Camouflage Object Detection Algorithm Based on Improved YOLOv8n

Camouflage objects pose a challenge in machine vision due to their adaptability to surroundings, reducing visual distinctiveness. Traditional image processing methods struggle with manual feature extraction, hindering robustness and efficiency. This study proposes the DCT-YOLOv8n algorithm, leveraging the YOLOv8n deep learning model to address these challenges. Key improvements include integrating a variable convolutional DCNv3 module to enhance feature extraction, incorporating a triplet attention mechanism to improve detection accuracy for occluded and multi-pose objects, and employing SIOU_Loss for enhanced sample matching. Experimental results demonstrate superior performance with a mAP of 89.1%, surpassing other models by 4% to 26.6% in detection accuracy. Additionally, the algorithm exhibits advantages in model parameter count and single-frame inference time, emphasizing its accuracy, lightweight nature, and training efficiency.