Suspicious article automatic detection method, terminal equipment, computer equipment and medium

Abstract

The invention discloses a suspicious article automatic detection method, terminal equipment, computer equipment and a medium. The detection method comprises: inputting a plurality of to-be-detected images into a target detection network model, and obtaining a candidate box and a confidence coefficient c of a suspicious article; fusing suspicious article candidate boxes in the to-be-detected imageto determine the number of candidate boxes corresponding to the same suspicious article; accumulating candidate box confidence coefficients corresponding to the suspicious articles in the to-be-detected image to calculate weighted confidence coefficients c' corresponding to the suspicious articles; and comparing the weighted confidence c' of each suspicious article with a preset threshold T, and outputting a candidate box corresponding to the suspicious article when c' is greater than or equal to T. According to the technical scheme of the invention, suspicious article detection of the to-be-detected image is carried out by using the target detection network model, and the invention has high detection precision and effectively suppresses false alarms.

Description

technical field [0001] The invention relates to the field of automatic detection of suspicious objects, in particular to a method for automatic detection of suspicious objects, terminal equipment, computer equipment and media. Background technique [0002] In recent years, the deep neural network has played an increasingly important role in the field of image target detection. It has end-to-end characteristics. After proper training, the deep neural network can learn the deep-level features of suspicious objects and give the suspicious object in the input image. Item location and properties. As a representative of the target detection method, FasterRCNN has excellent detection performance for suspicious objects hidden by the human body. The performance of deep neural networks depends on massive training data. The more training data and the wider the coverage, the better the performance of the detection model can be trained. However, in actual scenarios, the collection of ...

AI Technical Summary

Problems solved by technology

Millimeter-wave imaging itself is susceptible to noise interference, causing obvious texture, clutter and other pollution in the image, making it difficult to detect suspicious objects, and it is difficult to meet the high requirements of detection rate and false alarm rate

At the same time, due to the scanning hardware system, there will be differences in the imaging styles (such as light and dark, contrast, etc.) of each security inspection system and a single security inspection system in different environments, resulting in a decline in detection performance.

Images