A High-Voltage Electric Switch Classification System Based on K-Nearest Neighbor Classifier

Classification of high-voltage electric switches is an important operation in industrial manufacturing. However, the electrical shock hazards make it dangerous to human. Therefore, classifying high-voltage electric switches automatically is of great interest for factories. For this purpose, we designed a system based on k-nearest neighbor algorithm and bag of visual words model, which performs well in classifying 3 states of highvoltage electric switches. We achieve the classifying task by 3 steps: extracting features of high-voltage electric switch pictures by using SIFT algorithm; clustering SIFT features of all training pictures as visual words and set up a bag of visual words model; calculating the visual words frequency of each picture and using them as inputs of k-nearest neighbor classifier. With the trained model, we extract SIFT features and count visual words frequency of a new picture to be classified, then predict its state by looking for the k nearest training pictures. An experimental study performed on a set of pictures reveals some good performance of this system, compared to other classification methods such as SVM and VGG-16.