Contenu principal de l'article
Paddy leaf diseases cause major issues in yield production and quality of rice. The initial stage analysis and classification of paddy leaf diseases in prior knowledge can reduce the spread of pathogens across the field and increase yield production. This paper proposes an accurate classification of paddy leaf diseases, namely bacteria leaf blight, leaf spot, blast, hispa, and leaf folder based on the Deep Convolutional Neural Network and Random Forest. Novel data-set of 2,088 images collected using Canon EOS 1200D, FLIR E8’s camera, and stored in the paddy image repository. The leaf portion was extracted from a complex data-set using K-Means Clustering techniques. The count and quality of the image enhanced using Generative adversarial network techniques (GAN). GAN augmentation generates images of 18,317, features extracted using the Deep Convolutional Neural Network model, and diseases classified using Random Forest. The interpretation was made over the architecture models, namely AlexNet, GoogleNet, VGG, ResNet, and Inception-ResNet. The resultant value states that Inception-ResNet produces better accuracy of 95.166 compared with the remaining standard models. ResNet-50 model parameters are reduced by 24,769,690 compared with Inception-ResNet model. This research work indicates that the proposed system produces better accuracy with less error rate of 0.0230 in the classification of paddy leaf diseases.