Discrimination of filled and unfilled grains of rice panicles using thermal and RGB images

In recent days, the agricultural research community is focusing on the development of different varieties of aerobic rice, as it consumes less water for its growth. In general, the yield of a crop is considered as a critical performance metric to evaluate different varieties of rice. The count of filled grains in panicles provides a measure for the yield of a crop. The evaluation of yield is a laborious, tedious process and requires human intervention. Hence, this study aims to automate the process for differentiating filled and unfilled grains of rice across different genotypes/varieties and also to help agricultural scientists in the rapid evaluation of different varieties. More precisely, this paper proposes two novel methods that involve RGB and thermal images: (a) Discrimination based on RGB Images (DRI) (b) Discrimination based on Thermal Images (DTI). The study of proposed methods on 15 rice-panicles of different genotypes indicates that DRI method, which involves colour of grains, is found to be challenging to discriminate between filled and unfilled grains. Whereas, DTI method, which makes use of thermal images in differentiating filled and unfilled grains, is found to be profoundly convenient. The performance analysis demonstrates that the proposed DTI method, with averaged absolute errors (AAEs) in discriminating filled grains (2.66%) and unfilled grains (11.389%), outperforms the proposed DRI method with an AAEs in discriminating filled grains (10.664%) and unfilled grains (34.296%). The present investigation resulted in the development of DTI method to discriminate against the filled and unfilled grains across genotypes, and it can be used in rice improvement programs in the future.