| 刘柯楠,赵静涛,赵西宁,蔡耀辉,冯全.基于无人机高光谱遥感的马铃薯叶片含水率估测[J].干旱地区农业研究,2026,(3):255~264 |
| 基于无人机高光谱遥感的马铃薯叶片含水率估测 |
| Estimation of potato leaf water content based on unmanned aerial vehicle hyperspectral remote sensing |
| |
| DOI：10.7606/j.issn.1000-7601.2026.03.24 |
| 中文关键词: 马铃薯 叶片含水率 无人机高光谱 连续小波变换 |
| 英文关键词:potato leaf moisture content UAV hyperspectral continuous wavelet transform |
| 基金项目:国家重点研发计划项目（2021YFD190070400）；甘肃省教育厅高校青年博士支持项目（2025QB-044）；甘肃农业大学伏羲青年英才项目（GAUfx-04Y02）；甘肃农业大学青年导师扶持基金（GAU-QDFC-2025-14） |
|
| 摘要点击次数: 275 |
| 全文下载次数: 22 |
| 中文摘要: |
| 叶片含水率是评估作物水分状况、优化灌溉决策的关键指标。为探索基于无人机高光谱影像数据快速、准确获取马铃薯叶片含水率的方法，通过低空飞行获取马铃薯冠层的高分辨率光谱数据，并运用连续小波变换（CWT）对原始光谱进行多尺度分解，将各尺度小波系数与叶片含水率进行相关性分析，利用CWT光谱构建植被指数探讨其快速估测马铃薯叶片含水率的潜力。采用竞争自适应重加权（CARS）方法对光谱数据进行特征筛选，结合偏最小二乘回归（PLSR）、随机森林（RF）和极端梯度提升（XGBoost）实现叶片含水率的高精度估测。结果表明：从尺度１到尺度8，小波系数与马铃薯叶片含水率的相关性先升后降，中等分解尺度在提高相关性方面最佳（相关系数绝对值由0.59提升到0.84）。小波系数构建的经验植被指数在特定尺度下与含水率的相关性会被放大，各指数最佳尺度组合构建的模型优于原始光谱植被指数。基于单一尺度小波系数估测叶片含水率，5尺度CWT光谱经CARS筛选后建立的XGBoost模型性能最佳（R2=0.88，RMSE=1.235%）。尺度5与尺度7特征融合可进一步提升估测模型的性能（R2=0.90，RMSE=1.145%）。综上，通过CWT结合CARS方法选取的多尺度特征进行叶片含水率监测具有更高的估算精度，为马铃薯水分的精细化管理提供了有效手段。 |
| 英文摘要: |
| Leaf water content is a key indicator for assessing crop water status and optimizing irrigation decisions. To explore a rapid and accurate method for obtaining potato leaf water content based on UAV hyperspectral image data, high\|resolution spectral data of potato canopy were acquired through low\|altitude flight. Continuous wavelet transform was applied to decompose the original spectra at multiple scales, and the correlation between wavelet coefficients at each scale and leaf water content was analyzed. The potential of using CWT spectra to construct vegetation indices for rapid estimation of potato leaf water content was investigated. The competitive adaptive reweighted sampling method was employed for spectral feature selection, combined with partial least squares regression, random forest, and extreme gradient boosting to achieve high\|precision estimation of leaf water content. The results showed that from scale 1 to scale 8, the correlation between wavelet coefficients and potato leaf water content first increased and then decreased, with medium decomposition scales being optimal for improving correlation (the absolute value of the correlation coefficient increased from 0.59 to 0.84). The empirical vegetation indices constructed from wavelet coefficients showed amplified correlation with water content at specific scales, and models built using optimal scale combinations of each index outperformed original spectral vegetation indices. For estimating leaf water content based on single\|scale wavelet coefficients, the XGBoost model established after CARS screening of scale 5 CWT spectra performed best (R2=0.88, RMSE=1.235%). Feature fusion from scale 5 and scale 7 further improved the estimation model performance (R2=0.90, RMSE=1.145%). In conclusion, monitoring leaf water content using multi\|scale features selected by CWT combined with the CARS method provides better estimation accuracy, offering an effective approach for precision water management in potato cultivation. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
| | |