In this paper, we propose a real-time rice quality assessment mechanism based on deep learning. It integrates single-stage object detection, deep convolutional neural networks, and traditional machine learning techniques to perform rice variety identification, grain integrity grading, and grain turbidity assessment. Through experiments on a dataset of about 20,000 images of six widely cultivated rice varieties in China, we achieve 99.14% mean accuracy (mAP) in object detection, 97.89% accuracy in classification, and 97.56% average accuracy in grain integrity grading within the same variety. This enables more accurate and efficient rice classification and quality assessment than traditional manual methods.
