At present,the information entry method of the electric meter version still manually compares and verifies with the database,which has the problems of inefficiency and is difficult to make sure the accuracy.An accurate classification method for old and new versions of electric metersin the highly mixed background was proposed by using real-shot electricity meter images.Firstly,the version identification ROI was obtained,and the gray level run-length matrix (GLRLM) feature was extracted.Then normalization and principal component analysis (PCA) were performed on the data.Finally,the support vector machine (SVM) model of linear kernel function was used as the best model for classification experiments.Meanwhile,different texture feature algorithms combined with different classification models were used to evaluate the method performance.The experimental result shows that the GLRLM-SVM is better and faster than other classification models,the accuracy rate is as high as 98.95%,it can meet the quantity and accuracy requirements of the annual inspection of the electric meter.