Development and study of the algorithm for automated recognition of gas meter readings

The relevance of developing an algorithm for automatic recognition of gas meter readings for test benches and real-time monitoring has been substantiated. A review of modern methods for recognizing readings on meter images has been conducted. The choice of the YOLOv10s neural network architecture for recognizing meter readings has been justified. The efficiency of the chosen architecture was validated after training it on a created meter images dataset. Based on the testing results, a second stage of recognition results processing was added, allowing the removal of background fragment detections and sorting the sequence of meter reading digits. In the second stage, a developed convolutional neural network was also employed, ensuring repeated verification of readings and error detection during recognition.