Road defect detection method based on deep learning and self-attention mechanism

Abstract

The invention discloses a road defect detection method based on deep learning and a self-attention mechanism, and the method comprises the steps: 1, selecting an open road defect data set containing longitudinal cracks, transverse cracks, net cracks and pits, and dividing a training set, a verification set and a test set; 2, building a road defect detection neural network model based on a single-stage target detection method, wherein the final output of the network is the defect category and position predicted on the image; 3, training the road defect detection neural network model by using the training set to obtain a road defect detection model; and 4, for an input road image, adopting the road defect detection model to obtain a plurality of detection results, filtering the repeated detection results by using a non-maximum suppression algorithm, and finally obtaining the category and position of the road defect in the image. Based on a self-attention mechanism, the road defect detection precision is improved through fusion of feature maps in different stages, and the problem that the average detection accuracy is affected by large road defect size and shape changes is solved.

Description

technical field [0001] The invention belongs to the application field of computer vision, in particular to a road defect detection method based on deep learning and self-attention mechanism. Background technique [0002] At present, the daily inspection of urban roads mainly relies on inspectors to patrol the roads. For the road defects found, they are reported by manual operations such as filling out newspaper quality forms, which are inefficient. In addition, this method is highly subjective, and it is difficult for different inspectors to maintain a consistent perception of the degree of the same defect. In addition, the length of roads inspected by inspectors every day is generally between 30-40km, and the mileage is short. Therefore, the use of computer technology to intelligently analyze road surface defects has important practical significance. Existing technical means generally use road survey vehicles equipped with many sensors to conduct road surface condition su...

AI Technical Summary

Problems solved by technology

However, the current target detection algorithms usually have a better detection effect on targets with regular outlines and fixed sizes. When facing road defect targets such as cracks and potholes, the detection accuracy is often low due to factors such as various sizes and irregular shapes. Therefore, it is particularly important to construct a suitable detection network model for such targets.

Images