The classification of Toor Dal (pigeon pea) based on quality is essential for fair pricing, consumer trust, and efficient industrial processing. Traditional methods rely heavily on manual inspection, which is often subjective, labor-intensive, and inconsistent. This paper presents a comprehensive study of AI-driven approaches for automated Toor Dal classification using both classical machine learning and modern deep learning techniques. Various image preprocessing and feature extraction methods such as Gray Level Co-occurrence Matrix (GLCM), Histogram of Oriented Gradients (HOG), and histogram analysis are used to enhance visual inputs. Classical models including Support Vector Machines (SVM), K-Nearest Neighbors (KNN), and Decision Trees are evaluated alongside deep learning models like Convolutional Neural Networks (CNN), ResNet, and MobileNet. The You Only Look Once (YOLO) object detection framework is also implemented using annotated images via Roboflow, achieving an accuracy of 91%. Comparative analysis highlights performance based on accuracy, processing time, and F1-score, with deep learning models generally outperforming traditional approaches. Challenges such as dataset limitations, real-time deployment, and mobile optimization are addressed, and possible future directions such as hybrid AI models, edge deployment, and AI-IoT integration are discussed. The findings demonstrate the potential of AI in enhancing food quality assurance and establishing scalable grading systems in agriculture.