A dam crack detection method based on u-net network and sc-sam attention mechanism

Abstract

The invention discloses a dam crack detection method based on a U-net network and an SC-SAM attention mechanism. The method first collects a dam data set, and performs data expansion on the data set. Then build the deep learning segmentation network U-net model, and add SC-SAM attention mechanism on this basis, the attention mechanism consists of two parts, CAM improves the crack channel weight in the feature map, and SAM improves the feature The weight of the crack area on the spatial domain in the figure, the mutual cooperation of the two makes the accuracy of the model for dam crack detection greatly improved. Divide the expanded data set into training set and test set, use the deep learning segmentation network model with SC-SAM attention mechanism in the training set for training, and obtain the trained model; according to the trained model, input the test set into the training set Test in a good model to get the dam crack test results.

Description

technical field [0001] The invention relates to a dam crack detection method based on a U-net network and an SC-SAM attention mechanism, and belongs to the technical field of image recognition. Background technique [0002] Dams are an important part of my country's water conservancy construction. The safety and stability of dams play a key role in the development of water conservancy construction. Therefore, how to efficiently and accurately detect the problems of dams is a challenge we need to face. Based on the basic nature of the dam, due to the impact of water pressure and rainwater erosion, seepage, and erosion for a long time, the dam is prone to cracks, and the existence of cracks will lead to the deterioration, damage and collapse of the dam building. Therefore, in the dam Among the various problems that exist, the detection of cracks is the most to be solved. [0003] There are many detection methods for cracks. For example, Zhao Fang et al. proposed a new Canny e...

AI Technical Summary

Problems solved by technology

For example, Shuang Wang et al. used the entire image to reconstruct a 3D point cloud model, then used FCN to iteratively train the labeled data to obtain a crack detection model, and traversed the remaining original size images for crack prediction; Jianghong Tang et al. In order to improve Faster R-CNN Based on the detection accuracy of the model for multiple small targets, a multi-task enhanced dam crack image detection method based on Faster R-CNN (ME-Faster R-CNN) is proposed to adapt to the detection of dam cracks in different lighting environments and lengths ; Xu Hui et al. Aiming at the problem that the traditional crack detection algorithm can not overcome the underwater noise, they propose a method based on image saliency to detect dam cracks; Jiang Xiaoyan et al. face blurred dam crack images Due to the characteristics of unclear, uneven brightness, low contrast, and large random noise, a dam crack detection method based on Lattice Boltzmann Model (LBM) is proposed; because the dam defect image has low signal-to-noise ratio and extreme illumination distribution Inhomogeneity and other features, the recognition rate of the classification recognition algorithm is low, so Mao Yingchi et al. proposed a defect image recognition method based on the combination of image LBP features and Gabor features combined with CNN

Images