A CNN-based biological image processing and multimedia framework for rice grain purity analysis using enhanced mobile imaging
Rice grain purity assessment is a persistent challenge in food quality assurance, particularly in informal markets where laboratory-based inspection methods are inaccessible. Conventional manual inspection relies heavily on human perception and experience, leading to subjective, inconsistent, and error-prone evaluations. This paper presents a biological image processing and multimedia-based framework that leverages image enhancement, convolutional neural networks (CNNs), and mobile imaging systems to perform automated rice grain purity analysis using standard RGB images. Rice grains are treated as biological micro-objects whose morphological, textural, and chromatic features are enhanced and learned directly from images captured under real-world conditions. A dataset comprising 46,575 rice grain images across five commonly traded rice varieties was collected using mobile cameras under varying lighting and background conditions. Image enhancement and background augmentation techniques were applied to improve robustness and generalization. A MobileNetV2-based CNN achieved a peak classification accuracy of 98% under optimized training conditions, with consistently high per-class F1-scores. The trained model was integrated into a mobile-accessible multimedia application and evaluated through a user acceptance study based on the Unified Theory of Acceptance and Use of Technology (UTAUT). Results indicate strong technical performance, usability, and practical viability, demonstrating that biological image processing using consumer-grade devices can serve as an effective alternative to traditional rice purity verification methods.