Abstract
The early identification of crop diseases plays a crucial role in ensuring food security and minimizing agricultural losses. This study investigates the use of deep convolutional neural networks (CNNs) for the multiclass classification of apple leaf diseases. Five prominent CNN architectures, AlexNet, VGG16, ResNet50, DenseNet121, and GoogLeNet, were evaluated using a curated dataset, consisting of four classes: apple scab, black rot, cedar rust, and healthy leaves. Extensive experiments were conducted with two optimization algorithms (Adam and SGD) across multiple learning rates. DenseNet121 consistently outperformed the other models, achieving a classification accuracy of 99.75% with robust precision, recall, and F1-score values. The results demonstrate the effectiveness of transfer learning, optimizer tuning, and class balancing in enhancing disease detection performance. This work supports the application of DenseNet-based systems in real-world agricultural diagnostics and highlights the importance of architectural and hyperparameter choices in deep learning applications