Deep Learning Models Evaluated on Real-World African Crop Dataset

A recent study published on arXiv (2608.11053) evaluates six object detection models—YOLOv5, YOLOv8, YOLO11, YOLO26, Faster R-CNN, and RT-DETR—utilizing a real-world dataset named AgriAISeg, which was manually gathered from farms across Nigeria. This dataset consists of 3,382 images featuring sesame, cabbage, and tomato crops, taken under diverse environmental conditions, such as varying light, occlusion, and different angles. The models were trained and evaluated based on precision, recall, mAP@0.5, and mAP@0.5:0.95. Findings indicate that RT-DETR delivered the best overall results, achieving a precision of 0.768 and a mAP@0.5:0.95 of 0.624. The research underscores the promise of computer vision in agriculture, especially in less represented areas like Africa, while addressing the lack of controlled datasets that mirror actual farming scenarios.