Traditional rice grain classification methods are expensive, time-intensive, and require expert human evaluation. Due to the diversity in rice grain types, predefined morphological features in conventional computer vision approaches often fail to generalize effectively. This study introduces a deep learning (DL)--based framework for rice grain purity analysis, leveraging Convolutional Neural Networks (CNNs) for automatic feature extraction. Unlike traditional approaches, the proposed method eliminates manual feature selection by directly learning relevant patterns from rice images. A new dataset of technician-verified rice grain images is introduced, ensuring consistent background illumination for reliable model training. The CNN-based classifier is compared with traditional morphological classifiers, demonstrating a substantial improvement in classification accuracy. The proposed model achieves at least 25% higher accuracy in distinguishing native from foreign grains. This result highlights the superiority of automated deep learning techniques over conventional handcrafted methods in rice grain classification. By integrating CNNs for scalable and efficient quality assessment, this study presents a novel approach to automating rice grain purity detection. The proposed system reduces reliance on manual inspection, enhances consistency, and ensures high-accuracy classification. Additionally, the newly developed dataset advances future research in agricultural automation, providing a benchmark for machine-learning applications in rice grain analysis.