Identifying Sri Lankan traditional rice species and analyzing their quality using rice seed images together with deep learning

Abstract

" Image classification and identification have become significant use cases in most of the domains. When it comes to agriculture, it is a preferred method to analyze the quality of the crops. There are several studies on image classification to identify rice species around the world. However, it cannot use image datasets from other countries to perform classifications in the Sri Lankan context with contextual bottlenecks. As well as Sri Lanka has more than 3000 traditional rice varieties, and some are indigenous varieties. This study on two main things with these concerns. The first one is developing an image dataset of Sri Lankan traditional rice species. Also, under the dataset development, the research proposes methodologies for performing image acquisition, segmentation, and background removal of the images. The study’s initial version focuses on developing a dataset of six Sri Lankan traditional rice varieties with five thousand seed images for each rice variety. The second part is to develop a convolutional neural network image classification model to classify the Sri Lankan traditional rice varieties. The developed neural network model is embedded into a back-end application, and it is exposed as an API to perform the rice classification. As the final part, the research focuses on a front-end mobile application to work with the back-end application to perforce a rice identification and rice sample analysis. Finally, the research is a complete solution for rice image identification using a mobile application. "