Features Split and Aggregation Network for Camouflaged Object Detection

• DOI: 10.3390/jimaging10010024
• Publication Date: 2024-01-18T08:50:56Z
• Authors: Zejin Zhang; Tao Wang; Jian Wang; Yao Sun
• Keywords: Feature (linguistics); Backbone network; Camouflage

Higher standards have been proposed for detection systems since camouflaged objects are not distinct enough, making it possible to ignore the difference between their background and foreground. In this paper, we present a new framework Camouflaged Object Detection (COD) named FSANet, which consists mainly of three operations: spatial detail mining (SDM), cross-scale feature combination (CFC), hierarchical aggregation decoder (HFAD). The simulates three-stage process human visual mechanism when observing scene. Specifically, extracted five layers using backbone divided them into two parts with second layer as boundary. SDM module cursory inspection gather details (such edge, texture, etc.) fuses features create impression. CFC is used observe high-level from various viewing angles extracts same by thoroughly filtering levels. We also design side-join multiplication in avoid distortion use element-wise filter out noise. Finally, construct an HFAD deeply mine effective these stages, direct fusion low-level semantic knowledge, improve camouflage map cascade technology. Compared nineteen deep-learning-based methods terms seven widely metrics, our has clear advantages on four public COD datasets, demonstrating effectiveness superiority model.