Research on Digital Meter Reading Method of Inspection Robot Based on Deep Learning

Aiming at solving the issue of blurred images and difficult recognition digital meters encountered by inspection robots in process, this paper proposes a deep-learning-based method for image restoration LED identification. Firstly, fast Fourier transform (FFT) is used to perform blur detection on acquired images. Then, are recovered using spatial-attention-improved adversarial neural networks. Finally, meter region extracted polygon-YOLOv5 model corrected via perspective transformation. The digits YOLOv5s model, then recognized CRNN digit recognition. It experimentally verified that improved network achieves 26.562 PSNR metric 0.861 SSIM metric. missing rate reading proposed only 1% accuracy 98%. effectively overcomes blurring problem caused robot during process. This solves problems inaccurate positioning low complex changeable environments, provides new meters.