Recognition Algorithm Of Smearing Instrument Character Under Complex Lighting Conditions

During the periodical inspections of electronic meters,it is necessary to collect the image through the camera when the voltage changes and recognize instrument characters.However,uneven lighting under natural conditions and smears generated during the digital refresh of the meters cause difficulty in recognition of instrument characters,and there is an urgent need for a set of instrument character recognition methods that can be used for smear under complex lighting conditions.This article focuses on the digital instruments that appear smear during the refresh process,and systematically studies the methods of character image preprocessing,binarization,and character recognition including smearing instrument characters under complex lighting conditions.In view of that the existing methods can not remove the smear in the image and the influence of uneven illumination,etc,an instrument character recognition method with smear under complex lighting conditions is studied.Image enhancement algorithms for uneven illumination are studied.By estimating the illumination component of the image,and then calculating the fusion weights by Principal Component Analysis(PCA)for multi-scale fusion,and adding upsampling and downsizing to the algorithm to reduce the calculation amount of the algorithm,to achieve fast image enhancement.The proposed algorithm can effectively reduce the effect of lighting on the binarization of instrument characters with smear,improve the overall brightness and contrast of the image,and enhance the image to be clear,bright and natural.Its information entropy,average gradient,contrast,and natural image quality evaluator are better than other algorithms in the experiment.In combination with the proposed image enhancement algorithm for uneven illumination,a binary algorithm for smearing instrument characters with slight uneven illumination is proposed,that is,the image enhancement algorithm is used to overcome the effects of slight uneven illumination,and then the gray level distribution statistics of the gray image is used as the input of BP neural network,and the ideal global threshold is used as the output of the network for training and threshold prediction,so as to realize the binarization of the instrument characters with slight uneven illumination.The proposed algorithm has low hardware requirements,fast speed,and good binarization effect,which can effectively overcome the effects of slight uneven lighting and smear.In view of the problem of binarization of smeared instrument characters with serious illumination unevenness,a framework of convolutional neural network based on deep learning is proposed.The data set used by the network is the instrument image in the real environment.The basic idea is to perform dimension reduction and extraction on the input image,then deconvolve to reconstruct the image foreground,and finally output a binary map.The designed network is compared with traditional binarization methods.The experimental results show that the binary image trained by the networkhas clear numbers and no smear,and can efficiently binarize smeare d instrument image with serious illumination unevenness.In view of the traditional instrument character recognition,which needs to extract features and then use a classifier to identify them,the efficiency is low.In this paper,four improvements are made based on the LeNet-5 convolutional neural network.The sigmoid activation function are replaced with the ReLU function.The average pooling method was replaced with the maximum pooling method.The incomplete connection mode of the C3 convolutional layer was changed to the full connection mode.The RMSpro optimization algorithm was used to update the weights and offsets of the networks.The experimental results show that the proposed improved LeNet-5 network can automatically extract features and recognize them.The recognition rate is high,reaching 99.04%,the recognition speed is fast,and it meets the requirements for rapid recognition of instrument characters,which is superior to other algorithms in the experiments.