Infrared spectrum target recognition and positioning technology based on image segmentation algorithm

In the era of rapid development of digitalization and science and technology, infrared images are widely used in security, industry and other fields, but their target recognition and positioning face many difficulties. The existing image segmentation and recognition algorithms have low accuracy and large positioning errors when processing infrared images. Based on this situation, this paper proposes an infrared image target recognition and positioning technology based on multi-scale context perception, thermal radiation feature mining and integrated into the Transformer architecture. The study uses a comparative experimental method to compare the proposed model with several classic models such as U-Net and Mask R-CNN on the FLIR ADAS dataset and the KAIST multispectral pedestrian dataset. The experimental results show that the average intersection-over-union ratio of the proposed model in infrared image segmentation is 76. 6%, the comprehensive error of target positioning is only 1. 06 pixels, the average accuracy of military scene target recognition is as high as 88. 4%, and the comprehensive score in complex environments is 85. 3. Compared with other models, the performance is significantly improved. The research results enrich the processing theory of special images in the field of computer vision, provide effective technical support for the upgrade of infrared monitoring equipment in the fields of security, industrial monitoring, etc., and help improve the safety and reliability of related fields.