Quantitative analysis of watermelon fruit skin phenotypic traits via image processing and their potential in maturity and quality detection

文献类型： 外文期刊

第一作者： Gu, Qing

作者： Gu, Qing;Li, Tong;Gu, Qing;Li, Tong;Zhu, Yihang;Zhang, Xiaobin;Hu, Ziwei;Shi, Jun;Zhang, Leichen

作者机构：

关键词： Watermelon; Breeding; Phenotype; Fruit maturity; Fruit quality; Lacunarity

期刊名称：COMPUTERS AND ELECTRONICS IN AGRICULTURE （ 影响因子：8.9； 五年影响因子：9.3 ）

ISSN： 0168-1699

年卷期： 2025 年 230 卷

页码：

收录情况： SCI

摘要： Accurate phenotypic analysis is crucial for crop breeding and genetic research. Traditionally, watermelon fruit skin phenotypes have been evaluated manually, which limits precision, particularly for complex traits. This study aims to develop a quantitative approach for analyzing watermelon fruit skin phenotypic traits and to assess their potential in detecting fruit maturity and quality. The study primarily introduces the lacunarity algorithm for quantifying watermelon skin texture characteristics. The effectiveness of the lacunarity algorithm was validated through its application in classifying different texture patterns using various machine learning algorithms. In addition to the lacunarity algorithm, features based on the gray-level co-occurrence matrix (GLCM), color features, and stripe area ratio were extracted as skin phenotypic traits. The temporal dynamics of watermelon skin texture at different fruiting stages were evaluated using the lacunarity algorithm and stripe ratio, while all extracted features were used for fruit maturity and quality detection. Results showed that the support vector machine (SVM) outperformed others in texture pattern classification, achieving an accuracy of 0.89 and an F1 score of 0.88, highlighting the effectiveness of the lacunarity algorithm in quantifying watermelon skin texture. The extreme gradient boosting (XGBoost) model performed best for maturity level classification, with an accuracy of 0.76 and an F1 score of 0.77. Variable importance evaluation revealed that lacunarity values with scale windows of two and one ranked as the most critical features. For quality detection, central sugar content yielded the most accurate predictions among all quality indicators. The long short-term memory (LSTM) model demonstrated the best performance in predicting central sugar content, achieving an R2 value of 0.76, an rRMSE of 0.09, and a MAPE of 7.40 %. This study confirms the feasibility of a quantitative approach to watermelon fruit skin phenotypic analysis and provides valuable insights for advancing non-destructive detection techniques and optimizing breeding strategies.

分类号：

  • 相关文献
作者其他论文 更多>>

微信小程序