Camouflaged object detection using hybrid-deep learning model

Camouflaged Object Detection, or COD, is crucial in many real-world applications, such as surveillance, military reconnaissance, and wildlife monitoring. The two main areas where conventional self-attention methods can be enhanced are in capturing hierarchical contextual information and addressing the challenge of COD. It requires capturing both coarse and fine details due to the significant variation in the sizes and appearances of objects. This research work presents Hybrid-COD, a novel method for camouflaged object detection using Swin Transformer architecture with Enhanced Receptive Field (ERF) modules and Cross Scale Feature Fusion (CSFF) processes. Swin Transformer’s shifted windows reduce the computational burden by limiting self-attention to non-overlapping windows and then shifting them. The addition of ERF modules improves the network’s capacity to capture contextual information, enabling more precise distinctions between disguised objects and their backgrounds. The CSFF streamlines the integration of features from multiple scales, empowering the model to identify and categorise objects hidden at various scales. According to the accuracy metric, proposed model achieved a score of 90% on CAMO dataset, 98% CHAMELEON dataset, 91% COD10K dataset, and 97% NC4K dataset. It has been observed from experimental results that Hybrid-COD effectively captures hierarchical contextual information and fine details, improving detection of objects with varying sizes and appearances.