Apple foliar diseases create major agricultural productivity issues that require prompt detection through accurate means to manage diseases properly. One main drawback of traditional approaches is their requirement for human-powered manual checking, which causes time-related problems and has an inherent risk of human errors. This research examines a deep learning-based automatic system that diagnoses apple leaf diseases through analysis of the Plant Pathology 2020 dataset. There are a total of 3642 images belonging to healthy leaves or rust, scab, or multiple disease categories in this dataset. The research evaluated and trained five deep-learning network architectures, namely DenseNet121, EfficientNet, GoogLeNet, ResNet, and MobileNet. The augmentation technique like Canny edge detection together with flipping and combination with convolution and blurring helps to enhance the model’s flexibility and adaptability. From the tested model set-up GoogLeNet achieved maximum accuracy yet the performance of ResNet and MobileNet proved inadequate. Ensemble learning with model averaging was used to improve performance by combining predictions from the most effective models. The ensemble model gained 97.08% accuracy which surpassed the results of all single-model predictions. The results demonstrate how ensemble deep learning models excel at automating apple foliar disease classification yet set the foundation for smart agricultural disease detection systems that scale efficiently.