基于EMD-SD光谱的玉米叶片叶绿素含量GA-BP模型反演

叶绿素是作物进行光合作用所需的主要色素，BP神经网络（BPNN）是较为新颖的反演叶绿素含量的方法。为研究反演精度更高的叶绿素含量反演模型，将经验模态分解（EMD）与光谱微分（SD）结合来提高输入因子与叶绿素含量的相关性，并使用遗传算法（GA）优化BPNN得到GA BP模型以获得最优初始权值阈值。将光谱数据EMD后进行一阶微分变换得到EMD SD光谱，选择与叶绿素含量相关系数超过06的5个波段处的EMD SD值作为GA BP模型的输入因子，隐含层节点数为7，多次训练取最优个体适应度值最低的GA BP模型来反演玉米叶片叶绿素含量。GA BP模型反演得到的预测值与实测值之间的判定系数（R2）最高，达到0818，均方根误差（RMSE）仅为2442，平均相对误差（e）为5436%。研究表明，EMD SD光谱作为GA BP模型的输入因子，与线性模型MLR和未优化的BP模型相比反演精度最高，验证了基于EMD SD光谱的GA BP模型提高玉米叶片叶绿素含量反演精度的可行性。

Study on GA BP inversing modeling method of corn leaf chlorophyll content based on EMD and spectral derivative method

The chlorophyll is the main pigment for the photosynthesis of crops. The BPNN is a novel method of inversing chlorophyll content. In order to study chlorophyll content inversion model with higher precision， it was used to increase the correlation between input factors and chlorophyll content by combining empirical mode decomposition （EMD） with spectral derivative （SD）. And genetic algorithm（GA） was used to optimize BPNN building GA BP model to get the best initial weights and thresholds. The spectral reflectance of corn leaf was pretreated by the methods of EMD and derivative， getting the EMD SD spectrum. It was selected as the input factors of GA BP model that the EMD SD values in five bands whose correlation coefficients with chlorophyll content were over 06. Then the GA BP model with seven hidden layer nodes was established， selecting the network whose fitness of the best individual was the lowest to predict the chlorophyll content of corn leaf. R2 of GA BP model was the highest， at 0818， RMSE was 2442 and e was 5436%. The results showed that the predicting precision of GA BP model using EMD SD values as input factors was higher than MLR and BP model. It was verified feasible that using GA BP model based on EMD SD spectrum to improve the inversion accuracy of corn leaf chlorophyll content.