Dangerous article detection method and device for X-ray security check scene

Abstract

The invention discloses a hazardous article detection method and device for an X-ray security check scene. The method comprises the following steps: acquiring an X-ray security check image to be detected; training to obtain a neural network model of dangerous goods category information features irrelevant to the extraction domain; inputting a to-be-detected X-ray security check image into a pre-trained neural network model of dangerous goods category information features irrelevant to an extraction domain to obtain a feature map of the image, which is used for dangerous goods prediction and contains domain-irrelevant dangerous goods characterization information; and image features are extracted as input of a security inspection machine and are used for subsequent dangerous goods prediction. According to the invention, endogenous offset caused by different types of security inspection machines can be well processed.

Description

technical field [0001] The invention relates to a dangerous goods detection method for an X-ray security inspection scene, and also relates to a corresponding dangerous goods detection device, belonging to the technical field of safety inspection. Background technique [0002] With the increase of the number of people in public transport, security checks are becoming more and more important. In the security inspection scenario, X-rays are usually used to check whether there are dangerous goods in the luggage, and then manual inspection is used to check whether there are dangerous goods in the security inspection images generated by the security inspection machine. However, after watching complex security images with high attention for a long time, security inspectors may experience visual fatigue, resulting in a decrease in work efficiency. Therefore, manual security checks will have problems such as high labor costs, many missed detections and false detections, and low det...

AI Technical Summary

Problems solved by technology

However, after watching complex security images with high concentration for a long time, security inspectors may experience visual fatigue, resulting in decreased work efficiency

Therefore, manual security checks will have problems such as high labor costs, many missed detections and false detections, and low detection efficiency. It is necessary to use security inspection machines instead of manual security inspections.

[0003] Due to the differences in the hardware of different types of security inspection machines, there are inherent offsets that are difficult to detect with the naked eye in the X-ray security inspection images generated by different security inspection machines

The existence of this endogenous offset makes the dangerous goods detection neural network model trained on the X-ray images generated by one type of security inspection machine have a serious performance degradation when it is detected on another type of security inspection machine.

The existing cross-domain dangerous goods detection neural network model cannot handle the endogenous offset caused by different security inspection machine models well, and it is difficult to learn the domain-independent category features in dangerous goods images. The task of detecting cross-domain objects (that is, dangerous goods or prohibited items) poses challenges

Images