Detection of leaf diseases in ornamental plants using artificial intelligence methods

In this thesis, a novel and lightweight convolutional neural network (CNN) architecture is developed for the classification of rose leaf diseases using image processing and deep learning techniques. The model is trained on the RoseNet dataset, which consists of six distinct classes of rose diseases. A 9-channel input structure is formed by fusing RGB, HSV, and YCrCb color spaces to enhance feature diversity. To strengthen channel-wise attention, the proposed CNN integrates Squeeze-and-Excitation (SE) and Convolutional Block Attention Module (CBAM) blocks sequentially. During training, Z-score normalization, label smoothing, class weighting, and early stopping strategies are applied. The model achieved over 93% in Doğruluk, Kesinlik, Duyarlılık, and F1-score metrics, while maintaining low parameter complexity and fast inference time, making it suitable for resource-constrained environments. Compared to recent studies in the literature, the proposed model presents a competitive performance in terms of both Doğruluk and efficiency. Furthermore, its lightweight structure and explainability highlight its potential for integration into mobile agricultural systems and early disease detection applications.