Vision-attribute-based X-ray security inspection contraband identification method

Abstract

The invention relates to the field of computer vision, and particularly relates to a vision-attribute-based X-ray security inspection contraband identification method by adopting a deep learning framework. The method comprises the following steps: 1) training samples are collected and labeled, original single-channel 16-bit high and low energy X-ray grayscale image are acquired, and through vision-attribute-based preprocessing, 16-bit three-channel color images are obtained as a data set for model training and testing; 2) the images in the training set are inputted to a network for training: adarknet network is used to extract features from the input image, a feature map is outputted, and a yolo layer is adopted to perform boundary frame prediction on the feature map at multiple scales, and after training, the model supports identification on already-labeled 12 categories of contrabands; and 3) images in a test set are inputted to the model for testing, identification results are outputted, the labeled contrabands are displayed on the input image, and according to IoU and R-P curves, mAP is calculated and obtained. Compared with the prior art, the method has the advantages of highaccuracy, high intelligence, high compatibility and the like.

Description

technical field [0001] The invention relates to the field of computer vision, in particular to a vision-attribute-based X-ray security inspection contraband identification method. Background technique [0002] In public places with a large flow of people, there are dense and complicated people, and criminals often carry knives, guns, bombs and other prohibited items to commit crimes. In order to avoid heavy casualties and property losses, these public places will be equipped with security inspection systems, the most common of which is the X-ray security inspection system. [0003] For the luggage image generated by the X-ray security inspection system, most of the monitoring methods now use manual identification, and rely on the staff to identify and locate the contraband in the luggage with the naked eye. However, this method has obvious disadvantages: 1. The staff will inevitably experience fatigue when performing object recognition for a long time, resulting in a decrea...

AI Technical Summary

Problems solved by technology

However, this method has obvious disadvantages: 1. The staff will inevitably experience fatigue when performing object recognition for a long time, resulting in a decrease in concentration

This will have a certain impact on the monitoring results, reduce the identification rate of contraband, and give criminals an opportunity

2. The speed of staff's visual recognition is relatively slow, resulting in low efficiency of security inspection

3. The staff responsible for the security inspection and monitoring work need to undergo long-term training before taking up the job, which will consume a lot of manpower and material resources

However, the whole method needs to be based on a large number of physical conditions, such as nuclear density meters, spectral analyzers, etc., and the training process is complicated. When the category increases, the entire classifier needs to be retrained and the model modified. This method of individually training classifiers for each category is used in practice. difficult to achieve in the process

Images