EnhancedNet Deep Learning Model for Plant Disease Classification

Agricultural production vitally contribute to country's economic growth and food needs; however, it depends upon the production and quality of crops. Plants are affected easily due to climate changes and low-quality water. Plants health is examined manually by farmers or pathologists; however, it is not viable to examine each plant regularly due to larger areas of field and limited human resource. If the plant disease gets diagnosed at early stage, the precautionary measures can be taken on time and quality can be improved. There are several solutions available based on modern machine learning and deep learning techniques however, precise prediction and classification is challenging due to the limitations of low light or blurry images. We propose a deep learning-based approach which classifies the healthy and diseases plant using blurry images. Our method addresses the problem using improved layers activation, lesser parameters, using radiant pixels and active neurons. We improved the classification of smaller size and blur images by adding the pooling and dropout layers. Our solution addresses the vanishing gradient and high validation loss problems using a smaller number of parameters as compared to other convolutional neural networks. We evaluated the performance of the proposed method on plant village dataset consisting of a diverse collection of 20600 images. We achieved overall 98.3 % accuracy by minimizing the vanishing gradient, high validation loss and overfitting problem.