A Comprehensive Framework For Automated Tur Dal Variety Classification Using Computer Vision And Machine Learning
Tur dal, renowned as one of the most popular pulses globally, encompasses a wide range of varieties with significant variations in texture, colour, and other attributes. Accurately identifying tur dal varieties, is essential to satisfy consumer demands and uphold consumer rights simultaneously This research presents an intelligent computer vision framework designed for the automated and cost-effective identification of five distinct tur dal varieties: Asha, Nagpur, Navapur, Gulyal, and Latur. The methodology involves a systematic process of sample preparation, high-resolution image acquisition under controlled lighting, rigorous image preprocessing including resizing, noise reduction, segmentation, and data augmentation. The color, morphological and texture features are extracted from the processed images. These features, along with deep features from ResNet50, are utilized to train and evaluate three machine learning classifiers: ResNet50, Support Vector Machine (SVM), and Random Forest (RF). Experimental results demonstrate that the ResNet50 model achieves the highest classification accuracy of 97.92%, outperforming SVM and Random Forest. The study highlights the efficacy of the developed framework in standardizing the classification process, enhancing operational efficiency, and providing a reliable alternative to traditional manual inspection methods