Multi-Modal Transfer Learning with DenseNet and Random Forest for Accurate Detection of Powdery Mildew in Rice & Maize Crops

The early detection and accurate diagnosis of powdery mildew in rice and maize crops are essential for ensuring optimal yield and quality. This research study proposes a Multi-Modal Transfer Learning with DenseNet and Random Forest approach for the detection and classification of powdery mildew in rice and maize crops. Our proposed method employs transfer learning, where the DenseNet model is pre-trained on a large-scale image dataset to extract useful features from the images. Subsequently, the extracted features are fed into a Random Forest classifier to achieve precise classification of the powdery mildew disease in the crops. Moreover, the vegetation index is provided as an additional input to enhance the accuracy of the classification. Data augmentation techniques are additionally employed to enhance the diversity of the training dataset and bolster the model's capacity to generalize. Experimental results demonstrate that the proposed approach attains high levels of accuracy, precision, sensitivity, and specificity for the classification of powdery mildew in rice and maize crops. The use of transfer learning and vegetation index integration, along with data augmentation, contributes to the effectiveness of our proposed approach. This research study has significant implications in the agriculture industry, as it provides a reliable and efficient means of detecting and classifying powdery mildew disease in rice and maize crops.