An Improved Deep Learning Model for Detecting Rice Diseases

Abstract

Early detection of rice plant diseases could help to quickly eradicate numerous diseases, such as fungi, viruses, and bacteria, consequently increasing rice yield. Traditional techniques for performing this task may not be the best because they take a long time, require experienced personnel, and are susceptible to a variety of infections. As a result, machine learning and deep learning approaches have recently been utilized to overcome these issues and present a more accurate model for detecting rice plant diseases. However, the current machine learning (ML) and deep learning (DL) models for this task produce unsatisfactory results due to many constraints, including high computational expenses and overfitting. To address these limitations and obtain more accurate disease detection for rice plants, we present a hybrid model of MobileNet and DNN (HMobileNetDNN). The small size of MobileNet minimizes computing costs, while the depth and complexity of DNN improve the model's capacity to capture complicated features, yielding satisfying results. Furthermore, the proposed HMobileNetDNN is also compared to four transfer learning-based DL models, namely ResNetV2, InceptionV3, MobilenetV2, and DensNet121, using the Paddy Doctor dataset. We employ several performance metrics to assess the effectiveness and efficiency of the models, like accuracy, precision, recall, F1 score, and area under the curve. The proposed model outperformed the comparing models, achieving values of 0.918, 0.918, 0.907, 0.912, and 0.949 for accuracy, precision, recall, F1 score, and area under the curve, respectively.