Development of a mushroom harvesting assistance system using computer vision

Conventional mushroom harvesting relies on manual labors, which is one of the major reasons for increased production costs. Different maturation speed among individual mushrooms promotes farm workers' efforts in selective harvesting. To assist in robotic harvesting of mushrooms, a computer vision system was developed to detect an individual mushroom from mushroom clusters and evaluate maturity of the mushroom. Specific objectives of this study were to (1) solve the overlapping problem and differentiate each mushroom from a mushroom cluster, and (2) develop a machine vision algorithm to identify maturity using the size and shape of mushrooms. For mushroom identification, faster R-CNN model was developed to distinguish mushrooms from substrate. A 3D pointcloud of mushroom was acquired by a depth camera and used to segment an individual crop among the overlapped mushrooms in clusters. After the segmentation, the size of mushroom caps was calculated using the pointcloud and the shape of mushroom caps was quantified using normal vectors. The accuracy of maturity recognition reached 70.93 %. The results of this study can be extended to a commercial scale and enhance mushroom harvesting efficiency by reducing the overall cost of mushroom production.