Visual building crack recognition method based on attention mechanism and ResNet fusion

Abstract

The invention relates to a visual building crack recognition method based on an attention mechanism and ResNet fusion. The method comprises the following steps: (1) collecting building crack images by adopting an unmanned aerial vehicle to construct a crack data set; (2) carrying out data preprocessing and data enhancement on the crack image by adopting histogram equalization, bilateral filtering, image center random cutting and other methods; (3) building a building crack recognition model based on the combination of an attention mechanism and a deep residual neural network; (4) hierarchically and visually displaying an image recognition result by adopting a gradient weighting class activation heat map algorithm, adjusting a network structure and network parameters according to a visualization result, and building and optimizing a final model; (6) detecting the image in the actual field by using the optimized model. According to the method, the crack can be quickly and accurately recognized, a black box mechanism of the neural network in the recognition process can be effectively broken through, and a visual basis is provided for adjustment of a network structure.

Description

Technical field[0001]The present invention relates to the field of computer image processing, and more particularly to a method of visualizing a crack visualization recognition based on a focal mechanism and a RESNET fusion.Background technique[0002]In the process of engineering construction, the quality and safety inspection of buildings are extremely important. Among them, the detection of the exterior wall cracks is particularly important. It will not only affect the function and beauty of the building, but will make the structural security decrease, but the seismic performance is deteriorated, so the rapid and accurate detection identification of building cracks is the urgent problem in the current structural health test.[0003]Currently, construction crack detection recognition often uses artificial regular examination methods. However, this method has great subjectivity, consumption consumption, high cost, low work efficiency. In addition, a building with a more complex structu...

AI Technical Summary

Problems solved by technology

However, the recognition accuracy of the existing deep neural network model for crack detection is not high, and at the same time, it cannot overcome the black box mechanism of the neural network model. In addition, it is impossible to see the recognition results of each layer in the network layered and visualized, which will also lead to difficulties. Determine the optimal network model, so the current crack deep learning models are mostly designed based on experience

Images