Decision-Making Strategies for Robotic Harvesting of Button Mushrooms Using YOLOv5s and Spatial Analysis

Button mushrooms (Agaricus bisporus) are traditionally harvested manually worldwide to maintain their high quality in the fresh market due to their delicate skin. However, the U.S. mushroom industry faces rising production costs and labor shortages due to the labor-intensive nature of mushroom harvesting. The varying growth rates of mushrooms and their dense clustering complicate decision-making and picking strategies for harvesting mature mushrooms across multiple flushes. Effective robotic harvesting systems must determine picking strategies and execute precise and effective bending motions for mushroom detachment without collisions. This research focuses on developing a decision-based strategy to determine the optimal bending orientation for mushroom harvesting. Using the YOLOv5s model for mushroom detection and a pixel-based threshold for maturity assessment, the algorithm analyzed spatial arrangements to determine optimal bending directions. The results demonstrated that the developed strategy effectively identified mushroom locations with 92% precision, maturity levels with an R2-score of 97%, and spatial arrangements across various conditions with 100% success. During model evaluation, all mushrooms in the test images were successfully assigned the correct picking orientation. This strategy holds significant potential for automating robotic mushroom harvesting in future applications.