The application belongs to the technical field of
image processing, and discloses an RGB-T saliency target detection method based on cross-
modal interaction and full-scale aggregation, realizes fusion between
RGB image features and
thermal infrared image features, fully utilizes feature information of different depths, and can also completely detect targets of different scales in an outdoor scene disturbed by interference, and improves detection precision. The application uses a cooperative incentive learning module as a cross-
modal interaction mode to make two different
modes use effective information features of each other in a targeted manner, maximally suppresses interference information of both sides, that is, information on an image background will not be acquired as effective information, and a detection result with high accuracy can be obtained under the condition of scene interference.