Sun Jihong, Sun Chen, Li Zhaowen, Qian Ye, Li Tong
College of Agronomy and Biotechnology, Yunnan Agricultural University, Kunming, Yunnan, China.
The Key Laboratory for Crop Production and Smart Agriculture of Yunnan Province, Yunnan Agricultural University, Kunming, Yunnan, China.
PLoS One. 2024 Dec 13;19(12):e0312444. doi: 10.1371/journal.pone.0312444. eCollection 2024.
The efficacy of generalized sugarcane yield prediction models holds significant implications for global food security. Given that machine learning algorithms often surpass the precision of remote sensing technology, further exploration of machine learning algorithms in the development of sugarcane yield prediction models is imperative. In this study, we employed six key phenotypic traits of sugarcane, specifically plant height, stem diameter, third-node length (internode length), leaf length, leaf width, and field brix, along with eight machine learning methods: logistic regression, linear regression, K-Nearest Neighbors (KNN), Support Vector Machine (SVM), Backpropagation Neural Network (BPNN), Decision Tree, Random Forest, and the XGBoost algorithm. The aim was to establish an intelligent model ensemble for predicting two crucial phenotypic characteristics-stem diameter and plant height-that determine sugarcane yield, ultimately enhancing the overall yield.The experimental findings indicate that the XGBoost algorithm outperforms the other seven algorithms in predicting these significant phenotypic traits of sugarcane. Furthermore, an analysis of the sugarcane intelligent prediction model's performance under a specialized data environment, incorporating self-prepared data, reveals that the XGBoost algorithm exhibits greater stability. Notably, the data pertaining to these crucial phenotypic traits have a profound impact on the efficacy of the intelligent models. The research demonstrates that a sugarcane yield prediction model ensemble, incorporating multiple intelligent algorithms, can accurately forecast stem diameter and plant height, thereby predicting sugarcane yield. Additionally, this approach, combined with the principles of sugarcane cross-breeding, provides a valuable reference for the artificial breeding of new sugarcane varieties that excel in stem diameter and plant height, bridging a research gap in indirect yield prediction through sugarcane phenotypic traits.
通用甘蔗产量预测模型的有效性对全球粮食安全具有重大意义。鉴于机器学习算法往往超过遥感技术的精度，因此在甘蔗产量预测模型开发中进一步探索机器学习算法势在必行。在本研究中，我们采用了甘蔗的六个关键表型性状，即株高、茎径、第三节间长度（节间长度）、叶长、叶宽和田间锤度，以及八种机器学习方法：逻辑回归、线性回归、K近邻（KNN）、支持向量机（SVM）、反向传播神经网络（BPNN）、决策树、随机森林和XGBoost算法。目的是建立一个智能模型集成，用于预测决定甘蔗产量的两个关键表型特征——茎径和株高，最终提高总产量。实验结果表明，XGBoost算法在预测甘蔗这些重要表型性状方面优于其他七种算法。此外，在结合自行准备的数据的特定数据环境下对甘蔗智能预测模型性能的分析表明，XGBoost算法表现出更大的稳定性。值得注意的是，这些关键表型性状的数据对智能模型的有效性有深远影响。研究表明，结合多种智能算法的甘蔗产量预测模型集成可以准确预测茎径和株高，从而预测甘蔗产量。此外，这种方法与甘蔗杂交育种原理相结合，为人工培育茎径和株高优良的新甘蔗品种提供了有价值的参考，填补了通过甘蔗表型性状进行间接产量预测的研究空白。
