High-fidelity damage classification of milled rice grains using deep unsupervised learning

Abstract

The quality evaluation of processed rice grains is an important factor in determining market acceptance, pricing, storage stability, processing quality, and overall consumer approval. Damage classification based on visual symptoms of raw rice grains allows for very effective quality evaluation. In the current literature, the machine vision methods are predominantly based on supervised machine learning which is fundamentally dependent on manual labelling. However, manual labelling faces issues like erroneousness, subjectiveness and overlapping classes. There exists no work in the current literature which presents an unsupervised approach to classifying rice grain damages. In this study, a deep unsupervised method Contrastive-RC [KM1] is developed for fine-grained damage classification of processed white rice, leveraging contrastive self-supervised learning technique. In particular, self-supervised contrastive learning (SimCLR) is used for feature representation followed by dimensionality reduction (UMAP) and clustering (HDBSCAN). For this, a large dataset of 20,134 high magnification (24 MP) images of individual rice grains spread across different damages was collected. I have been successful in clustering the rice grains into six main cluster based classes with well defined attributes. The class names along with the number of corresponding instances are: normal-damage (5599), chalky-discoloured (4987), discoloured (3215), half-chalky (2386), healthy (2061), and broken (931). Further, it is also presented how the method can be extended to subclassify these damages according to the user’s needs by providing a low-level control, enabling the method to be used in multiple use-cases. The method is fast, versatile and robust towards changes in messy variables like brightness, grain orientation, etc., making it ideal for real world use and extension to other varieties of white processed rice. Overall, this study presents a deep unsupervised method Contrastive-RC for fine-grained damage classification of processed white rice, leveraging contrastive self-supervised learning technique, which could be utilised as a tool for better and more objective quality assessment of the damaged rice grains at market and trading locations.