现有的针对PCB裸板的缺陷检测方法存在精确度低、实时性差且难以在移动端部署等问题,本文以YOLO(you only look once)v4算法为基本框架并对其进行改进,提出了一种专门针对PCB裸板的缺陷检测算法。针对YOLOv4算法难以在移动端部署的问题...现有的针对PCB裸板的缺陷检测方法存在精确度低、实时性差且难以在移动端部署等问题,本文以YOLO(you only look once)v4算法为基本框架并对其进行改进,提出了一种专门针对PCB裸板的缺陷检测算法。针对YOLOv4算法难以在移动端部署的问题,采用GhostNet取代CSPDarknet53以轻量化整个检测网络。为弥补YOLOv4算法在多尺度特征融合方面的性能不足,提出了一种双向自适应特征融合网络AF-BiFPN取代PANet网络。为进一步提高模型的检测精度,在AF-BiFPN特征融合网络的采样的过程中插入m-ECANet通道注意力机制。实验结果证明,改进后的YOLOv4算法的模型大小为18.64 MB,检测的平均精度(mean average precision,mAP)为98.39%,检测速度为62.23 FPS,可为实际PCB裸板检测提供理论指导。展开更多
TraditionalCTR recommendation models have concentrated on howto learn low-order and high-order characteristics.The majority of them make many efforts at combining low-order and high-order functions.However,they ignore...TraditionalCTR recommendation models have concentrated on howto learn low-order and high-order characteristics.The majority of them make many efforts at combining low-order and high-order functions.However,they ignore the importance of the attentionmechanism for learning input features.The ECABiNet model is proposed in this article to enhance the performance of CTR.On the one hand,the ECABiNet model can learn the importance of features dynamically via the LayerNorm and ECANET layers.On the other hand,through the use of a biinteraction layer and a DNN layer,it is capable of effectively learning the feature interactions.According to the experimental results on two public datasets,the ECABiNet model is more effective than the previous CTR model.展开更多
文摘现有的针对PCB裸板的缺陷检测方法存在精确度低、实时性差且难以在移动端部署等问题,本文以YOLO(you only look once)v4算法为基本框架并对其进行改进,提出了一种专门针对PCB裸板的缺陷检测算法。针对YOLOv4算法难以在移动端部署的问题,采用GhostNet取代CSPDarknet53以轻量化整个检测网络。为弥补YOLOv4算法在多尺度特征融合方面的性能不足,提出了一种双向自适应特征融合网络AF-BiFPN取代PANet网络。为进一步提高模型的检测精度,在AF-BiFPN特征融合网络的采样的过程中插入m-ECANet通道注意力机制。实验结果证明,改进后的YOLOv4算法的模型大小为18.64 MB,检测的平均精度(mean average precision,mAP)为98.39%,检测速度为62.23 FPS,可为实际PCB裸板检测提供理论指导。
基金supported by Hainan Province Science and Technology Special Fund,which is Research and Application of Intelligent Recommendation Technology Based on Knowledge Graph and User Portrait (No.ZDYF2020039).
文摘TraditionalCTR recommendation models have concentrated on howto learn low-order and high-order characteristics.The majority of them make many efforts at combining low-order and high-order functions.However,they ignore the importance of the attentionmechanism for learning input features.The ECABiNet model is proposed in this article to enhance the performance of CTR.On the one hand,the ECABiNet model can learn the importance of features dynamically via the LayerNorm and ECANET layers.On the other hand,through the use of a biinteraction layer and a DNN layer,it is capable of effectively learning the feature interactions.According to the experimental results on two public datasets,the ECABiNet model is more effective than the previous CTR model.
