Application of Deep Residual Neural Network to Water Meter Reading Recognition

Aiming at water meter reading pictures taken at arbitrary angles in various complex environments, we propose a robust automatic identification method. This method is based on a deep residual neural network, uses a self-made water meter reading image data set, uses an inverse gradient stochastic descent method to optimize the weights of the RFCN model, and is trained by the network. After repeated iterations, the RFCN model for target detection of water meter reading characters is finally obtained, and then this model is used to detect and identify the characters in the water meter reading box in the water meter picture, and the output result is sorted by the central sort algorithm to achieve a water meter reading Method of identification. The experimental verification of this method shows that its recognition rate is fast and the accuracy is high, which can meet the needs of general companies.