Development and Testing of a CNN-Based Smart Web Application Deep Learning for Early Detection of Plant Diseases

The primary aim of the current research is to solve this problem that rises at early stage thereby making detection and management easier for plant diseases in hibiscus leaf using deep learning techniques specifically ResNet50 model. The dataset for the study comprised 2500 imges, including healthy leaves and affected ones by powdery mildew, downy mildew or botrytis. These images were preprocessed significantly (resizing, normalization, augmentation) to be ready for being trained. Three deep learning models VGG16, VGG19 and ResNet50 were used for the comparison of these three techniques on disease classification with best performance achieved by ResNet 50 (96.89% accuracy). The research methodology especially around training each model on the dataset, testing them with structured data accuracy along precision and recall metrics using Confusion Matrices provided by libraries such as yellowbrick as well, also integrating ResNet50 within a web application. This app (real time disease identification) would enable the user to upload an image of a hibiscus leaf and would display how accurate resnet50 is able classify healthy or diseased leaves. The results also demonstrate the superiority of ResNet-50 in disease detection compared to other models and prove that this model is capable, strong generalizable feature extractors for agricultural uses. The findings presented here contribute towards the development of precision agriculture as they supply a useful new tool for disease detection at an early stage, enabling easy and rapid intervention mitigating crop losses and promoting agricultural sustainability.