DetectPLD: A Federated CNN Approach for Collaborative Potato Leaf Disease Detection

The potato stands out as one of the most commonly cultivated crops globally. The spread of diseases such as potato leaf disease can significantly impact its quality and yield worldwide. Federated learning is a machine learning approach that allows multiple entities to collaboratively train a model without needing to share their individual data. In the context of potato farming, our research proposes federated learning (FL) to detect potato leaf disease across two clients without sharing sensitive data between them. Each client is trained with the CNN model and its different architectures using their dataset in this approach. Then the models are aggregated to create a global model and this global model is then used to detect potato leaf disease across multiple farms, improving the accuracy of disease detection and enabling early intervention to limit its spread. For every round of the updated global model, the accuracy of our model has improved significantly. After three rounds of communication using the Inception-V3 model on the server, the accuracy and F1-score were 88% and the precision and recall were around 89% respectively. These experiments were conducted using different custom and pre-trained CNN architectures to understand the improvement in results after collaborating with different datasets.