Deep learning for image-level camouflaged object detection: A review of progress, challenges and prospects

Camouflaged Object Detection (COD) represents an emerging and highly challenging visual detection task dedicated to identifying and segmenting objects that blend seamlessly into surrounding complex environments. With the continuous advancement of deep learning techniques, image-level COD has garnered widespread research attention, with practical application value demonstrated in military reconnaissance, target tracking, industrial defect detection, and wildlife conservation. However, the rapid progress of this field has outpaced existing review works, which fail to cover and systematize the latest theoretical and technical achievements comprehensively. To address this gap, this paper presents a systematic and detailed review of deep learning-based image-level COD technologies. First, the review categorizes existing deep learning-based COD models, which are thoroughly summarized across multiple dimensions such as network architecture and learning paradigm. Second, the study systematically organizes and analyzes current mainstream benchmark datasets and evaluation metrics. Subsequently, the research selects representative models for benchmarking. An in-depth analysis of detection performance from both quantitative and qualitative perspectives is conducted. Finally, the article explores the bottlenecks and challenges of current COD technologies in actual complex scenarios and proposes potential solutions, forecasts future development trends, and research directions. This review aims to establish a comprehensive and accurate reference framework for researchers in COD, thereby advancing the theoretical development and practical application of COD technologies. Relevant resources: https://github.com/BaiyuanLiang/Awesome-COD-Survey .