A Lightweight Electronic Water Pump Shell Defect Detection Method Based on Improved YOLOv5s

Abstract

For surface defects in electronic water pump shells, the manual detection efficiency is low, prone to misdetection and leak detection, and encounters problems, such as uncertainty. To improve the speed and accuracy of surface defect detection, a lightweight detection method based on an improved YOLOv5s method is proposed to replace the traditional manual detection methods. In this method, the MobileNetV3 module replaces the backbone network of YOLOv5s, depth-separable convolution is introduced, the parameters and calculations are reduced, and CIoU_Loss is used as the loss function of the boundary box regression to improve its detection accuracy. A dataset of electronic pump shell defects is established, and the performance of the improved method is evaluated by comparing it with that of the original method. The results show that the parameters and FLOPs are reduced by 49.83% and 61.59%, respectively, compared with the original YOLOv5s model, and the detection accuracy is improved by 1.74%, which is an indication of the superiority of the improved method. To further verify the universality of the improved method, it is compared with the results using the original method on the PASCALVOC2007 dataset, which verifies that it yields better performance. In summary, the improved lightweight method can be used for the real-time detection of electronic water pump shell defects.

A lightweight improvement of YOLOv5 for insulator fault detection

“…used the YOLOv2 network, combined with edge detection, line detection, image rotation, and other methods to realize the defect diagnosis of transmission line insulators, and finally, the average recognition accuracy reached 80.1%, and the recognition speed reached 30 frames/s. Wu et al [5] used the lightweight YOLOv3 to screen the fault defects of insulators, and the results also achieved an average accuracy of nearly 80%, and the speed reached 26 frames/s.…”