Utilizing X-ray radiography for non-destructive assessment of paddy rice grain quality traits

Abstract

Abstract Background Agricultural systems are under extreme pressure to meet the global food demand, hence necessitating faster crop improvement. Rapid evaluation of the crops using novel imaging technologies coupled with robust image analysis could accelerate these improvements. This study assesses the potential of X-ray imaging for non-destructive evaluation of rice grain traits. By analyzing 2D X-ray images of paddy grains, we aim to approximate their key physical grain traits important in rice breeding: (1) T 1 chaffiness, (2) T 2 chalky rice kernel percentage (CRK%), and (3) T 3 head rice recovery percentage (HRR%). Successful integration of X-ray imaging and data analysis into the breeding process could prospectively revolutionize rice breeding and improve global agricultural productivity. Results The study aimed to predict the key rice traits (chaffiness, CRK%, HRR%) using 2D radiographs obtained from high resolution X-ray imaging systems. The accuracy of trait inference algorithms was evaluated by comparing the predicted values with ground-truth measurements. We showed that all three traits can be predicted with reasonable accuracy (chaffiness: R 2 = 0.9987, RMSE = 1.302; CRK%: R 2 = 0.9397, RMSE = 8.91; HRR%: R 2 = 0.7613, RMSE = 6.83). Conclusions Our study demonstrated that multiple key physical grain traits important in rice breeding (chaffiness, CRK%, and HRR%) can be inferred from single 2D X-ray images of whole paddy grains. Such a non-destructive rice grain trait inference is expected to improve the robustness of paddy rice evaluation, as well as to reduce time and possibly costs for rice grain trait analysis. Furthermore, the described approach can also be transferred and adapted to other grain crops.

Plant phenotyping relevance

X線画像からイネ籾の品質形質を非破壊推定する画像解析手法を開発・精度評価しており、表現型取得が研究の中心である。

Code and data availability

The supplied blocks describe X-ray radiography of paddy rice grains and PCA/SVM-based trait inference, but contain no data availability statement, deposited phenotype/image datasets, or author analysis code with a public URL. The only URLs present (Fraunhofer CTportable and Volex10 pages) are commercial equipment pages

No evidence-backed public reproduction asset is currently recorded.