Weld Defects Images Classification and Recognition Based on Sparse Deep Belief Network

Because weld defects images obtained by X-ray are difficult to be identified in industrial production, a Deep Belief Network (DBN) model composed of three-layer Restricted Boltzmann Machine(RBM) is proposed.Due to the sparse constraint introduced into the depth learning model algorithm, the weld defects information can obtain the target sparse representation in more effective way and the weld defects information can be identified through more concise and effective training. The comparison experiment of layered network model shows that the Sparse Deep Belief Network can ensure the accuracy of weld defects images recognition and classification while improving the performances of Deep Belief Network by avoiding the overfitting.