Research on a Segmentation and Location Algorithm Based on Mask RCNN for Agaricus Bisporus

This paper proposes a recognition and localization algorithm based on Mask RCNN(Mask Region-based Convolutional Neural Network) for Agaricus bisporus to address the problems caused by the poor environmental adaptability of traditional vision methods, such as high mushroom unrecognized rate especially for small mushrooms and sticky mushrooms, poor robustness of recognization, low accuracy of centroid localization, etc. Firstly, the algorithm uses Mask RCNN to segment Agaricus bisporuses and obtain the masks of each Agaricus bisporuses. Next, the Agaricus bisporus mask edges are extracted, then the ellipse is used to reconstruct the contour depending on the data of the mask edges, and the center point of the ellipse is taken as the center point of the Agaricus bisporus. The experiments show that the proposed algorithm in this paper has stronger robustness, it greatly improves the success rate of recognition of Agaricus bisporus under different light intensities and complex growth environments, and the generated contours much more fit with the actual contours, the size of the recognized Agaricus bisporus and the localization accuracy are higher, with the segmentation accuracy AP50 reaching 95.061%, which can better meet the requirements of Agaricus bisporus picking and its growth status monitoring.