基于红边特性和连续小波变换的冬小麦叶绿素含量估算

为充分利用高光谱数据红边区域对冬小麦叶绿素含量进行估算，以关中地区冬小麦为研究对象，基于红边波段反射率的一阶导数进行连续小波变换，对变换后得到的小波系数与叶绿素含量进行相关性分析，选取相关性较好的小波系数分别结合偏最小二乘法（PLS）、BP神经网络（BPNN）算法、随机森林（RF）算法和XGBoost算法构建冬小麦叶绿素含量估算模型。结果表明：（1）通过对建模数据和验证数据的决定系数（R）、均方根误差（RMSE）和相对分析误差（relative predictive derivation，RPD）进行比较，利用XGBoost算法构建的估算模型表现最好；（2）通过XGBoost算法的特征重要性分析得到13个有效小波系数，将其与7个红边指数共同作为自变量代入XGBoost算法发现，优化后的模型精度得到显著提高，建模集决定系数（R=0.91）和验证集决定系数（R=0.802）分别提高了1.34%和11.54%。这说明该方法可以作为一种挖掘高光谱敏感特征信息的途径来估算冬小麦叶绿素含量。

Estimation of Chlorophyll Content in Winter Wheat Based on Red Edge Characteristics and Continuous Wavelet Transform

In order to make full use of hyperspectral data to estimate the chlorophyll content of winter wheat in the red edge region and monitor the growth status of winter wheat,taking winter wheat in Guanzhong area as the research object,continuous wavelet transform was carried out based on the first derivative of reflectance in the red edge band,and the correlation between the transformed wavelet coefficient and chlorophyll content was analyzed. The wavelet coefficients with good correlation were selected,and the estimation model of chlorophyll content in winter wheat was constructed by combining Partial Peast Square method(PLS),Back Propagation neural network(BPNN) algorithm,Random Forest(RF) algorithm and Extreme Gradient Boost(XGBoost) algorithm. The results showed that:(1)By comparing the coefficient of determination(R),root mean square error(RMSE) and relative predictive derivation(RPD) of the modeling data and the validation data,the estimation model constructed by XGBoost algorithm showed outstanding performance.(2)The 13 effective wavelet coefficients obtained by the feature importance analysis of the XGBoost algorithm and the 7 red-edge exponents meeting the significance test of 0.01 level were substituted into the XGBoost algorithm as independent variables,and it was found that the accuracy of the model optimized by the wavelet coefficients is improved remarkably. The determination coefficients of modeling set(R=0.91) and verification set(R²=0.802) were increased by 1.34% and 11.54%,respectively. The results indicate that this modeling method can be used as a way to mine hyperspectral sensitive feature information to estimate the chlorophyll content of winter wheat.