Optimizing Convolutional Layer in YOLOv8: Enhancing Accuracy and Efficiency for Camouflaged Object Detection in Complex Environments
Joshu Leonardy, Mohd Rashidi Salim, Randy Erfa Putra, Agus Virgono, Umar Ali Ahmad, Mera Kartika Delimayanti, Wildan Panji Tresna · 2025
Camouflaged object detection (COD) in complex environments remains an important challenge for computer vision systems, especially in military and surveillance applications. This research addresses the unexplored impact of hyperparameter tuning on YOLOv8 performance for COD in forest environments. We propose an optimized YOLOv8n framework, systematically evaluating the effects of epoch configuration (10-75) and batch size (2,4,6) on detection accuracy and efficiency. A data set of 15 original images of camouflaged military objects is expanded to 90 samples through rotation, inversion, and contrast adjustment. Training on an NVIDIA GTX 1080 GPU revealed that a smaller batch size (4) and higher epochs (50-75) significantly improved the mAP@50 and mAP@50-95 average precision and precision in multiple object detection. These findings highlight the importance of optimizing the parameters before training the data for the best results and offer practical guidance for real-world applications.