School of Agriculture and Biology, Shanghai Jiao Tong University, Shanghai, People's Republic of China.
PLoS One. 2019 Aug 19;14(8):e0221259. doi: 10.1371/journal.pone.0221259. eCollection 2019.
The internal phenotypes of netted muskmelon (Cucumis melo L. var. eticulates Naud.) are always associated with its external phenotypes. In this study, the parameters of external phenotypic traits were extracted from muskmelon images captured by machine vision, and the internal phenotypes of interest to us were measured. Pearson analysis showed that most external phenotypic traits were highly correlated with these internal phenotypes in muskmelon fruit. In this study, we used the random forest algorithm to predict muskmelon fruit internal phenotypes based on the significantly associated external parameters. Carotenoids, sucrose, and total soluble solid (TSS) were the three most accurately monitored internal phenotypes with prediction R-squared (R2) values of 0.947 (root-mean-square error (RMSE) = 0.019 mg/100 g), 0.918 (RMSE = 3.233 mg/g), and 0.916 (RMSE = 1.089%), respectively. Further, a simplified model was constructed and validated based on the top 10 external phenotypic parameters associated with each internal phenotype, and these parameters were filtered with the varImp function from the random forest package. The top 10 external phenotypic parameters correlated with each internal phenotype used in the simplified model were not identical. The results showed that the simplified models also accurately monitored the melon internal phenotypes, despite that the predicted R2 values decreased 0.3% to 7.9% compared with the original models. This study improved the efficiency and accuracy of real-time fruit quality monitoring for greenhouse muskmelon.
网纹甜瓜的内部表型总是与其外部表型相关联。在这项研究中，使用机器视觉从甜瓜图像中提取外部表型特征参数，并测量我们感兴趣的内部表型。Pearson 分析表明，大多数外部表型特征与甜瓜果实中的这些内部表型高度相关。在这项研究中，我们使用随机森林算法，根据显著相关的外部参数，预测甜瓜果实的内部表型。类胡萝卜素、蔗糖和总可溶固形物（TSS）是三种监测最准确的内部表型，预测 R2 值分别为 0.947（均方根误差（RMSE）=0.019mg/100g）、0.918（RMSE=3.233mg/g）和 0.916（RMSE=1.089%）。此外，基于与每种内部表型相关的前 10 个外部表型参数构建并验证了简化模型，并使用随机森林包中的 varImp 函数对这些参数进行了筛选。与简化模型中每个内部表型相关的前 10 个外部表型参数并不相同。结果表明，尽管简化模型的预测 R2 值与原始模型相比降低了 0.3%至 7.9%，但仍能准确监测瓜类内部表型。本研究提高了温室甜瓜实时果实品质监测的效率和准确性。
