Target Detection of Pointer Instrument based on Deep Learning

Abstract

This paper proposed an instrument target detection algorithm based on yolov3 network for the drawbacks caused by manual inspection of pointer instruments in complex industrial environments. Firstly, the algorithm improved the model convergence speed by introducing the k-means++ algorithm to cluster out 9 sets of initial anchor boxes suitable for the pointer meter data set. Moreover, by combining the channel attention mechanism with spatial attention mechanism in the yolov3 backbone network, the extraction of shallow features was further improved by adding two residual blocks to the second residual block, then a new model yolov3-CBAM (Convolutional Block Attention Module) was formed. In addition, the mean average accuracy (map) of the training and testing of the three types of instruments on the data set reaches 90.8% by the results, which is about 2.1% higher than the original yolov3. This algorithm has obvious advantages in the patrol inspection and identification of industrial instruments.