改进YOLOv7的零件目标检测算法  

Research on part target detection of improved YOLOv7

作　　者：孙可欣 王影[1] SUN Kexin;WANG Ying(College of Information and Control Engineering,Jilin Institute of Chemical Technology,Jilin 132022,Jilin,China)

机构地区：[1]吉林化工学院信息与控制工程学院,吉林吉林132022

出　　处：《智能计算机与应用》2025年第2期175-181,共7页Intelligent Computer and Applications

基　　金：吉林市科技局项目(201750244,20190502118);吉林化工学院科研项目(2018064);吉林化工学院重大科技项目(2016033,2018017)。

摘　　要：为了在不增加计算复杂度的情况下提升零件的识别精度,本文提出一种基于改进YOLOv7的零件目标检测算法。首先,引入SA注意力机制帮助模型更好地聚焦于图像中的关键特征,抑制不必要的特征,使模型能够更精确地聚焦于图像的最重要方面,从而提高检测精度;其次,引入高效卷积模块ConvNeXt替换部分ELAN模块,降低模型的计算复杂度;最后,引入NAM注意力机制构建MP-NAM模块,提高算法对零件的检测能力。实验结果表明,改进YOLOv7算法在自制数据集上的精确率和均值平均精度分别达到0.964和0.925,相比原始YOLOv7算法有较大的提升;同时计算复杂度有显著的下降,约为原始算法的37.4%;与主流目标检测算法相比,改进YOLOv7算法具有更好的检测性能,也证明了改进算法在零件目标检测的有效性。In order to significantly enhance the recognition accuracy of various parts without increasing the overall computational complexity,an advanced part target detection methodology predicated on the improved YOLOv7 algorithm was introduced.This method incorporates several key advancements to optimize performance.In the first place,the Shuffle Attention Mechanism is introduced to help the model better focus on key features in the image and suppress unnecessary features.This adjustment allows the model to focus more precisely on the most important aspects of the image,leading to improved detection accuracy.Futher more,the efficient ConvNeXt convolutional module is introduced to replace some ELAN modules to reduce the computational complexity of the model.The ConvNeXt module helps streamline processing,making the model more efficient and faster.In conclusion,the NAM attention mechanism is introduced to construct MP-NAM module to improve the detection ability of the algorithm.The effectiveness of the improved YOLOv7 algorithm is clearly demonstrated through comprehensive experiments conducted on a custom dataset.The results show that the enhanced algorithm achieves a precision of 0.964 and a mean average precision(mAP)of 0.925.These metrics indicate a substantial improvement over the original YOLOv7 model.At the same time,the computational complexity of the improved model is reduced to approximately 37.4%of that of the original,highlighting a significant reduction in resource usage.Compared to other mainstream target detection algorithms,the improved YOLOv7 algorithm demonstrates notably superior detection performance,which effectively proves the high effectiveness and reliability of the improved algorithm in parts target detection.

关 键 词：零件 目标检测 YOLOv7 注意力机制 ConvNeXt 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术] TP399[自动化与计算机技术—计算机科学与技术]