Gap Detection of Aircraft Bell-Shaped Tube Based on Convolutional Neural Networks

Abstract

AbstractThe traditional manual inspection of aircraft bell-shaped tube has the problems of low accuracy and speed. In order to solve this problem, we propose an improved algorithm based on YOLOv4 to detect the gaps of aircraft bell-shaped tube. First, we adjust the size of the anchor boxes through the cluster analysis to match the characteristics of small targets and complex structures better in the detection of gaps. Then, convolutional layers are added after the different feature layers output by the backbone feature extraction network and the spatial pyramid pooling structure to improve the network's ability to extract defect features. Experimental results show that the AP value of the improved YOLOv4 algorithm in gaps detection is 92.21% which is 16.67% higher than the original YOLOv4 algorithm, while the average detection time of a single image is basically same as the original algorithm, and the detection performance is also better than the traditional Canny algorithm and Faster R-CNN.KeywordsYOLOv4Gap detectionAircraft bell-shaped tubeCNNs