Study Of Remote Sensing Classification Based On Self Organizing MAP And Feature Extraction

In recent years,with the continuous development of earth observation technology,the spatial resolution of remote sensing satellite data has been improved to sub-meter level,and the spectral resolution has been improved to 5 nm.The rapid development of remote sensing technology has greatly promoted the application of remote sensing data.At the same time,the demand for remote sensing data processing methods is also increasing,more advanced data needs more advanced technical methods.In this context,some scholars have proposed some new algorithms,such as support vector machine,neural network and so on.However,most of the research directions are focused on supervised classification algorithms.For supervised remote sensing classification algorithms,the accuracy of supervised remote sensing classification algorithms is affected by training data sets with category markers.It often takes a lot of manpower and material resources to select a training data set which can well represent the research area,and different experimenters choose different training sets,the final classification accuracy of the classification algorithm is also different,and even may be very different.In this paper,the unsupervised classification is studied,and the following results are obtained:(1)Different unsupervised remote sensing classification algorithms are studied.The experimental results show that the classification accuracy of traditional unsupervised classification algorithms,such as ISODATA and K-means,is about 80%.For the self-organizing mapping(SOM)algorithm in neural network,because of the topological ranking mapping ability of SOM in different dimension space,its classification accuracy is significantly higher than that of the traditional unsupervised classification algorithm,and the classification accuracy is more than 85%.(2)Aiming at the pixel-based SOM algorithm,this paper proposes to use feature extraction to optimize SOM,so as to improve the classification accuracy and classification efficiency.The SOM algorithm for pixels inevitably produces salt and pepper noise in the classification results,and the feature extraction method can consider the influence of pixels around a pixel and the texture information of ground objects.The classification accuracy of the algorithm based on self-organizing mapping and feature extraction is 94%.In order to further verify the classification accuracy and efficiency of the proposed algorithm,this paper sets up some other comparison algorithms,namely: BP neural network,support vector machine,ISODATA,K-means and pixel-based SOM algorithm.The experimental results show that the classification accuracy of the proposed algorithm is higher than that of the contrast algorithm,and the classification time is less than that of the popular supervised classification algorithm.Once the process is established successfully,it can classify different areas of the same type of remote sensing data in order to achieve the purpose of rapid classification.