Detection of Weld Defects Based on Incremental Two-Dimensional Principal Component Analysis

Abstract

Aiming at the requirements of the real-time detection of weld defects in the industry, the candid covariance-free incremental PCA (CCIPCA) algorithm was introduced into the Two-Dimensional Principal Component Analysis (2PCA) algorithm, therefore an incremental 2DPCA algorithm (I2DPCA) is proposed. Firstly, the images captured by camera are preprocessed to improve the quality of the images because the captured images might be affected in the process of image acquisition. Then the 12DPCA algorithm is used to achieve feature extraction, and the dimension of the images is reduced. Finally, the recognition of weld defects is implemented using the k-nearest neighbor algorithm (KNN). Compared with the methods of traditional recognition by extracting the sizes of several geometric defects, the proposed algorithm has the advantage of the small memory spaces, the high recognition rate and the strong real-time recognition. The experimental results show that the proposed algorithm has higher recognition rate and is more practical than the 2DPCA, and the recognition rate reaches 94%, which can meet the requirements of the real-time detection.