A hybrid deep learning model for robust and efficient plant leaf disease detection using ResNet50, PCA, and SVM.

Agricultural productivity sustains itself by detecting diseases in the leaves of plants, especially in poor nations where economic growth is greatly affected by delayed or incorrect detection of diseases. Although high classification accuracy is achieved by the application of deep learning techniques like VGG16 and DC-GAN-based architecture, high computational complexity is still an issue. The optimization-oriented hybrid model for the classification of plant leaf diseases developed in this research emphasizes the viability of deployment and computational efficiency over algorithmic innovation. In the model, high-level semantic information is extracted from the images of the plant leaves using a ResNet50 network that was pretrained as a feature extractor. Then, the deep feature representation is reduced in size using Principal Component Analysis (PCA), which decreases the dimensions of the deep feature representation to reduce information redundancy and prevent overfitting. Finally, the multi-class illness classification is performed using a Support Vector Machine (SVM) classifier. For evaluating the model, the publicly accessible PlantVillage data set containing 38 different classes of both normal and diseased leaves was used. The model was found to achieve a training accuracy of 98.9% and a validation accuracy of 89.4% when a standard train-validation split was applied. In order to further assess the robustness of the proposed model, five-fold stratified cross-validation was carried out to attain an average accuracy of 98.63%. In an ablation study, the maximum accuracy obtained was 98.79%. As suggested by the experimental results, the balance between the accuracy of the classification and the computing economy can be achieved using the integration of deep feature extraction and dimensionality reduction with machine learning classifiers. The findings show that, even if the evaluation is done using a controlled set of data, the suggested architecture can serve as an efficient framework for creating plant disease diagnosis systems in precision agriculture with the use of effective computer resources.