Early Detection of Disease in Millet Crops Using Transfer Learning

Abstract

Millet crops play a foremost role in ensuring food security and sustainable agriculture; however, their productivity is significantly affected by various leaf diseases such as Rust, Smut and Blast. Early and accurate detection of these diseases is key to effective crop management. In this study, a hybrid Transfer Learning–Convolutional Neural Network (TL–CNN) framework is designed for automated millet disease detection. The model uses pre-trained architectures such as MobileNetV2, Xception and EfficientNetB0V2, for feature extraction, followed by additional convolutional blocks to capture discriminative features. The novelty of this work lies in the development of a millet-specific deep learning framework, integrated with CNN-based feature optimization and a reliable evaluation strategy. Unlike classic approaches, the proposed model uses both single-run training and stratified 5-fold cross-validation to ensure reliable and unbiased performance assessment. A comprehensive comparative study is conducted across multiple transfer learning models within an integrated framework. Experimental data show that the proposed EfficientNetB0V2 + CNN model achieves superior performance, with higher accuracy and better precision, recall and F1 Score across all classes. Confusion matrix analysis additionally confirms minimal misclassification and strong class separability. The data highlight the effectiveness of the suggested approach in detecting complex disease patterns. Overall, the proposed TL–CNN framework delivers a reliable and efficient solution for millet disease detection. This solution provides a strong prospect for real-world agricultural applications and intelligent crop-monitoring systems.