Research on Recognition Algorithm of Industrial Instrument Based on Convolutional Neural Network

Aiming at the problem of low accuracy of image classification and positioning recognition of industrial instruments, an instrument reading recognition algorithm based on bilinear difference improved Mask-RCNN was proposed by taking pointer instrument as the research object. Firstly, the algorithm adopted Mask-RCNN instance segmentation network to distinguish the pointer region and the dashboard region. Then, the position and direction of the instrument panel were corrected by the perspective transformation algorithm based on irregular ellipse. Finally, the principal component analysis (PCA) algorithm was used to fit the pointer line, and the angle method was utilized to recognize the pointer reading. The results showed that compared with the Mask-RCNN algorithm before improvement, the target positioning accuracy and instance segmentation accuracy of the improved Mask-RCNN algorithm are improved by about 3% and 2%, respectively. The average relative error and average quoted error between the proposed algorithm and manual reading values are 1.14% and 0.253%, respectively, and the instrument recognition error is small, which indicated that the proposed method can effectively improve the accuracy of instrument image classification and positioning, and can be widely applied in the field of industrial instruments.