Existing weld seam recognition based on sub-region BP_Adaboost algorithm

This paper presents a sub-region BP_Adaboost algorithm. Compared with the sub-region BP algorithm, it can raise the recognition accuracy of existing weld seam from 90% to 94%. The algorithm firstly obtains various tilt angles of three types of weld seam and builds samples set which consists of weld seam and equal number of non-weld seam sub-regions. 5000 samples are obtained by Matlab. 4000 samples are selected as the training data while 1000 samples are chosen as testing data. For training samples, the final strong classifier is obtained by adjusting the node number of hidden layer and the number of weak classifiers. The strong classifier is applied to test 1000 group of samples. The experiment shows that the classification accuracy is increased by 4%. The algorithm has good result. The network structure is simple due to less input vector dimensions and only four weak classifiers can improve the recognition accuracy of existing weld seam.