Application of Neural Network Technology in Defect Image Recognition

Abstract

Pressure vessels need non-destructive testing regularly, which requires more accurate and faster detection methods. The wall-climbing robot with visual sensor can carry out magnetic particle testing at the same time, and take real-time magnetic particle testing results. It has broad application prospective and application values. This paper presents a study on the crack defects of weld seams in pressure vessels. A neural network algorithm is introduced to recognize and classify a large number of pictures taken by a wall-climbing robot. An experiment was conducted. The results showed that the CNN method is efficient and economic way to complete the detection tasks. It is feasible to be integrated to the wall-climbing robot for the further automatic detection of defect images.