Improved YOLOx-s industrial pointer instrument detection method

Abstract

Aiming at the problems of slow detection speed and low detection accuracy of existing industrial field pointer instruments, a detection method of industrial pointer instruments based on improved YOLOx-s is proposed. This method is based on YOLOx-s. Firstly, the backbone network of the original model is optimized, and a new feature extraction network is proposed based on the idea of Ghost Module, which improves the detection speed of the network without changing the accuracy. Secondly, introduce attention mechanisms in the backbone network to improve the network's feature extraction capabilities. Finally, the Neck module is improved, streamlined, reduced the complexity of the model and increased the detection rate. Experiments on the self-made industrial field pointer instrument data set show that the improved YOLOx-s algorithm is 1.17% higher than the original algorithm mAP, and 3.2% higher than the two-stage algorithm Faster R-CNN mAP, and it has obvious advantages in detection speed, and 65.45 pictures can be detected per second. The experimental results show that the algorithm has good robustness and universality in industrial pointer instrument detection, and provides guarantee for subsequent accurate acquisition of instrument readings.