Scholarlink

<p>Phenotypic selection plays a vital role in rice breeding programs by enabling the identification of superior rice genotypes exhibiting desirable traits and predicting their potential yield. This process is crucial for enhancing the efficiency of breeding programs, as it allows breeders to focus on the most promising lines. This study focuses on evaluating the phenotypic performance and yield prediction of elite rice gene pools across two contrasting Indonesian environments, aiming to provide valuable insights for optimizing rice breeding strategies. A smartphone-based image retrieval system and logistic regression model were used to classify rice yield based on panicle appearance by acquisition of 1050 sets of data images. The study found a training accuracy of the processed image of 0.854 with a validation accuracy of 0.724. A significant genotype-environment interaction was observed for yield, while showing no notable differences in other traits. Principal Component Analysis of 108 genotypes showed the first principal component accounted for 48.35% of the variance, while the second principal component explained 19.71%. Correlation analysis indicated a positive relationship between maturity and grain yield (<i>r</i> = 0.64), suggesting that late-maturing genotypes may yield higher, albeit with trade-offs in traits such as plant height, tiller number, and 1000-grain weight. Among the tested lines, BP35264b-Sub-2-Sub-2-8-4 (9.84 ton ha<sup>−&#xa0;1</sup>), BP35264b-Sub-3-Sub-3-13-9 (9.33 ton ha<sup>−&#xa0;1</sup>), BP35264b-Sub-3-Sub-3-37-1 (8.82 ton ha<sup>−&#xa0;1</sup>), BP35264b-Sub-2-Sub-2-8-3 (8.75 ton ha<sup>−&#xa0;1</sup>), and BP35264b-Sub-2-Sub-2-8-5 (8.68 ton ha<sup>−&#xa0;1</sup>) exhibited superior performance compared to all check varieties. A 3D PCA plot revealed distinct clusters across yield categories, underscoring the diverse characteristics of rice genotypes and the environmental influences on productivity. Smartphone-based image retrieval and logistic regression have demonstrated effectiveness in classifying rice yield based on panicle appearance, presenting a rapid yield assessment method. Predictive models derived from multi-environmental trials offer insights into factors influencing rice yield, potentially improving breeding strategies and scalability.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Smartphone-based image analysis, trait correlation and PCA for rice yield prediction in multi-environment trials

  • 期刊论文

  • 发表日期：

    2025年10月27日

查看全文
  • Trias Sitaresmi,
  • Muhammad Aqil,
  • Willy Bayuardi Suwarno,
  • Hajrial Aswidinnoor,
  • Munif Ghulamahdi,
  • Arlyna Budi Pustika,
  • Nani Yunani,
  • Yudhistira Nugraha

摘要

Phenotypic selection plays a vital role in rice breeding programs by enabling the identification of superior rice genotypes exhibiting desirable traits and predicting their potential yield. This process is crucial for enhancing the efficiency of breeding programs, as it allows breeders to focus on the most promising lines. This study focuses on evaluating the phenotypic performance and yield prediction of elite rice gene pools across two contrasting Indonesian environments, aiming to provide valuable insights for optimizing rice breeding strategies. A smartphone-based image retrieval system and logistic regression model were used to classify rice yield based on panicle appearance by acquisition of 1050 sets of data images. The study found a training accuracy of the processed image of 0.854 with a validation accuracy of 0.724. A significant genotype-environment interaction was observed for yield, while showing no notable differences in other traits. Principal Component Analysis of 108 genotypes showed the first principal component accounted for 48.35% of the variance, while the second principal component explained 19.71%. Correlation analysis indicated a positive relationship between maturity and grain yield (r = 0.64), suggesting that late-maturing genotypes may yield higher, albeit with trade-offs in traits such as plant height, tiller number, and 1000-grain weight. Among the tested lines, BP35264b-Sub-2-Sub-2-8-4 (9.84 ton ha− 1), BP35264b-Sub-3-Sub-3-13-9 (9.33 ton ha− 1), BP35264b-Sub-3-Sub-3-37-1 (8.82 ton ha− 1), BP35264b-Sub-2-Sub-2-8-3 (8.75 ton ha− 1), and BP35264b-Sub-2-Sub-2-8-5 (8.68 ton ha− 1) exhibited superior performance compared to all check varieties. A 3D PCA plot revealed distinct clusters across yield categories, underscoring the diverse characteristics of rice genotypes and the environmental influences on productivity. Smartphone-based image retrieval and logistic regression have demonstrated effectiveness in classifying rice yield based on panicle appearance, presenting a rapid yield assessment method. Predictive models derived from multi-environmental trials offer insights into factors influencing rice yield, potentially improving breeding strategies and scalability.