Deep Learning for the Classification of Rose Leaf Disease

Abstract

Rose cultivation is essential to the world's floriculture industry. But because it is frequently plagued with various leaf diseases, both the quality and quantity produced on roses will decline.Early discovery and prompt and correct identification are crucial to effective control of these diseases. This paper presents a deep learning approach to the rose leaf disease diagnosis problem based on convolutional neural networks (CNNs) . A set of 2000 rose leaf images was collected and pre-processed by resizing, normalization and data augmentation techniques to improve model robustness . Five cutting-edge CNN models— VGG16, ResNet50V2, InceptionResNetV2, DenseNet121 and EfficientNetB0—were trained and tested to distinguish between infected and normal leaves. The experiment results indicate that InceptionResNetV2 achieved the best classification accuracy (97.57%) followed by DenseNet121 (97.03%) and ResNet50V2 (95.14%) , while VGG16 and EfficientNetB0 achieved the comparative worse results (88.24% and 90.62%) . Our experimental results justify the necessity of deeper and more complex CNN structures for the plant disease detection problem over the earlier models. This investigation shows the promise of deep learning for automatic, accurate, scalable, and broadcast rose disease classification and decision support in support of farmers and researchers toward intelligent agricultural systems. In future we aim to extend our dataset, and deploy the top performing models to real-time use in mobile or web application for practical usage.