Advanced Image Recognition for Mango Leaf Disease Diagnosis: A CNN-Based Study

The abstract presents an all-rounded evaluation of the model crafted for classifying various mango leaf diseases, building its performance metrics, including precision, recall, F1-score, support, support proportion, and accuracy. Model Performance With a precision score of 97.32%, Red Rust was considered outstanding, with Black Tip garnering the second-highest at a precision score of 97.39%. Other diseases such as Sooty Mold, Leaf Gall, and Dieback displayed slightly lower precision values with 96.89%, 96.95%, and 96.91% respectively. With 10,380 photos in the dataset, the model's overall accuracy was a remarkable 97.10%. The recall values were also equally as impressive as the precision values, thus indicating a good model in terms of correctly identifying positives within each class as well since the F1-scores indicate that the performance is well-balanced. The excellent level of generalization ability was outlined by an overall accuracy of 99% for all classes of disease; it can be noted that Bacterial Canker and Phoma Blight had a precision of 97.01% and recall of 97.30%, respectively. The support values ranged from 1,445 for Sooty Mold to 1,530 for Black Tip dataset is fairly well spread. Support ratios were maintained at around 0.14 to 0.15 by each of the classes, thus pointing toward a significant enough presence of instances. Thereby, results produced reveal only the effectiveness of the model in anticipating and classifying cases of mango leaf diseases with accuracy but at the same time throw light on the imperative this model may have in the realm of agriculture for mechanisms to be chosen well in time for disease management. This development in predictive models presents a power tool for mango growers by promising healthier and higher crops.