Method for autonomously learning and identifying novel contraband

Abstract

The invention discloses a method for autonomously learning and identifying a novel contraband. The method comprises the following steps of delimiting a picture p of the novel contraband, and marking adanger level; forming combined information by the corresponding picture p, the classification n and the large class S, and uploading the combined information to an autonomous learning preliminary data set a1; extracting preliminary features t1 of the novel contraband according to the picture p; performing automatic networking retrieval on related pictures including the preliminary features t1, and extracting and concluding deep features t2 of the novel contraband according to the picture set; forming an autonomous learning deep data set a2; carrying out sharing sending and learning updating of an inter-device autonomous learning deep data set a2 according to different frequencies; and generating a new contraband detection model Z. According to the method for autonomously learning and identifying the novel contraband, the novel contraband can be marked in time, the automatic networking retrieval is carried out to extract deep features, the data sharing, sharing sending and learning updating frequency between equipment is reasonable, and the method is very worthy of popularization.

Description

technical field [0001] The invention relates to the technical field of contraband identification, in particular to a method for autonomous learning and identification of new contraband. Background technique [0002] In the prior art, the application number is "201810220068.9", an intelligent contraband identification method based on X-ray security inspection machine images, including establishing a data set for training; training a unified contraband identification model, including shape classification models and color Classification model; use the shape classification model to detect the picture to be detected, and output the shape detection result; after calculation according to the detection result, input the color classification model, and output the color detection result; use the shape detection result and color detection result to calculate the probability and location of dangerous objects ; Comparing the model of the detectable dangerous object with the image to be d...

AI Technical Summary

Problems solved by technology

[0003] However, there are still obvious defects in its use: the above-mentioned invention does not have the function of self-learning, and it cannot list new contraband as identification items in a timely manner, and it is time-consuming and labor-intensive to rely entirely on artificial regular training, and the lag time is relatively long. Long time, not suitable for large-scale security inspection places

Images