Mycelium phenotypic characteristics of edible mushroom are an important basis for the evaluation of edible mushroom germplasm resources and scientific breeding. To address the problems of traditional threshold segmentation method to extract mycelium regions which are easily disturbed by uneven light, irregular growth of mycelium and metabolites produced in the petri dishes, an image dataset of edible mycelium was made and a deep learning-based automatic measurement method for edible mycelium phenotype parameters was proposed. The U-Net network encoder was partially replaced with the first 13 convolutional layers of VGG16, and pre-training weights were introduced to construct a VGG-UNet model applicable to mycelium segmentation. The average cross-merge ratio of this model reached 98.18%, which was 0.93 percentage points higher than that of the original U-Net model. After obtaining mycelium segmentation images by this model, the five phenotypic parameters of radius, perimeter, area, coverage, and roundness of mycelium were calculated by using OpenCV correlation functions. A linear regression analysis was performed between the manual measurement method, and the R2 of mycelium radius, perimeter, area and coverage were 0.9795, 0.9915, 0.9750 and 0.9750, respectively, and the RMSE were 2.20mm, 4.73mm, 176.74mm2 and 3.16%, respectively. The method was tested to accurately accomplish the task of automatic measurement of phenotypic parameters of edible mycelium, which provided a theoretical basis for the study of phenotypic analysis of edible mushrooms.