Advancing Foliar Disease Detection in Apple Orchards: Evaluating the Efficacy of the EfficientNetB3 Model in Machine Learning-Based Classification

The performance of a machine learning model in identifying foliar diseases in apple trees is investigated in this work. 23,000 high-quality RGB photos of apple foliar diseases were used to train, verify, and test the model, which is based on the EfficientNetB3 architecture. The model performed admirably throughout training, with a low loss of 0.292115 and a high accuracy of 0.984884. Similar performance metrics were seen in the test and validation stages, with accuracies of 0.875465 and 0.875492 and losses of 0.636863 and 0.654590, respectively. A classification report was also produced in order to evaluate the model's effectiveness for each of the illness groups separately. Out of all of them, “Healthy” had the best recall (0.97) and accuracy (0.98), indicating that it was able to correctly identify and collect healthy leaves. On the other hand, the illness category “rust frog_eye_leaf_spot” had the lowest accuracy, recall, and F1-score (all 0.00), suggesting considerable difficulties in accurately categorizing it. The inverse class frequency approach was employed in the study to tackle the problem of class imbalance within the dataset. This method gives each class a weight that is determined by the inverse of how often that class occurs in the training set. In order to lessen the effects of class imbalance and enhance the model's capacity to correctly identify uncommon classes, the method incorporates these weighted class frequencies into the training process. the study shows how well the model works to identify different foliar diseases that affect apple trees, but it also points out key areas that still need work, especially when it comes to resolving the challenges related to certain disease classes' classification. The performance of the proposed model has been compared with SOTA and got to be the highest accurate model.