Disease Detection in Finger Millet Plant Using Fine-Tuned Transfer Learning Models

Plant diseases are becoming one of the major threats to global food security, particularly in carbohydrate rich staple crops like finger millet (Eleusine coracana) which requires minimal water and can be grown in diverse climate conditions, highlighting the need for early disease detection to minimize yield losses and promote sustainable agriculture. We employed web scraping for data collection from different open source, data augmentation techniques like flip, rotation, zoom to enhance the model's performance and enhance generalization capabilities. This research focuses on developing an effective disease detection system for finger millet using Convolutional Neural Networks (CNNs) and transfer learning methodologies. We analyzed with pretrained CNN architectures, such as DenseNet121, VGG16, and VGG19, using a dataset of more than thousands of labelled images collected from diverse agricultural settings. Fine tuning these models ensures optimising model's performance and achieving better accuracy. Among the tested models, DenseNet121 achieved the highest accuracy, reaching 96.88%. Additionally, we developed a user-friendly real-time application to disease detection using Streamlit library helping the farmers to detect the disease in mobile for remote access. This study demonstrates the significant use of deep learning techniques like CNN architectures, particularly transfer learning, for developing better crop disease detection systems. By detecting the disease early and prevention at earlier stage will save tonnes of crops. This solves the major issues like food security and also malnutrition.