为进一步提高遥感图像水质反演的精度，提出了一种基于知识和遥感图像相结合的神经网络水质反演模型。该模型利用遥感图像数据以及与水质相关的知识数据作为BP神经网络的输入，经训练后，确定神经网络的结构，在训练好的BP神经网络基础之上对水质进行反演。以中国太湖为例进行实证研究，实验中，使用的知识数据包括太湖的地理信息知识和先对太湖TM图像上的水域解译进而对水质进行分类的知识。实验结果表明，本文提出的水质反演模型较常规的线性回归模型和传统的神经网络模型有更高的反演精度。

In order to improve water quality retrievals of remotely sensed image accurately, this paper puts forward a neural network model for water quality retrievals using knowledge and remotely sensed image. The model uses remotely sensed image data and water quality related knowledge as input of BP neural network, then trains neural network, after that water quality is retrieved by the trained neural network. The proposed model is applied to the water quality retrievals of Tai Lake in China. In experiment, knowledge used includes Tai Lake geography information knowledge and classification knowledge of water quality by interpretation of TM image. The result of experiment shows that the developed model has more accuracy than the routine linear regression model and traditional neural network model.