Overstepping Existing Classifications: Deciphering Five Severity Tiers in Sugar Maple Tar Spot Disease Through Integrated CNN-GAN Models

This study classifies the Tar Spot disease severity of leaves from Sugar Maple using a large dataset of 7500 high-resolution images. Using transfer learning, the study deploys a convolutional neural network (CNN) architecture and generative adversarial network (GAN) to analyze personally collected images. The developed model achieves an overall accuracy of 96.43%, which inhabits strong qualification in discriminating among the five severity levels. Performance metrics such as precision, recall, and specificity afford a more fine-grained assessment of the model's discriminative capacity for each severity class. A closer examination of the confusion matrix provides a glimpse into how good the model is at distinguishing between levels and points out where improvements are needed. Although it is based on a model with high accuracy, discussions point to the need for unceasing improvement to overcome the subtleties of disease processes and environmental influences. This research recognizes limitations and next steps. Although the model has huge potential in assessing the severity of Sugar Maple leaf disease, refinement is still needed to enhance its practical applicability for forestry management.