PCBs is becoming increasingly common,which presents significant challenges for quality control processes.Given the potential impact that even minute defects can have on signal traces,the surface inspection of PCB remains pivotal in ensuring the overall system integrity.To address the limitations associated with manual inspection,this research endeavors to automate the inspection process using the YOLOv8 deep learning algorithm for real-time fault detection in PCBs.Specifically,we explore the effectiveness of two variants of the YOLOv8 architecture:YOLOv8 Small and YOLOv8 Nano.Through rigorous experimentation and evaluation of our dataset which was acquired from Peking University’s Human-Robot Interaction Lab,we aim to assess the suitability of these models for improving fault detection accuracy within the PCB manufacturing process.Our results reveal the remarkable capabilities of YOLOv8 Small models in accurately identifying and classifying PCB faults.The model achieved a precision of 98.7%,a recall of 99%,an accuracy of 98.6%,and an F1 score of 0.98.These findings highlight the potential of the YOLOv8 Small model to significantly improve the quality control processes in PCB manufacturing by providing a reliable and efficient solution for fault detection.展开更多
在复杂的自然环境中绿色柑橘生长形态各异,颜色与背景色相近,为有效识别绿色柑橘,提出一种基于混合注意力机制并改进YOLOv5模型的柑橘识别方法。首先,改进YOLOv5的网络结构,在主干网络中添加混合注意力机制,即在主干网络中的第2层嵌入SE...在复杂的自然环境中绿色柑橘生长形态各异,颜色与背景色相近,为有效识别绿色柑橘,提出一种基于混合注意力机制并改进YOLOv5模型的柑橘识别方法。首先,改进YOLOv5的网络结构,在主干网络中添加混合注意力机制,即在主干网络中的第2层嵌入SE(squeeze and excitation)注意力,第11层嵌入CA(coordinate attention)注意力;其次,改进网络模型特征融合结构,将YOLOv5模型Concat特征融合操作的下层分支放在模型C3模块之前,再与另一条上层分支进行特征融合;最后,改进模型分类损失函数,将YOLOv5模型的分类损失函数改成Varifocal Loss函数,加强绿色柑橘特征信息的提取,提高绿色柑橘检测精度。根据自然环境和柑橘自身的特点,对自建数据集进行分类,设计3组不同分类场景下柑橘的对比试验以验证其有效性。试验结果表明,改进后的YOLOv5-SC模型准确率为91.74%,平均精度为95.09%,F1为89.56%,在自然环境下对绿色柑橘的识别具有更高的准确率和更好的鲁棒性,为绿色水果智能采摘提供技术支持。展开更多
[目的]本文旨在解决在自然环境下不同成熟度苹果目标检测精度较低的问题。[方法]提出了一种改进的YOLOv5s模型SODSTR-YOLOv5s(YOLOv5s with small detection layer and omni-dimensional dynamic convolution and swin transformer bloc...[目的]本文旨在解决在自然环境下不同成熟度苹果目标检测精度较低的问题。[方法]提出了一种改进的YOLOv5s模型SODSTR-YOLOv5s(YOLOv5s with small detection layer and omni-dimensional dynamic convolution and swin transformer block),用于不同成熟度苹果检测。首先改进YOLOv5s的多尺度目标检测层,在Prediction中构建检测160×160特征图的检测头,提高小尺寸的不同成熟度苹果的检测精度;其次在Backbone结构中融合Swin Transformer Block,加强同级成熟度的苹果纹理特征融合,弱化纹理特征分布差异带来的消极影响,提高模型泛化能力;最后将Neck结构的Conv模块替换为动态卷积模块ODConv,细化局部特征映射,实现局部苹果细粒度特征的充分提取。基于不同成熟度苹果数据集进行试验,验证改进模型的性能。[结果]改进模型SODSTR-YOLOv5s检测的精确率、召回率、平均精度均值分别为89.1%、95.5%、93.6%,高、中、低成熟度苹果平均精度均值分别为94.1%、93.1%、93.7%,平均检测时间为16 ms,参数量为7.34 M。相比于YOLOv5s模型,改进模型SODSTR-YOLOv5s精确率、召回率、平均精度均值分别提高了3.8%、5.0%、2.9%,参数量和平均检测时间分别增加了0.32 M和5 ms。[结论]改进模型SODSTR-YOLOv5s提升了在自然环境下对不同成熟度苹果的检测能力,能较好地满足实际采摘苹果的检测要求。展开更多
