Camouflaged Object Detection with a Feature Lateral Connection Network

We propose a new framework for camouflaged object detection (COD) named FLCNet, which comprises three modules: an underlying feature mining module (UFM), texture-enhanced (TEM), and neighborhood fusion (NFFM). Existing models overlook the analysis of features, results in extracted low-level texture information that is not prominent enough contains more interference due to slight difference between foreground background object. To address this issue, we created UFM using convolution with various expansion rates, max-pooling, avg-pooling deeply mine textural features eliminate interference. Motivated by traits passed down through biological evolution, NFFM, primarily consists element multiplication concatenation followed addition operation. obtain precise prediction maps, our model employs top-down strategy gradually combine high-level information. Using four benchmark COD datasets, proposed outperforms 21 deep-learning-based terms seven frequently used indices, demonstrating effectiveness methodology.