Valve status recognition method based on saliency map and improved CNN

Abstract

Valves are widely used in industrial production, and real-time acquisition of valve status is of great significance for production control. Because the manual acquisition method is time-consuming and laborious, this paper proposes a valve status recognition method based on the saliency map and improved convolutional neural network using the valve image: Based on the improved homomorphic filtering method, the valve image is preprocessed to reduce the uneven illumination; The FT algorithm is used to generate the image saliency map and extract the valve body from background image; By stretching the image multi-directional and multi-scale, training set is extended. The improved CNN is constructed and trained to realize the valve status recognition. The experimental results show that the proposed method can effectively identify the status of different valves in the actual environment.