文摘Printed Circuit Boards(PCBs)are materials used to connect components to one another to form a working circuit.PCBs play a crucial role in modern electronics by connecting various components.The trend of integrating more components onto PCBs is becoming increasingly common,which presents significant challenges for quality control processes.Given the potential impact that even minute defects can have on signal traces,the surface inspection of PCB remains pivotal in ensuring the overall system integrity.To address the limitations associated with manual inspection,this research endeavors to automate the inspection process using the YOLOv8 deep learning algorithm for real-time fault detection in PCBs.Specifically,we explore the effectiveness of two variants of the YOLOv8 architecture:YOLOv8 Small and YOLOv8 Nano.Through rigorous experimentation and evaluation of our dataset which was acquired from Peking University’s Human-Robot Interaction Lab,we aim to assess the suitability of these models for improving fault detection accuracy within the PCB manufacturing process.Our results reveal the remarkable capabilities of YOLOv8 Small models in accurately identifying and classifying PCB faults.The model achieved a precision of 98.7%,a recall of 99%,an accuracy of 98.6%,and an F1 score of 0.98.These findings highlight the potential of the YOLOv8 Small model to significantly improve the quality control processes in PCB manufacturing by providing a reliable and efficient solution for fault detection.
文摘在复杂的自然环境中绿色柑橘生长形态各异,颜色与背景色相近,为有效识别绿色柑橘,提出一种基于混合注意力机制并改进YOLOv5模型的柑橘识别方法。首先,改进YOLOv5的网络结构,在主干网络中添加混合注意力机制,即在主干网络中的第2层嵌入SE(squeeze and excitation)注意力,第11层嵌入CA(coordinate attention)注意力;其次,改进网络模型特征融合结构,将YOLOv5模型Concat特征融合操作的下层分支放在模型C3模块之前,再与另一条上层分支进行特征融合;最后,改进模型分类损失函数,将YOLOv5模型的分类损失函数改成Varifocal Loss函数,加强绿色柑橘特征信息的提取,提高绿色柑橘检测精度。根据自然环境和柑橘自身的特点,对自建数据集进行分类,设计3组不同分类场景下柑橘的对比试验以验证其有效性。试验结果表明,改进后的YOLOv5-SC模型准确率为91.74%,平均精度为95.09%,F1为89.56%,在自然环境下对绿色柑橘的识别具有更高的准确率和更好的鲁棒性,为绿色水果智能采摘提供技术支持。
文摘[目的]本文旨在解决在自然环境下不同成熟度苹果目标检测精度较低的问题。[方法]提出了一种改进的YOLOv5s模型SODSTR-YOLOv5s(YOLOv5s with small detection layer and omni-dimensional dynamic convolution and swin transformer block),用于不同成熟度苹果检测。首先改进YOLOv5s的多尺度目标检测层,在Prediction中构建检测160×160特征图的检测头,提高小尺寸的不同成熟度苹果的检测精度;其次在Backbone结构中融合Swin Transformer Block,加强同级成熟度的苹果纹理特征融合,弱化纹理特征分布差异带来的消极影响,提高模型泛化能力;最后将Neck结构的Conv模块替换为动态卷积模块ODConv,细化局部特征映射,实现局部苹果细粒度特征的充分提取。基于不同成熟度苹果数据集进行试验,验证改进模型的性能。[结果]改进模型SODSTR-YOLOv5s检测的精确率、召回率、平均精度均值分别为89.1%、95.5%、93.6%,高、中、低成熟度苹果平均精度均值分别为94.1%、93.1%、93.7%,平均检测时间为16 ms,参数量为7.34 M。相比于YOLOv5s模型,改进模型SODSTR-YOLOv5s精确率、召回率、平均精度均值分别提高了3.8%、5.0%、2.9%,参数量和平均检测时间分别增加了0.32 M和5 ms。[结论]改进模型SODSTR-YOLOv5s提升了在自然环境下对不同成熟度苹果的检测能力,能较好地满足实际采摘苹果的检测要求。
