Multi-feature Fusion With Attention and Receptive Field for Camouflaged People Detection

Abstract

Camouflaged object is usually visually similar to the surrounding environment. Although camouflaged object detection is more challenging than general object detection, camouflaged object detection has been used in medicine, biology and military, so the research on camouflaged object detection is also of great significance.Since the camouflage object can be regarded as the "salient object" in the camouflage image, this paper combines the characteristics of salient object detection and camouflage object detection, mainly from the three aspects of object edge information enhancement, channel attention mechanism and expansion of the receptive field. Inspired by the idea of U-Net, we design a new camouflage object detection network. Finally, we use the camouflage people data set, and carry out training and testing under the same conditions with other networks, and verify the effectiveness of the method by comparing the results.