Automated plant diagnosis has a lot of promise to increase agricultural productivity, but the adoption of drone-based solutions is hampered by issues including the trade-off between processing speed and image resolution and the scarcity of labeled training data. To address these challenges, this research presents a novel two-step machine learning approach that uses Convolutional Neural Networks (CNN). Our approach guarantees the production of representative data from UAV photos while efficiently addressing class imbalance in datasets. Our method, which focuses on a dataset of apple trees with class imbalance, entails preprocessing the images, building a CNN architecture with dropout layers strewn in between convolutional and pooling layers to mitigate overfitting, and then training the model to distinguish between images that are diseased and those that are not. Our model then performs a two-step approach to identify possibly unhealthy plants and offer actual diagnosis. The experimental results provide a remarkable 80.90% accuracy rate on training data and 74.79% on test data, demonstrating the efficacy of our CNN-based drone technology for automated crop disease diagnosis and providing a viable substitute for Labor-intensive diagnostic techniques.