<![CDATA[Attacks by long-tailed macaques (Macaca fascicularis) on corn farmland cause a drastic decline in yield from 7–9 metric tons to only 4–5 metric tons per hectare. Traditional control methods have proven ineffective, necessitating solutions based on smart technology. This study explores the RTMDet (Real-Time Detection Transformer) algorithm combined with the SAHI (Slicing Aided Hyper Inference) framework to evaluate wildlife detection performance. Four RTMDet variants RTMDet-tiny, RTMDet-s, RTMDet-m, and RTMDet-l were tested using a dataset of 2,271 COCO annotated images. Evaluation on the test set showed that the standalone RTMDet-s achieved the best performance with a mAP@0.5 of 0.9070, a Recall of 0.9585, and an F1-Score of 0.9321. Applying SAHI reduced accuracy and caused computational delays (down to 3.02 FPS) because the monkey objects occupy a relatively large area within the image frame. Based on these results, the standard RTMDet-s is highly recommended for integration as a real-time automated trigger system within IoT-based wildlife deterrent devices to instantly mitigate attacks and protect farmers’ crop yields.]]>
Copyrights © 2026
