Security CT Image Enhancement And Target Recognition System

Safety inspection has always received widespread attention from the government and the public,and the awareness of security inspection has been integrated into the development of various industries,attracting more and more attention from researchers.With the maturity of CT security inspection technology,image-based security inspection systems are widely used in customs inspections,station security inspections,cargo inspections,and public security inspections.Early security inspection systems were mainly operated manually,which had many disadvantages,such as high risk,low detection efficiency,poor accuracy,and high rates of missed and false alarms.With the development of artificial intelligence,hardware and algorithm research and development and improvement,the application of AI image processing algorithms to CT security inspection image detection has improved the intelligence recognition of CT security inspection systems to a certain extent.However,for certain specific places,objects,and environments,there are still situations such as missed detection,false alarms,and misjudgment due to the influence of CT security inspection imaging principles,which sometimes also bring about waste of resources for subsequent situation processing.This paper develops an efficient,precise,and intelligent security CT image enhancement and target recognition system based on relevant artificial intelligence algorithms and techniques.The content of this paper mainly includes the following aspects:(1)Apply the ADN algorithm for medical metal artifact reduction to the task of CT security inspection metal artifact elimination,to enhance the legibility of CT images and facilitate the extraction of image features in later stages.(2)Propose a single-image super-resolution algorithm,CoT-SR,to enhance the image details of blurry parts,and facilitate the subsequent image recognition work.(3)Improve the lightweight FCOS-M network by combining the characteristics of FCOS and MobileNet,to perform prohibited and taxable item recognition on the images after the image enhancement process.(4)Divide the CT imaging into six segments to obtain image data,while preserving different angle information of imaging parameters such as isotopes and voxels,to establish an CT security inspection image dataset.In the object detection results,the average accuracy of detecting tangible(solid)objects is 98.8%,while the average accuracy of detecting intangible(liquid/gas)objects is 86%.The experimental results indicate that the CT image enhancement and target recognition system can effectively enhance the information of CT images and complete image recognition,with advantages in accuracy,security,and efficiency.It can significantly reduce labor costs.