Land Cover Classification Method Based On High-resolution Imagery

With the rapid development of remote sensing sensor technology,the spatial resolution of remote sensing images acquired by humans has been continuously improved.High spatial resolution remote sensing images have been widely used in urban planning,land,surveying and mapping,geology,agriculture,forestry and environmental industries.The automatic classification of remote sensing images by computers automatically recognizes and extracts the information of ground objects,which is the basis for the application of remote sensing image information.Because high spatial resolution images have serious "homologous" phenomena,traditional image classification methods based on spectral features encounter difficult difficulties.For this reason,people introduce object-oriented thinking to classify high spatial resolution images.The technology has achieved good results,it has been adopted by mainstream commercial remote sensing software,such as ENVI,ERDAS and so on.Among them,the object-oriented classification technology based on SVM(Support Vector Machine)is a representative of this type of method.The main idea is to divide the image into a series of internal uniforms according to the principle of "maximal homogeneity and minimum heterogeneity" based on the image spectral elements firstly.Then extract the spectrum of the object,texture and geometric features,and finally use the SVM classifier to classify the features.Although the classification result of this method is excellent,it has the following disadvantages:(1)The texture information used in the classification is only single-scale information,ignoring the multi-scale characteristics of the texture information in space;(2)The classifier used The parameter is empirical and lacks optimality.Based on the above understanding of this article to start research,the specific results are as follows:(1)Applying panchromatic and multispectral image fusion can provide high spatial resolution multispectral images for image classification,which can effectively improve data quality and improve classification accuracy.Since the remote sensing image fusion method has different performance for different fusion purposes and data sources,in order to select the optimal image fusion method based on classification purposes,this paper selects IHS,PCA,Wavelet fusion,Gram-Schmidt fusion,IHS+Wavelet transform fusion and PCA+Wavelet transform fusion,integrate the image data of ZY3 and GF2,and evaluate the overall brightness range,sharpness,information,spectral correlation,and object-oriented classification accuracy of the fused image.Finally,the PCA fusion image is suitable for ZY3 image classification.For GF2 images,the Wavelet fusion algorithm has the best fusion effect.(2)In order to make up for the defect of the object-oriented classification method based on SVM,this paper improves it from two aspects.One is to use Log-Gabor wavelet to extract the multi-scale texture information of the image,and use PCA to use it.The information extracted by the object-oriented method is used for information integration;the second is the use of Particle Swarm Optimization(PSO)to optimize the SVM classifier parameters.Through visual classification of ZY3 and GF2 fusion images,visual comparison and objective statistical analysis show that the proposed method is better than SVM classification methods based on pixel and spectral features,object-oriented SVM classification methods and Log-Gabor wavelet textures.The SVM classification method has improved its classification accuracy to varying degrees.