Crack recognition method and device based on neural network, equipment and storage medium

Abstract

The invention discloses a crack recognition method and device based on a neural network, equipment and a storage medium, and the method comprises the steps of constructing a structural feature extraction module and a detail feature extraction module, extracting structural features and detail features through the structural feature extraction module and the detail feature extraction module and fusing same, carrying out the up-sampling of the fusion features, obtaining a crack probability graph, and processing the crack probability graph to obtain a crack identification result. The invention ishigh in automatic processing level, directly outputs the quantitative features and masks of the crack diseases from the to-be-recognized image in an end-to-end manner, can greatly reduce the workloadof an operator, and improves the crack recognition efficiency and accuracy.

Description

Technical field [0001] The present invention belongs to the field of computer science, and specific to crack identification methods, devices, devices, and storage media based on neural networks. Background technique [0002] Cracks are one of the common diseases of objects, directly related to its quality, safety, wood plate, road bridge, etc., and quantify the accurate identification and quantification of cracks and other diseases. One of the important links in quality and status assessment. At present, the diagnosis of cracks and other diseases are still artificial, time-consuming, strong subjectivity, high cost, and the artificial testing steps are more cumbersome and require rich expertise. In order to improve the efficiency of crack detection, it is necessary to study rapid and accurate automated crack detection methods. [0003] With the continuous development and application of computer vision technology and deep learning technology, the neural network represented by convo...

AI Technical Summary

Problems solved by technology

In order to pursue accuracy, the segmentation network often uses complex network structures such as high-resolution input images, multi-scale feature fusion, more feature channels, and deeper networks, but these methods will bring a large amount of model parameters and calculations. The amount reduces the model reasoning speed, which is not suitable for practical applications; on the other hand, in order to improve the speed, the segmentation network can reduce the resolution of the feature map, reduce the depth of the network or the number of feature channels, etc. These methods will lose the local details of the fracture area. will cause a decrease in accuracy

In addition, the cracks in the image show a large global size change and irregular shape, but the local area is small. The common segmentation network cannot balance the speed and accuracy well in this task.

Images