Efficiency-optimized monkey detection for wildlife monitoring: A comprehensive YOLOv8s evaluation

Human-wildlife conflicts in rapidly urbanizing regions necessitate the use of automated monitoring systems for effective mitigation strategies. Monkey populations cause significant agricultural damage and urban safety concerns, yet manual monitoring remains impractical for continuous surveillance. This study implements the YOLOv8s architecture for automated monkey detection, balancing detection accuracy with computational efficiency essential for field deployment. The model was trained on 2,244 annotated images spanning diverse environmental conditions—urban settings, forest canopies, and varied illumination from dawn to dusk. Training utilized 150 epochs with augmentation including rotation (±15°), scaling (0.8-1.2×), mosaic (probability=1.0), and mixup (α=0.15). YOLOv8s improved mean Average Precision at IoU 0.5 (mAP@0.5) from 0.48 to 0.52 (+8.3%), achieved a precision of 0.89 with a recall of 0.78, and reduced inference time from 6.3 ms to 5.5 ms (−12.7%). The precision-recall curve achieved an Area Under the Curve (AUC) of 0.867, confirming robust detection performance. These improvements enable deployment on edge devices with limited computational resources, facilitating real-time wildlife monitoring in resource-constrained environments while maintaining detection reliability for practical conservation applications.