A Novel Improved Framework for Multiclass Rice Disease Detection using Deep Learning

Abstract

The rice yield is poorly impacted due to lack of expertise in identifying the rice diseases in the field. Deep learning architectures are applied for classification of different crop diseases in some studies but they suffer performance degradation, less accuracy and overfitting posing a challenge for implementation in the real rice fields. To overcome the above challenges this study aims to propose a novel framework by fusing Visual Geometry Group16 (VGG16) with Convolutional Neural Network (CNN). The improved framework consists of 18 layers. The Convolution layer is added after pretrained VGG16 with max pooling layer to prevent overfitting. The set of optimal hyperparameters applied to the proposed framework is obtained through rigorous experimentation. The batch normalization and dropout layers are added with focus on improving accuracy and preventing overfitting. The proposed framework is evaluated in two stages. In stage 1 the proposed framework is compared with fine-tuned state-of-the-art VGG16, Inceptionv3, GoogLeNet, Resnet50, DenseNet121 and MobileNetV2. For stage 2 comparative analysis transfer learning models are optimized and compared. The proposed improved framework outperforms all the above-mentioned models in both the stages of comparative evaluation achieving the testing accuracy of 99.66%. The proposed framework performs without any sign of performance degradation and overfitting when tested on different datasets.