A weak supervision X-ray image contraband inspection method based on layered propagation and activation

Abstract

The invention discloses a weakly supervised X-ray image contraband inspection method based on layered propagation and activation, and the method comprises the steps: firstly obtaining X-ray image dataand a corresponding image category, and forming a training sample set and a test sample set; Obtaining a feature map from top to bottom on the training sample set through a convolutional neural network (CNN) hierarchical structure; According to the layered propagation activation method, promoting the activation of each layer of characteristic pattern through the propagation of interlayer and intra-layer confidence coefficients from top to bottom, and finally obtaining the accurate position of the prohibited article. According to the method, weak supervised annotation information is used for learning an image recognition model; when the image is labeled, only simple specified images are needed to have prohibited goods or not and the category information of the prohibited goods; the specific position of the contraband in the image does not need to be accurately labeled, so that the cost of manual labeling is greatly reduced, and the method is of great significance in realizing intelligent detection of the contraband target and reducing unnecessary repeated work in the security inspection process.

Description

technical field [0001] The invention belongs to the field of security inspection, relates to image processing and classification, and in particular to a method for inspecting contraband in weakly supervised X-ray images based on layered propagation and activation. Background technique [0002] Automatic X-ray image security inspection technology is an important means of ensuring public safety, and plays an important role in ensuring the safety of people, airports, ships, cargo circulation and key infrastructure. At present, the main inspection method for prohibited items in the field of public security is to use manual inspection of X-ray images. With more and more passengers traveling around the world and wider concerns about transportation safety, the work intensity of security personnel is gradually increasing, and the long-term X-ray image inspection work has brought huge burden and pressure to them. [0003] How to alleviate the contradiction between the low efficiency...

AI Technical Summary

Problems solved by technology

[0007] In order to overcome the above-mentioned problems, the present inventors have carried out dedicated research, aiming at the problems existing in X-ray images, proposed a weakly supervised X-ray image contraband inspection method based on layered propagation and activation, taking advantage of the particularity of X-ray imaging, namely The color space of X-rays is more uniform than that of optical data sets, and objects of the same material will be projected into similar colors. The mechanism of layered propagation in the scheme proposed by the present invention can make good use of this feature to find the same The similarity between targets in the category, using the layered activation method to activate the salient area (area with high response value) in the feature image, can effectively alleviate the problem of different target sizes and serious occlusion; use the defined weighted loss function method, there are few positive samples, a larger loss weight is assigned, and there are many negative samples, a smaller loss weight is assigned to alleviate the problem of unbalanced positive and negative images

Images