AI in Agriculture: A Federated Learning CNN Approach to Detecting Almond Leaf Disease

The proposes an original method of detecting and classifying almond leaf diseases with a CNN-federated learning framework. Aside from the four severity levels of almond leaf diseases, there are also four kinds of clients. There is no loss of independent client-side confidential data or subversion of a global model based on localized data input. A primary objective of the project was to increase accuracy and efficiency in disease classification for almond cultivation with the help of distributed computing power. Four different clients with different local data characteristics were used to train the CNN model. Diseases are graded based on severity, facilitating accurate and timely detection, which is critical to crop management success. The model's performance has been carefully evaluated locally and globally for its accuracy and reliability. The macro averages across clients also reflect the model's class-wise performance. The highest score was 94.68. The weighted average (which takes class frequency into account) also reached 94.69, showing consistency across classes. As usual, the highest of all the micro averages is 94.68, which is what people need to understand performance when class distributions are not balanced. This study constitutes an important step in agricultural AI, particularly in detecting and classifying plant diseases. In addition to being statistically significant, these results also have practical implications. As a result of this accomplishment, agriculture can now benefit from more advanced and decentralized AI that preserves privacy.