水体透射光谱的多特征融合COD含量估算研究

化学需氧量（chemical oxygen demand，COD）是一项可以快速检测有机污染物的参数，能够很好地反映水污染的程度。提出一种基于透射光谱测量的多特征融合水体COD含量估算模型，透射高光谱法采集100组COD水体光谱信息，对光谱数据进行预处理以及特征波段的选取，分析不同预处理方法对模型精度的影响并进行特征融合，建立BP神经网络模型，通过比较模型的精度选择最优模型进行水体COD含量的检测。结果显示，基于多特征融合BP神经网络模型决定系数R2高达0.991 64，均方根误差RMSE为0.030 9，与偏最小二乘法相比，该模型拟合优度更大，精确度更高。基于多特征融合的BP神经网络高光谱检测方法能够实现水体中COD含量的检测，并运用到水体其他成分的检测中。

COD content estimation of multi-feature fusion based on water transmitted spectrum

The chemical oxygen demand (COD) is a parameter that can quickly detect the organic pollutants and can well reflect the degree of water pollution. A COD content estimation model of multi-feature fusion water based on transmitted spectrum measurement was proposed. The transmission hyperspectral method collected 100 groups of COD water spectral information, preprocessed the spectral data, selected the characteristic band, analyzed the influence of different preprocessing methods on the model accuracy, carried out the feature fusion, and established the BP neural network model. By comparing the accuracy of the model, the optimal model was selected to detect the water COD content. The results show that the decision coefficient R2 of BP neural network model based on multi-feature fusion is 0.991 64, and the root-mean-square error (RMSE) is 0.030 9. Compared with the partial least square method, the model has higher goodness of fit and higher accuracy.The multi-feature fusion based BP neural network hyperspectral detection method can realize the detection of COD content in water, and can be applied to the detection of other components in water.