DeepRoseNet: An Explainable Hybrid Deep Learning Approach for Precise Rose Leaf Disease Classification

Rose cultivation faces significant risks due to leaf diseases, resulting in reduced production and potential nation-wide export restrictions. While numerous studies have explored the detection of rose leaf diseases using deep learning techniques, existing methods have yet to achieve consistently high accuracy, emphasizing the need for more effective detection approaches. This research presents a robust hybrid deep learning model that integrates EfficientNetV2 and ResNet-152 with customized layers to accurately classify rose leaf diseases. Additionally, explainable AI (XAI) techniques, such as Grad-CAM++ and LIME, enhance model transparency and reliability. This study utilizes a rose leaf disease dataset containing 7,350 images across three categories: healthy rose leaf, rose rust, and rose sawfly or rose slug, with each category comprising 2,450 images. The dataset was preprocessed using resizing, data augmentation, and batch normalization to optimize it for deep learning models. We then implemented a hybrid deep learning model integrating EfficientNetV2 and ResNet-152 models to construct the classification system. The proposed method achieved an impressive average classification accuracy of 97.39% using 3-Fold cross-validation, outperforming existing techniques. This study addresses the critical challenge of detecting rose leaf disease, contributing to improved disease management in rose cultivation.