Comprehensive Analysis of Deep Learning Models for Plant Disease Prediction

The major crop across the world is wheat. The growth of the wheat crop is significantly affected by various types of plant diseases. Technology advancements are instrumental in the recognition and prediction of plant diseases by diagnosing the health of plant leaves. To cut losses and achieve intelligent, healthy farming, the use of computer vision and pattern recognition to identify disease has been researched. The rapid and precise automatic detection of diseases is now possible with image recognition techniques. This work focuses on developing methods for wheat plant disease identification using deep learning models. There are many deep learning models proposed by researchers, but the majority provide poor testing results if some variation (rotation, tiling, and other abnormal image orientations) is there in the images; moreover the models do not store relative spatial relationships among the features captured. Thus the intent of this work is to implement the Whe-C-Net hybrid model, which combines features of VGG16 and CapsNet. The VGG16 model is initially employed for feature extraction. After that, misalignment issues with the current deep learning models are dealt with using CapsNet layers. Later, dropouts, sigmoid activation functions, and fully connected layers are employed. To avoid overfitting and create a Whe-C-Net model that is more broadly applicable, dropouts are used. On the dataset of wheat plant photos, the effectiveness of Whe-C-Net is confirmed. Compared to competing models like pre-trained MobileNet, it obtains a better validation accuracy of 98%, which is noteworthy. The accuracy rates for Xception, ResNetAQ1, MobileNet and VGG16 were 96%, 96%, 65%, and 93%, respectively.