Description of Invention
This research proposes a computer vision-based methodology where the YOLOv8 segmentation model has been used in identifying and isolating oyster mushroom caps from their cultivation bed image data. A manual-annotated polygon-based annotation dataset has been used for training the YOLOv8 segmentation model. This model also provides infestation detection which enable the system to have targeted treatment at early stage to minimize overall damage and maintain the productivity of the cultivation environment. The model was tested with standard detection metrics and resulted in a mean Average Precision (mAP@0.5) of greater than 75% on the primary class. The model's ability to mask single mushroom caps size area was established using test images of different classes, from the beginning stage of small cap size to ready harvest till overgrown large mushroom cap and that resulted in detection. This work explains the setup of the model pipeline consisting of data preparation, model training, and inference. Early post-processing was explored, and the trained model is set out to build a reliable detection framework to be used as a reference point for future maturity analysis, tracking, and harvesting reasoning. The results demonstrate the potential of using deep learning segmentation in mushroom cultivation to enable scalable low-labour monitoring.