PDS-MCNet: a hybrid framework using MobileNetV2 with SiLU6 activation function and capsule networks for disease severity estimation in plants

Advanced technologies like deep learning have been widely implemented in various agricultural applications, including disease severity estimation. In this study, the authors have leveraged the computational capabilities of MobileNetV2 and capsule networks (CapsNet) to effectively evaluate plant disease severity. In this paper, the authors implemented a hybrid framework constituting multilateral MobileNetV2 for feature extraction and CapsNet for classification, applied to the problem of estimating disease severity in plants, executed on two datasets: one for tomato late blight disease and another for tomato early blight disease. Furthermore, the authors improved the MobileNetV2 architecture by replacing ReLU6 with the SiLU6 activation function, resulting in improved outcomes. To test the robustness of the proposed methodology, the authors added salt-and-pepper noise to the original images and computed the performance measures. Additionally, the impact of data fusion, i.e., combining two datasets, one tomato early blight disease from the PlantVillage dataset and an original tomato early blight (TEB) dataset collected by the authors, was also investigated for disease identification (binary classification) as well as disease severity assessment. Our proposed framework was compared with 15 traditional pre-trained CNN models where classification accuracy, Cohen’s kappa, precision, recall, F1-score, and loss were recorded and compared. A rank, on the basis of performance, was assigned to all models for all implementations. Based on the mean rank values, it was observed that the proposed PDS-MCNet model outperformed all other models, followed by DenseNet169, the best amongst the state-of-the-art CNN architectures, for plant disease severity evaluation. The proposed approach was also validated using three other datasets while establishing the generalizability of the technique. Accurate and timely assessment of disease severity can benefit growers economically by enabling them to take precise corrective actions. Early detection and precise application of agrochemicals can also benefit the environment and maintain the ecological balance.