Rice analyzer: morphological and structural data engineering for deep learning-based rice classification application

Rice is a vital food source that is consumed worldwide. With increasing genetic varieties, mixing of rice has become a major issue in Pakistan’s export market, leading to economic losses and market distortion. The multiple varieties of rice create a challenge for efficient classification, particularly in distinguishing premium Basmati varieties. The work proposes a deep learning application for rice classification using visual images called RiceAnalyzer. The RiceAnalyzer uses the Convolution Neural Network (CNN) based deep learning model that classifies four varieties of Basmati rice: Basmati 2000, Sella 1509, Kainat 1121, and Sella Basmati 1121. The data set comprises 4561 rice grain samples collected from local rice mills. The sample images are captured using a high-resolution scanner at 600 DPI. Novel data engineering techniques are used to analyze morphological, structural, and colorimetric patterns of rice grains. The model achieved an accuracy of 95.13%, with precision rates between 0.94-0.98 for individual rice varieties. The application is deployed on the PakSupercomputing cluster, which provides remote access.