Early and accurate diagnosis of lemon leaf diseases is crucial to improve crop yield and quality. In this study, the potential of DL models in detecting and classifying lemon leaf diseases is explored, specifically the Inception V3 model. Various DL architectures like ResNet50, VGG16, MobileNetV2, DenseNet121, and EfficientNetB0 were evaluated using training accuracy, validation accuracy, training loss, validation loss, and overall training time. According to experimental data, Inception V3 is a dependable model for illness prediction, with a small validation loss of 0.4412 and an outstanding validation accuracy of 86.94%. VGG16 had the longest training time (4920.5885 s), whereas MobileNetV2 had the best training accuracy (95%). The paper demonstrates the promise of deep learning in precision agriculture and makes recommendations for more advancements by using explainability AI and data augmentation strategies to increase model performance and interpretability.