基于神经网络的SPOT数据模拟真彩色非线性方法研究

THE NONLINEAR SIMULATION MODEL FOR SPOT TRUE COLOR IMAGE BASED ON NEURAL NETWORKS

在总结现有真彩色模拟方法的基础上,提出了一种基于神经网络非线性逼近机制的真彩色模拟方法。该方法充分利用神经网络强大的学习能力,建立波段之间复杂的非线性映射关系。在具体实现上,首先通过对具有真彩色波段TM/ETM+数据的学习训练,得到各个网络节点的权重参数;然后将该组权重参数应用到SPOT影像,模拟出蓝光波段,实现SPOT真彩色合成。在新疆伊犁地区林地调查应用结果表明,非线性方法要明显优于传统的真彩色模拟方法。

This paper sums up the existent methods of SPOT true color image simulation and puts forward a new simulation method based on neural networks,which uses the powerful function of the neural networks in learning from samples.Firstly,the model studies TM/ETM+ data.Secondly,after repeated training,the model obtains the appropriate model parameters between the TM B(blue) band and other bands such as NIR(near infrared),R(red) and G(green) band.Thirdly,the model is applied to SPOT images.According to the NIR(near infrared),R(red)and G(green) bands of SPOT,the B(blue) band is simulated.The final step is the synthesis of the SPOT true color image.Test result shows that the method is superior to the traditional methods in simulation of SPOT true color image.