Dangerous goods detection method and device based on deformable convolution, and equipment

Abstract

The invention discloses a hazardous article detection method and device based on deformable convolution, equipment and a readable storage medium. The method comprises the steps of obtaining a to-be-detected hazardous article image; performing feature extraction on the dangerous goods image by using a deformable convolutional network to obtain an original feature map; inputting the original featuremap into an RPN layer, and determining a region of interest and whether the region of interest is a classification result of the dangerous goods or not; inputting the region of interest of which theclassification result is dangerous goods into an ROI pooling layer to obtain a target feature map of a preset size; and determining the category and the detection box of the dangerous goods accordingto the target feature map by using the full connection layer. Visibly, according to the scheme, the deformable convolution is used for carrying out feature extraction on the dangerous goods image; because the deformable convolution can increase the rotation invariance of the convolution kernel through learnable affine transformation, the method can achieve the target recognition under the condition that the shape or position of the dangerous goods is uncertain, and remarkably improves the reliability and accuracy of a dangerous goods recognition result.

Description

technical field [0001] The present application relates to the technical field of target recognition, and in particular to a method, device, equipment and readable storage medium for dangerous goods detection based on deformable convolution. Background technique [0002] With the development of the transportation industry, the population flow is more and more frequent. In order to ensure public safety, it is necessary to detect dangerous goods such as weapons, explosives, and drugs in public safety areas. The traditional object detection algorithm always learns the shape of the fixed dangerous goods during the learning process, so the algorithm is not robust enough. For example, the target detection algorithm can recognize a horizontally placed knife after learning, but cannot recognize a vertically placed knife. [0003] It can be seen that how to reduce the impact of target position changes on target recognition and improve the robustness of target recognition is an urgent...

AI Technical Summary

Problems solved by technology

[0004] The purpose of this application is to provide a dangerous goods detection method, device, equipment and readable storage medium based on deformable convolution, which is used to solve the problem that the traditional target recognition algorithm has poor robustness, and the recognition result is relatively affected by the change of the target position. big problem

Images