Fine-Tuned Xception Model for Potato Leaf Disease Classification

Renowned for their nutritional density, potatoes boast higher levels of protein and iron compared to many other vegetables. Nevertheless, farmers face substantial economic risks due to the susceptibility of potatoes to diseases like early blight, late blight, and various leaf-related afflictions. These vulnerabilities pose significant threats to agricultural livelihoods, food security, and community well-being. To tackle the challenges we face, we need to approach them comprehensively, covering both agricultural and social aspects. This issue can be effectively solved by utilizing pre-trained models. These models have already been extensively trained on large datasets and can be tailored to specific tasks. To refine a model, first unfreeze it or a portion of it, and then train it with a low learning rate on new data. Optimizing performance involves fine-tuning the Xception architecture. This goal can be accomplished by changing the parameters, such as the Adam optimizer and fine-tuning learning rate to 1e-4. The four models that will be examined include VGG19, DenseNet121, Xception, and the proposed fine-tuned Xception model. This study uses images of potato leaves from the PlantVillage dataset to evaluate different models. Potato leaf classification into categories like healthy, early blight, or late blight is contingent upon assessing performance metrics such as recall, F1-score, and precision. Drawing from the outcomes of experiments, the proposed fine-tuned Xception model achieves 95.37% accuracy, which is superior to other models. This model is capable of classifying potato leaves and identifying different states of health or disease in potato plants.