- Article
18 Pages
To enable the rapid and accurate acquisition of phenotypic parameters during distinctness, uniformity, and stability (DUS) testing of edible mushroom varieties, and to address the low efficiency, strong subjectivity, and insufficient standardization of conventional manual measurement methods, this study used the fruiting bodies of Oudemansiella raphanipes as the research object and proposed an automatic phenotypic parameter extraction method based on deep learning instance segmentation. An instance-segmentation dataset containing 473 O. raphanipes images and 1419 annotated objects was constructed, and the YOLOv11s-seg model was improved by introducing the lightweight feature-extraction network StarNet_s050 to construct the YOLOv11s-star-seg model, thereby improving model inference efficiency and meeting practical application requirements. Under standardized imaging conditions, pileus and stipe regions were obtained from the instance-segmentation results, and image-scale calibration was combined to automatically extract DUS phenotypic parameters, including pileus diameter, pileus context thickness, stipe length, and stipe diameter, while quantitatively describing stipe shape. The experimental results showed that the YOLOv11s-star-seg model performed well in the instance segmentation of O. raphanipes, achieving an mAP50 of 99.1%, an F1 value of 98.9%, and an mIoU of 97.86%; compared with the YOLOv11s-seg model, the number of parameters was reduced by 39% and the inference time was reduced by 27%, thereby achieving model lightweighting while maintaining high segmentation accuracy. The automatically extracted results showed high consistency with the manual measurements; the coefficients of determination (R2) from linear regression fitting for stipe length, stipe diameter, and pileus diameter all exceeded 0.96, indicating that the method can effectively replace part of the manual measurement process. The results can provide technical support for the digitalization of DUS testing of O. raphanipes, intelligent breeding of edible mushrooms, and accurate phenotypic identification.
J. Fungi
16 October 2026



![Application–mechanism–endpoint map constructed from the selected original studies. Supporting studies address inoculum production [3,4,5,21], crop performance [2,7,8,9,10,11,12,23,24,25,27], contaminated soils [13,14,15], and plant metabolites [16,17,18,19,20]. M denotes mechanistic evidence, C controlled proof of concept, and F field observations; these categories describe study setting and scope, not a formal risk-of-bias grade. The figure does not pool effect sizes or imply equivalent support across domains.](https://mdpi-res.com/cdn-cgi/image/width=281%2Cheight=192/https://mdpi-res.com/jof/jof-12-00777/article_deploy/html/images/jof-12-00777-ag-550.jpg)



