Crack defect detection system based on deep learning

Abstract

The invention relates to a crack defect detection system based on deep learning, and belongs to the field of product crack defect detection. Comprising an image annotation module, an image preprocessing module, a data enhancement module, a deep neural network module and an algorithm prediction module. Through reasonable design of a neural network architecture, multi-task identification of rectangular coordinates of a target area, a target category and a target pixel area is realized, the accuracy of target area identification by a target detection algorithm is effectively improved based on the advantages of a pixel-level segmentation identification algorithm, target pixel-level classification information is given, and the identification accuracy of the target area is improved. The method reduces the recognition error rate of a part of complex shape target areas, achieves the precise positioning of a target efficiently and reliably, improves the requirement for the high accuracy of a target detection algorithm in the practical engineering application, and provides a powerful algorithm support for a high-precision target recognition application scene.

Description

technical field [0001] The invention belongs to the field of product crack defect detection, and relates to a crack defect detection system based on deep learning. Background technique [0002] With the development of computer vision technology, target detection technology is more and more widely used in production and life. With the expansion of application fields, the requirements for target detection accuracy in engineering applications are getting higher and higher, especially in the aerospace field. Target detection methods cannot meet its high-precision requirements. In order to meet the challenges of target detection tasks in practical application scenarios, it is urgent to introduce more advanced algorithms to improve detection accuracy. In the field of industrial inspection, the current main detection methods still rely on a large amount of manpower, with low detection efficiency and high economic costs. With the development of deep learning technology, high-precisi...

AI Technical Summary

Problems solved by technology

[0003] From the literature, most of the object detection methods currently used in engineering are oriented to rectangular area prediction algorithms. Although a few methods use pixel-level segmentation methods, they have not achieved multi-scale training and cannot meet the target multi-scale. identify

Images