Convolutional Neural Network Approach for Early Detection of Green Mold Disease in Button Mushroom (<i>Agaricus bisporus</i>)

The white button mushroom (Agaricus bisporus), that contributes for 35–45% of global production, is the most commercially significant edible fungus in the world. Trichoderma harzianum caused green mold disease, which may decrease output by as much as 67%, poses a serious threat to its cultivation. Sustainable mushroom production depends on the early and precise detection of this disease because conventional diagnostic techniques are ineffective and time-consuming. This study uses a customized VGG16 Convolutional Neural Network (CNN) to suggest an automated method for identifying green mold disease in mushrooms. 2,316 photos of healthy, green mold initial and green mold final stage were collected from two significant cultivation locations in Pakistan to make up the dataset. To improve generalization and avoid overfitting, data augmentation techniques were used during the model's training. Its performance was assessed using metrics such as ROC analysis, F1-score, confusion matrix, recall, accuracy, and precision. It obtained 90% training accuracy, 82% validation accuracy. The study demonstrates how AI-based methods may be used to reliably detect green mold disease early on, facilitating prompt management and increased mushroom yield.