Plant diseases pose a significant threat to agricultural productivity and the health of green plants if not accurately diagnosed at an early stage. Traditional disease detection methods are often time-consuming, costly, and insufficiently effective. Therefore, there is a growing need for advanced intelligent approaches capable of improving the accuracy and efficiency of plant disease diagnosis. In this context, this study addresses the challenges and research gaps in plant disease detection and classification using deep learning techniques, particularly Convolutional Neural Networks (CNNs). The study proposes an EfficientNet-B3-based model for multiclass plant disease detection and classification involving 38 categories of healthy and diseased plant leaves. The model was trained and evaluated using a publicly available plant disease dataset obtained from Kaggle. The proposed framework integrates the EfficientNet-B3 architecture with gradual fine-tuning strategies, the AdamW optimization algorithm, and regularization techniques to enhance classification performance and reduce overfitting, The performance of the proposed model was evaluated using several standard metrics, including Accuracy, Precision, Recall, F1-Score, and ROC-AUC Score. Experimental results demonstrated that the EfficientNet-B3 model achieved outstanding performance, attaining an accuracy of 99.92% on the training and validation datasets. The findings indicate that the combination of transfer learning, gradual fine-tuning, AdamW optimization, and regularization methods significantly improves the robustness and effectiveness of the model compared with traditional approaches and previously reported studies................. Keywords:.................AdamW, Efficient Net B3, Gradual Fine Tuning, plant diseases.