Design of intelligent grading and sorting system for <i>Lyophyllum decastes</i> based on YDNet

The size of Lyophyllum decastes varies during growth, and its maturity detection and quality grading still primarily rely on manual operations. Issues such as inconsistent classification standards and high sorting damage rates severely impact production efficiency and quality stability. To address these challenges, this paper proposes a dual-stage deep learning model, YDNet, based on YOLO and Deeplabv3+, to enable maturity detection, quality grading, and low-damage sorting of Lyophyllum decastes . YDNet consists of two stages: in the first stage, a target detection algorithm is used to perform maturity grading, with depthwise separable convolutions and DIoU-NMS improving detection efficiency and resolving the occlusion problem between mushrooms. In the second stage, semantic segmentation is employed to segment mushrooms, and quality grades are defined based on the mean area, with lightweight MobileNetV2 and focus loss optimizing recognition speed and addressing sample imbalance issues. The system is equipped with a binocular depth camera and a flexible end-effector to achieve precise positioning and low-damage grasping. Experimental results show that the system achieves a sorting speed twice that of manual operations, reduces the error classification rate by 8%, and achieves an average damage rate of only 2.75%, providing an efficient and reliable solution for the industrial production of Lyophyllum decastes .