Evidence-driven reference guided camouflage target detection method and system

Abstract

This invention discloses an evidence-driven, reference-guided method and system for detecting camouflaged targets, relating to the field of computer vision technology. The method includes: acquiring a camouflaged target image and extracting multi-scale basic features; acquiring a reference image and extracting global common reference features; using a reference-guided deformable encoder, based on the global common reference features, performing directional modulation and spatial adaptive enhancement on the multi-scale basic features to obtain enhanced multi-scale sequence features; using an uncertainty-aware evidence decoder, based on the enhanced multi-scale sequence features, modeling and segmenting data uncertainty through evidence reasoning, and using uncertainty as a guide to achieve progressive fusion and directional enhancement of the enhanced multi-scale sequence features to obtain prediction maps at each scale; and using a boundary-aware refinement module to correct the prediction boundaries of the multi-scale prediction maps to obtain the final prediction map. This invention improves the recognition rate of camouflaged targets.

Images