Machine vision-based detection method for key phenotypic information of shiitake mushroom stipes.

Metadata-grounded summary

Citation abstract

INTRODUCTION: In the field of shiitake mushroom breeding, the difficulty and inaccuracy in measuring key stipe traits severely limit improvements in breeding efficiency and product quality. This study aims to address this technical challenge.

METHODS: A key trait detection method for mushroom stipes based on integrated image processing technology was proposed: 1) Developed the ACmix-ADown-YOLOv11n stipe detection model; 2) Input the detection bounding box into the EfficientSAM network for precise segmentation,then used OpenCV-based techniques to calculate 12 key phenotypic features; 3) Constructed the CoTAttention-YOLOv11n-Ghost-pose algorithm to predict the stipe centerline.

RESULTS: The ACmix-ADown-YOLOv11n model achieved an AP of 93.7% and a detection speed of 23.97 ms; the CoTAttention-YOLOv11n-Ghost-pose algorithm achieved an AP of 97.2%, a recall rate of 96.1%, and a detection speed of 22.09 ms. For different stipe length categories, the R² between predicted and actual values was 0.989 (extremely short, RMSE=0.030), 0.992 (short, RMSE=0.023), 0.989 (middle, RMSE=0.028), and 0.978 (long, RMSE=0.043).

DISCUSSION: Experimental results confirm the effectiveness and reliability of the proposed method. This study provides an efficient and accurate approach for detecting key stipe traits, offering significant support for advancing intelligent shiitake mushroom breeding and enhancing cultivation quality.