Improved Faster R-CNN for the Detection Method of Industrial Control Logic Graph Recognition

Shilin Wu; Yan Wang; Huayu Yang; Pingfeng Wang

doi:10.3389/fbioe.2022.944944

Improved Faster R-CNN for the Detection Method of Industrial Control Logic Graph Recognition

Front Bioeng Biotechnol. 2022 Aug 4:10:944944. doi: 10.3389/fbioe.2022.944944. eCollection 2022.

Authors

Shilin Wu^{1

2}, Yan Wang², Huayu Yang², Pingfeng Wang²

Affiliations

¹ Hubei Key Laboratory of Digital Textile Equipment, Wuhan Textile University, Wuhan, China.
² School of Mechanical Engineering and Automation, Wuhan Textile University, Wuhan, China.

Abstract

In the process of developing the industrial control SAMA logic diagram commonly used in the industrial process control system, there are some problems, that is, the size of logic diagram elements is small, the shape is various, similar element recognition is easily confused, and the detection accuracy is low. In this study, the faster R-CNN network has been improved. The original VGG16 network has been replaced by the ResNet101 network, and the residual value module was introduced to ensure the detailed features of the deep network. Then the industrial control logic diagram dataset was analyzed to improve the anchor size ratio through the K-means clustering algorithm. The candidate box screening problem was optimized by improving the non-maximum suppression algorithm. The elements were distinguished via the combination of the candidate box location and the inherent text, which improved the recognition accuracy of similar elements. An experimental platform was built using the TensorFlow framework based on the Windows system, and the improved method was compared with the original one by the control variable. The results showed that the performance of similar element recognition has been greatly enhanced through an improved faster R-CNN network.

Keywords: K-means; SAMA logic diagram; faster R-CNN; improved faster R-CNN; non-maximal value suppression algorithm.