Baboon Detection for Smart Agriculture Using Deep Learning Algorithms

Agriculture, contributing 30-40% to Africa’s GDP and supporting over 60% of its population, faces severe threats from wildlife foraging, with baboons causing significant crop losses. In Uganda alone, 65% of small-scale farmers report substantial yield reductions due to baboon raids, with losses estimated at ${\$}$ 100-${\$}$ 500 per smallholder farmer annually. Traditional deterrent methods, such as fences and sentries, are laborintensive, costly, and often ineffective due to baboons’ adaptability. This study presents an approach to mitigate baboon-related crop damage using state-of-the-art deep learning algorithms and the Internet of Things (IoT). We trained and compared multiple iterations of You Only Look Once (YOLO) models: YOLOv8-S, YOLOv9-S and YOLOv10-S for real-time baboon detection, utilizing a custom dataset of images collected from various locations in Rwanda. The best-performing model, YOLOv10-S, achieved the highest accuracy in baboon detection. Our system integrates this deep learning model with edge computing capabilities on a Raspberry Pi 4 Model B, enabling inference with the model even in remote farm fields. Upon detecting baboons, the system activates acoustic deterrents and instantly alerts field guards via the short messaging service (SMS) through a GSM module. Our results show that this approach can offer an efficient and scalable solution to baboon foraging in agricultural settings in Africa while minimising human-wildlife conflict. This research not only aims to enhance food security and economic stability for African farmers but also to provide a replicable framework to address wildlife-related agricultural challenges globally.