Rigid tank guide fault detection algorithm based on improved YOLOv7

Considering the problems of difficult target detection and recognition and low accuracy caused by factors such as uneven illumination, poor working conditions, complex structure of tank guide and narrow space in coal mine. This paper simulates the complex working environment of the underground mine to carry out different fault conditions experiments, and establishes four categories of channel fault picture data sets. In order to improve the detection accuracy and speed, the following improvements are made on the basis of the YOLOv7 algorithm, and our algorithm is constructed: (1) attention mechanisms are added at different locations of the network; (2) replacement loss function; (3) the original coupling detection head of YOLOv7 is replaced by an efficient decoupled head with implicit knowledge learning. The experimental results show that the mean average precision (mAP) of our algorithm model proposed in this paper reaches 93.2% when the Intersection over Union (IoU) threshold is 0.5, which is 3.2% higher than that of YOLOv7 itself, and the detection speed is also relatively improved by 15.76 frames per second (FPS), reaching 107.50 FPS. While solving the problem of unbalanced improvement of detection accuracy and speed, it also effectively reduces the number of parameters and calculation of the network, which verifies the feasibility of the improved algorithm in this paper.