基于改进YOLOv7的电机磁瓦缺陷检测方法  

Motor Magnetic Tile Defect Detection Method Based on Improved YOLOv7

作　　者：李欣宇 夏兴华[2] 韩忠华[1] 

机构地区：[1]沈阳建筑大学电气与控制工程学院,辽宁 沈阳 [2]沈阳建筑大学计算机科学与工程学院,辽宁 沈阳

出　　处：《计算机科学与应用》2024年第6期62-71,共10页Computer Science and Application

摘　　要：电机磁瓦在生产过程中会产生残次品从而影响电机性能,为解决磁瓦表面缺陷检测时存在的精检度低、效率低下等问题,提出一种基于改进YOLOv7的磁瓦表面缺陷检测方法。首先在YOLOv7网络中引入了MobileOne模块对网络进行轻量化,加快检测速度。然后引入ECA注意力机制提取更重要的特征,提高检测精度。同时改进SPPCSPC模块,提高对密集缺陷的检测性能。最后引入α-IoU损失函数更加准确的定位目标。实验结果显示,改进后的检测算法相较原网络mAP提升2.1%,参数量减少13%,FPS值为68,与其他主流算法对比能够快速准确地对磁瓦表面缺陷进行检测。In the production process of the magnetic tile of the motor, defective products will be produced, which will affect the performance of the motor. In order to solve the problems of low precision and low efficiency in the surface defect detection of the magnetic tile, a magnetic tile surface defect detection method based on improved YOLOv7 was proposed. Firstly, MobileOne module is introduced into the YOLOv7 network to lighten the network and accelerate the detection speed. Then ECA attention mechanism is introduced to extract more important features and improve the detection accuracy. At the same time, the SPPCSPC module is improved to improve the detection performance of dense defects. Finally, the α-IoU loss function is introduced to locate the target more accurately. The experimental results show that compared with the original network mAP, the improved detection algorithm is 2.1% higher, the number of parameters is 13% lower, and the FPS value is 68. Compared with other mainstream algorithms, the improved detection algorithm can detect the surface defects of magnetic tiles quickly and accurately.

关 键 词：缺陷检测 深度学习 注意力机制 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]