Research and Application of Pointer Meter Reading Algorithm Based on Attention U2-Net

As one of the most common measuring instruments, the pointer-type analog meter is widely used in factories and substations because of its simple structure and high stability. However the mechanical structure makes it almost impossible to provide digital interfaces directly, thus the most common way of getting reading is still recording by individual, but this could cause high labor costs and lack of reliability. To address this issue, we proposed a compact attention based salient object detection model, which combined the U ² -Net with a spatial attention module to make the model pay more attention to the pointer and scale area. And the model has been trained and tested on the collected pointer meter datasets from a power substations. In addition, the model has been deployed on the AIoT device to achieve a set of high accuracy and robust remote pointer meter reading system. The test on real scenarios proves the accuracy and robustness of our system.