Intelligent Ultrasonic Image Classification of Artillery Cradle Weld Defects Based on MECF-QPSO-KELM Method

The precise qualitative evaluation on the ultrasonic testing of the artillery cradle weld defects can effectively eliminate its security risk induced by the potential hazardous defects. Nevertheless, it is difficult to determine the detailed reflection properties of ultrasonic waves because of the effect of defect size, shape, orientation and its surface roughness. This study focuses on the intelligent analysis on the ultrasonic testing image of the artillery cradle weld defect. An intelligent classification method was proposed based on the small sample conditions. Thus, in this article, a significance classification feature evaluation algorithm was first proposed based on the multiple evaluation criteria fusion (MECF). No matter which pattern recognition algorithm was used, using the classification feature set (M7, unevenness of gray scale distribution, differential moment, standard deviation of entropy, directionality, long run advantage, contrast, standard deviation of energy, mixed entropy) to intelligently recognize the defect types has a high precision. Especially while using the kernel extreme learning machine (KELM), the classification accuracy reaches 96.4%. Thus, a multi-class classification model of the weld defect recognition termed quantum-behaved particle swarm optimization-based kernel extreme learning machine (QPSO-KELM) was further proposed. The corresponding classification accuracy was raised to 98%. Finally, comparative experiments were done with convolutional neural network (CNN) ResNet-34. The results show that compared with CNN ResNet-34, the proposed method exhibits obvious advantages, and more accurate classification results also indicate the proposed intelligent classification method can be used for the intelligent identification of artillery cradle weld defect during ultrasonic testing.