Automatic Recognition of Pointer Meter Reading Based on Yolov4 and Improved U-net Algorithm

Pointer meters are widely used in life and industrial production due to their simple structure, good reliability, high-cost performance, and intuitive and convenient reading. With the improvement of artificial intelligence and automation level, the application of drones in substation inspections has become more and more extensive, which has brought convenience to image collection, but the problem of the accuracy of reading recognition of pointer meters has not been well resolved. Using the current mainstream algorithms, due to the complexity of the surrounding environment, dust pollution, and other factors, the obtained image contains a lot of noise, which affects the accuracy of reading recognition. In this paper, YOLOV4 is used to detect the position of the dial in the images and classify the meters; then combined with the characteristics of the meter image, the improved U-Net image segmentation technology to effectively extract the pointers in the area, accurately identify the meter readings, and use the collected power equipment sample training data set to test the algorithm. The improved U-Net improves the learning ability of the shallow information of the image, reduces the loss of edge features, and improves the segmentation effect of the target.