State Recognition of Electric Control Cabinet Switches Based on CNNs
Xingang Mou, Conglin Leng, Xiao Long Zhou, Yichao Cai · 2018
When using the CNNs for image processing, it is more easy to learn the invariant features of the image without too much preprocessing. This is important for the task of image classification and recognition which need to extract deep features of the object. This paper proposes a switch state recognition algorithm which combines digital image processing technique and CNNs for the electrical control cabinet switches in substations. Firstly, use perspective transformation method to rectify the original image which is positioned and segmented using the projections of horizontal and vertical gradient. Then label the segmented images to get sample sets for training CNNs later. After that, design the architecture of CNNs with image processing knowledge and use the designed CNNs to extract the key features of switch state, build and train the CNNs model using labeled “disconnected” and “closed” switches. Finally, the state of the switch is recognized by the trained model and showed in the image. The experimental results show that the algorithm has the characteristics of high recognition accuracy, good robustness and fast recognition speed.