Recognition Of Communications Equipment Malfunction Based On Machine Vision

Machine vision is very important for the identification and detection of objects, because of high detection precision, non-contact property, stability and so on. The system for communication equipment malfunction detection based on machine vision could detect the devices in real time, moreover it can reduce the consumption of human resources and the malfunction location time significantly.The thesis has studied the malfunction detection for large communications equipment by combining image processing algorithms with neural networks, after analyzing the current research on machine vision. It includes the segmentation and classification for communications equipment tool boards, and the malfunction identification for network ports and ISV3 board’s dual interfaces.In order to solve the recognition problem of large-scale communication equipment components, the classification method combining SIFT and SVM is proposed. Extract the features of sample images by SIFT algorithm, and make cluster analyzing by Kmeans algorithm, Even more, gain the input data based on the minimum Euclidean distance. After analyzing the relationship between the SVM detection accuracy and kernel function type, y value, penalty factor c, the SVM classifier with the best performance is used to detect the target images, and the accuracy rate is more than 92%.Identifying whether the net ports are plugged through the detection algorithm combining gray histogram and BP neural network. Extract the feature points by SIFT algorithm, and make feature matching with using the nearest neighbor method. Then correct the detected image using the optimal transformation matrix gained by RANSAC algorithm. Moreover, get the input data by making histogram statistics for the net port samples. After analyzing the impact of hidden layer nodes and the numbers of iterations on BP neural network, which with the best performance is used to detect the target net ports, and the accuracy rate is reached 96%.Character recognition method is used to determine whether the cables and net ports are matched. Make graying and binarizing for the original image, and separate the characters by finding boundary points according to the distribution of pixels on the binary image. Letters’ sizes are all normalized and divided into 5 rows and 4 columns. Further, get the feature vector of a letter by connecting the numbers of pixels whose gray value is 255 for each small grid. LVQ neural network is trained for testing equipment picture, and marking the wrong matches between cables and net ports, and the accuracy rate is reached 95%.The detection method based on the RGB color model can identify the malfunction of ISV3 board’s dual interfaces effectively. Extract the features of positive and negative sample images as the input by HOG algorithm. Gain the area of four effective interfaces through RGB color information, and judge that whether the interfaces are properly connected by locating and color recognizing for four interfaces, with using the basic methods of image processing.In this thesis, different recognition algorithms are proposed for different test objects. The good results are achieved in the experiments, which lay a good foundation for the next study.