Image Classification Using Transfer Learning

Grapevine diseases such as Black Rot, Black Measles, and Leaf Blight can quickly damage vineyards, reducing both yield and farmer income. Deep learning—especially CNNs—offers a strong way to automate disease detection, but training these models from scratch needs large datasets. Transfer learning solves this by reusing pre-trained networks for faster, more accurate classification. This study evaluates VGG16, ResNet101V2, DenseNet121, InceptionV3, Xception, and MobileNet for grapeleaf disease detection. Earlier work reports DenseNet121 reaching 99.67% accuracy [1]. Our analysis compares model behavior and training patterns, showing that transfer learning is a reliable and efficient approach for agricultural image diagnosis.