基于神经网络和TM图像的大连湾海域悬浮物质量浓度的反演

Retrieval of suspended sediment concentration in Dalian Bay on basis of neural network model and TM imagery

水体中的悬浮物是评价海洋水质的重要指标之一。采用卫星遥感方法可以获得大范围连续的悬浮物质量浓度。以1999年5月10日大连湾海上现场-卫星同步实验数据为基础,采用神经网络模型技术模拟了陆地卫星TM传感器中心波长分别为485,560和660nm3个波段的辐射亮度值与在该海域现场获取的悬浮物质量浓度之间的传递机理。以TM图像的3个可见光波段作为输入的神经网络模型的相关系数达0.79,在反演海水表层悬浮物质量浓度方面比传统的统计分析方法表现要好。这说明神经网络方法在模拟非线性关系进行遥感反演方面更具应用前景。

Suspended sediment concentration is important for assessment of water quality in oceans. Remote sensing can provide large-scale and continuous suspended sediment concentration information. On the basis of satellite and synchronous in situ measurement data, Neural network was constructed to model the transfer function between in situ suspended sediment concentration and the radiances of 3 TM bands whose center-wavelengths are 485nm, 560nm, and 660nm respectively. The correlation coefficient of neural network model with 3 visible-light bands as inputs is 0.79. It performs better than statistic models in retrieving surface suspended sediment concentration. It is shown that neural network is promising in modeling non-linear function in remote sensing retrieval.