Precision agriculture has significantly benefited from advancements in deep learning and machine learning techniques for plant disease detection and classification. This study focuses on developing a real-time plant disease detection system utilizing a newly curated dataset covering three major crops—rice, wheat, and maize. The dataset comprises multiple disease categories, including early, advancing, and severe stages, under complex real-world conditions. Various pre-trained deep learning models, including Xception, MobileNetV2, and InceptionV3, were fine-tuned and evaluated on the dataset. Additionally, a novel Convolutional Neural Network (CNN) architecture, Rice Maize, Wheat-NN model (RMW-NN), was proposed and trained from scratch. Experimental results demonstrate that the RMW-NN outperformed existing state-of-the-art models, achieving a testing accuracy of 97.06%, 97.04%, and 98.08% for rice, maize and wheat disease classification, respectively. The study highlights the efficacy of deep learning(DL) in automated plant disease detection, emphasizing the need for high-quality, real-world datasets to improve model generalization. Future directions include expanding the dataset to include more crop varieties and integrating object detection models for precise localization of disease-affected regions.