RGB-D image-based CLANet steel rail surface defect detection system and method

Abstract

The invention relates to a CLANet steel rail surface defect detection system and method based on RGB-D images. The system comprises a feature extraction module, a cross-modal information fusion module and a defect positioning and segmentation module. The method comprises the steps that the feature extraction module extracts multi-level features of an input RGB image and a depth image to obtain RGB-D feature information; a space refinement branch module SRB of the cross-modal information fusion module disperses RGB-D feature information to four dimensions, and fnRGB and fnDepth are obtained; the cross-modal information fusion module carries out cross-modal fusion to obtain fi; a fusion result is transmitted to a double-flow decoder DSD to obtain a saliency defect prediction map; and calculating a loss value of the collaborative learning attention network CLANet. According to the method, feature fusion and cross-modal fusion of two modals are better realized, so that the detection effect of the image is further improved, and the accuracy of the surface of the steel rail is improved.

Description

technical field [0001] The invention relates to the technical field of rail surface defect detection and image detection technology, in particular to a CLANet rail surface defect detection system and method based on RGB-D images. [0002] technical background [0003] In recent years, due to the rapid development of visual attention mechanism and deep learning, salient object detection has gradually become a very popular research direction in the field of computer vision. Saliency detection is an extremely important task in computer vision. It relies on deep learning algorithms to simulate human visual characteristics and achieve specific target detection. The human visual attention mechanism relies on human prior knowledge to selectively acquire salient targets, that is, targets of interest. Saliency detection has important application value in object recognition, image and video compression, image retrieval, image redirection and other directions. The prototype of the mod...

AI Technical Summary

Problems solved by technology

In 2017, the first model that applied convolutional neural network to RGB-D image saliency detection was proposed, but this model only used the shallow information of the network for saliency prediction, and did not make full use of the deep information.

[0008] Artificial visual method, eddy current detection method, magnetic particle method and ultrasonic detection method are common methods for rail defect detection at present. These methods are lower than machine vision methods in terms of detection accuracy and efficiency, and image processing technology has not been widely used.

Images