基于作物多光谱荧光特征的冠层叶绿素含量原位检测系统设计

Design of In-situ Canopy Chlorophyll Content Detection System Based on Crop Multispectral Fluorescence Characteristics

摘要:

为了满足田间作物营养快速检测的需求，通过对叶部多光谱荧光特征筛选设计了冠层叶绿素含量原位检测系统。基于封闭试验采用 Relief、RFE 及逐步回归 3 种方法筛选出 UV690、B740、UV690/B740核心变量，明确了叶部叶绿素荧光探测特征参数，为设备光学通道设计提供依据;优化封闭式探测为半开放式用于冠层原位测量，硬件包括 Raspberry Pi 4B主控与双目宽动态摄像头模组，集成均匀激发光源同时开发双目协同采集软件。田间原位试验结果表明，UV520、UV690、B740、UV690/B740为冠层尺度叶绿素探测的敏感性参数，其中UV520荧光可反映蓝绿荧光区对植物胁迫及环境差异的响应特征，与红光荧光参数形成互补，有助于提升冠层尺度叶绿素含量检测模型的稳定性与抗干扰能力。构建支持向量机（SVM）叶绿素含量回归模型，训练集R2C达0.82，验证集R2V达0.72，RMSE为2.86 mg/L，平均相对误差MRE为8.8%。研究结果实现了从叶片测量原型"到"冠层群体检测"应用，为小麦叶绿素含量的实时无损监测提供实用工具。"

Abstract:

Aiming to meet the demand for rapid detection of crop nutrients in the field, systematic research was conducted on the screening of leaf multispectral fluorescence characteristics and the development of an in-situ canopy chlorophyll detection system.Firstly, based on controlled experiments, three methods, including Relief, recursive feature elimination (RFE) and stepwise regression were adopted to screen out core variables such as UV690, B740 and UV690/B740.The characteristic parameters for leaf chlorophyll fluorescence detection were clarified, which provided a basis for the optical channel design of the device.Secondly, the closed detection mode was optimized to a semi-open mode for in-situ canopy measurement.The hardware system was composed of a Raspberry Pi 4B main controller and a binocular wide dynamic camera module, with an integrated uniform excitation light source and independently developed binocular cooperative acquisition software.Field experiments demonstrated that UV520, UV690, B740, and UV690/B740 were sensitive fluorescence parameters for chlorophyll detection at the canopy scale.Among them, the UV520 fluorescence reflected the response characteristics of blue-green fluorescence to plant stress and environmental variations, which complemented red fluorescence parameters and improved the robustness of canopy chlorophyll detection models.A support vector machine (SVM) regression model for chlorophyll content was constructed.The model achieved a coefficient of determination (R2) of 0.82 for the calibration set and 0.72 for the validation set, with RMSEC and RMSEV values of 3.42 mg/L and 2.86 mg/L, respectively, and a mean relative error (MRE) of 8.8%.This research realized the application transformation from the "leaf measurement prototype" to "canopy population detection", providing a practical tool for real-time and non-destructive monitoring of wheat chlorophyll content.