Research on Overlapping Infrared Forensic Target Extraction Algorithms Based on Deep Learning
In forensic image processing,the arbitrariness and subjectivity of human activities lead to the problem of fingerprint overlap in infrared images,making the extraction of overlapping targets in such images more complex.This paper addresses the difficulty of extracting overlapping infrared forensic targets and proposes a ResEU-Net network model for locating and segmenting fingerprint target areas as well as overlapping target ar-eas.Furthermore,through coordinate transformation and image fusion algorithms,individual overlapping finger-prints are effectively extracted and restored.The algorithmic experimental results and comparative experiments with the proposed network model validate the accuracy and superiority of our approach in extracting overlapping infrared forensic targets.The evaluation metrics of the ResEU-Net network model,including mIoU,mAP,and accuracy,outperform other models.Compared with other target extraction methods,our approach demonstrates higher accuracy and integrity in extracting overall fingerprint targets,overlapping fingerprint targets,and individ-ual fingerprint targets.
NETL
NSTL
万方数据
