Synergistic Solutions: Federated Learning Meets CNNs in Soybean Disease Classification
Ankita Suryavanshi, Vinay Kukreja, Sushant Chamoli, Shiva Mehta, Ashish Garg · 2024
Modern agricultural demands call for creative, scalable, and flexible solutions, particularly for the early diagnosis and control of crop diseases. Using a Convolutional Neural Network (CNN) model and federated learning across five clients—O1 to O5—this research study explores a novel method for categorizing five kinds of soybean leaf diseases. Using federated averaging techniques, the local models, each trained on a different data set, are combined into a global model, ensuring data privacy while improving predicted accuracy. Our study of the results indicates numerically solid values that support the model's effectiveness. Client O1 showed a precision of 86.71, recall of 86.47, and F1-score of 86.55, with an accuracy of 0.95 for the federated averaging metrics. These numbers were 83.30, 83.22, 83.24, and 0.93 for client O2. Client O4 had 92.34, 92.06, 92.19, and 0.97; client O5 had 90.31, 90.37, 90.34, and 0.96; and client O3 had 88.20, 88.11, 88.12, and 0.95. The aggregate averages further show the robustness of the model. The weighted averages covered the range from 83.39 to 92.24, while the macro averages for all customers varied from 83.25 to 92.20. The micro standards, which ranged from 83.40 to 92.24, were likewise strongly correlated. In conclusion, the federated learning-based CNN model represents a paradigm shift in categorizing soybean leaf diseases. Its high precision, recall, F1 scores, and accuracy rates show the model's potential for real-world applications. It provides an efficient, privacy-preserving, and scaleable solution for regional and worldwide agricultural landscapes.