| 杨海洋,杨佳浩,吕鹏远,格桑曲珍,王文海,王继涛,马骅,曹云娥.基于高光谱技术与机器学习的黄瓜叶片水分含量快速检测研究[J].干旱地区农业研究,2026,(3):265~274 |
| 基于高光谱技术与机器学习的黄瓜叶片水分含量快速检测研究 |
| Rapid detection of cucumber leaf water content based on hyperspectral technology and machine learning |
| |
| DOI：10.7606/j.issn.1000-7601.2026.03.25 |
| 中文关键词: 黄瓜 叶片含水量 高光谱 机器学习 |
| 英文关键词:cucumber leaf water content hyperspectral machine learning |
| 基金项目:宁夏回族自治区国(境)外智力引进计划项目（宁科外专发〔2023)2号）；2025年自治区科技领军人才培养项目（2025GKLRLX22）；林芝市科技项目（LZZX-06） |
|
| 摘要点击次数: 258 |
| 全文下载次数: 13 |
| 中文摘要: |
| 针对传统植物水分含量测定方法耗时费力、需破坏性取样且难以实现田间实时连续监测等问题，基于高光谱成像技术，结合机器学习算法，构建黄瓜叶片水分含量快速无损监测模型。以开花期和结果期黄瓜叶片为研究对象，提取叶片光谱反射率，并采用卷积平滑、多元散射校正（MSC）、移动平均、基线校正和标准正态变量变换（standard normal variate，SNV）等方法进行光谱预处理；进一步结合竞争自适应重加权采样、连续投影算法、无信息变量消除法（uninformative variable elimination，UVE）和遗传算法偏最小二乘法筛选特征波长，分别建立随机森林、极限学习机和卷积神经网络模型。结果表明，基于MSC预处理和UVE特征波长筛选构建的卷积神经网络（CNN）模型表现最优，其在开花期和结果期的预测决定系数分别为R2p=0.844和0.781，RMSE分别为0.021 g·g-1和0.005 g·g-1。研究表明，高光谱成像技术结合MSC-UVE-CNN方法可实现黄瓜叶片水分含量的快速、无损检测，可为黄瓜水分状态监测及精准灌溉管理提供技术参考。 |
| 英文摘要: |
| Traditional methods for determining plant water content are often time\|consuming, labor\|intensive, destructive, and difficult to apply for real\|time continuous monitoring under field conditions. To address these limitations, this study developed a rapid and nondestructive monitoring model for cucumber leaf water content based on hyperspectral imaging technology combined with machine learning algorithms. Cucumber leaves at the flowering and fruiting stages were used as the research objects, and leaf spectral reflectance was extracted. Several spectral preprocessing methods were applied, including convolution smoothing, multiplicative scatter correction (MSC), moving average, baseline correction, and standard normal variate (SNV) transformation. Subsequently, feature wavelengths were selected using competitive adaptive reweighted sampling, successive projections algorithm, uninformative variable elimination (UVE), and genetic algorithm–partial least squares. Based on the selected feature wavelengths, random forest, extreme learning machine, and convolutional neural network models were established. The results showed that the convolutional neural network (CNN) model constructed using MSC preprocessing combined with UVE\|based feature wavelength selection achieved the best performance. The prediction coefficients of determination at the flowering and fruiting stages were 0.844 and 0.781, respectively, while the corresponding RMSE values were 0.021 g·g-1 and 0.005 g·g-1. These results indicate that hyperspectral imaging combined with the MSC-UVE-CNN method can achieve rapid and nondestructive detection of cucumber leaf water content, providing a technical reference for cucumber water status monitoring and precision irrigation management. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
| | |