Convolutional Neural Networks based Defect Detection Method for Smart Meter LCD Screen

In this paper, we introduce and experimentally demonstrate a novel defect detection method for the LCD screen of the smart meter. The images from the camera firstly undergo a series of image processing, such as rotate, grayscale, binarization, denoise and region of interest (ROI) positioning. Thus, the information of the images, which are unrelated to defect detection, is effectively removed. And then, the Convolutional Neural Network (CNN), which includes four convolution layers, four pooling layers and two fully connected layers, is designed and trained to recognize the defects in the processed images. A testbench is made, in which four LCD screens can be tested simultaneously. The experimental results show that the Accuracy, Precision, Recall and F1-score of our method are 97.25%, 97.49%, 97% and 97.24%, respectively, which is better than support vector machine (SVM) model. The method can be widely applied in the defect detection of the LCD screen.