Remote Sensing Image Classification Processing Based On Neural Network And Contourlet

As the application of remote sensing images widely used,more and more efficient and automated classification methods are exploited.The traditional classification method depend on statistical characteristics of image values,threshould or trained sample data to identify and classify. But for complicated Landmarks,it's difficult for similar spectral characteristics classifying efficiently and accurately,it will make great error and mistake. With the development of artificial neural networks,neural network classification techniques is widely used in the sensing image classification field.Neural network classification can solve the shortage of the traditional classification methods,the method has better self-learning ability and adaptive ability,more accurate and also improved the accuracy of the classification.This thesis firstly introduces the background and significance of the research,status and trends at home and abroad.Then introduces some basic knowledge of classification and research,Remote sensing images used by the US Landsat 5 multispectral image, Carried out in the ENVI and Matlab simulation,making a combination of the image processing and image classification.Do remote sensing image processing using the same image in different ways,and then make image classification.Make confusion matrix analysis and statistical analysis of the progress to all these processing methods,then make a conclusion.The study showed that the overall accuracy based on Contourlet and BP neural network classification up to 93.046%,which Kappa coefficient was 0.8874.Artificial neural network classification accuracy is better than traditional supervised classification and unsupervised classification accuracy.The final conclusion show that based on contourlet and neural network classification method is the best in this remote sensing region,and it is a feasible method.