Ensemble-based feature fusion for accurate plant disease classification using pre-trained models

Abstract

Agricultural productivity remains seriously threatened by the attacks of plant diseases, even though it is the bedrock of global food security. These diseases, if ignored, can lead to massive crop losses and economic setbacks. Therefore, the development of accurate and efficient plant disease detection systems is essential to preserve plants and promote sustainable agricultural practices. This study examined the potential of ensembled deep learning models that incorporates feature-level fusion to enhance the robustness and accuracy of plant disease detection. Specifically, we utilized pre-trained deep learning architecture (VGG16, Residual Network 50 (ResNet50), and GoogleNet (InceptionV3)) to extract distinctive features representations from plant leaf images. These extracted features were combined to enhance the performance of our ensemble model. For disease classification, the fused features were passed through a Dense layer with 128 units and ReLU activation, followed by a SoftMax classification layer to predict the probabilities of each plant disease class. Experiments were conducted using the New Plant Diseases Dataset. It contains 87,867 image samples of various plant disease species for 38 classes and 14 different crop species. The ensemble model achieved notable results, with an accuracy of 97.0%. This reveals the capability of feature-fusion ensembled learning in improving detection stability and accuracy. The knowledge and application of feature fusion in disease detection can help ensure more accurate, timely, and eco-friendly interventions. This, in turn, can support sustainable agricultural practices.