Real-time mushroom detection and maturity classification using YOLO-Tiny on Raspberry Pi platform

Mushroom growing is peerless in providing healthy and fresh mushrooms, aside from its tremendous economic contribution and livelihood among farmers. This paper discusses the efficacy of a state-of-the-art real-time object detector, YOLO, in particular YOLOv3-tiny and YOLOv4-tiny, in detecting oyster mushrooms in a greenhouse environment and at classifying their stages of maturity using low-power embedded devices. These depict that the models detected both versions of mushrooms and their maturity level. Among these, YOLOv4-tiny outperformed its variant, YOLOv3-tiny, in terms of mAP, accuracy, precision, recall, and F1-score. The results for accuracy showed the achievement of YOLOv4-tiny with 83.9% while YOLOv3-tiny attained 80.3%. This has pointed toward the extent such tuned models could go with smart farming systems for real-time monitoring, automated harvesting, and improving operational efficiency.