Application Of Clustering Algorithm Based On Wavelet Analysis In Remote Sensing Image Classification

The key point is the investigation and statistics of the area of cultivated land.Based on the high spatial resolution remote sensing image taken by the high-resolution one(GF-1)remote sensing satellite,this paper focuses on the investigation and monitoring of the global ecological environment.High-resolution remote sensing image classification method,algorithm and its improvement,to improve the classification of different types of land in the cultivated area to achieve the accuracy of the remote sensing image pixel processing and classification,and then intelligent extraction,classification and identification of geomorphological features,And provide the basis for estimating the coverage area of each cultivated land.First of all,clustering analysis technology,ISODATA algorithm,K-means algorithm as a representative of the unsupervised classification algorithm,because of its unique classification advantages and wide application,become preferred;but because of the initial clustering center randomly selected,easily lead to The results of the classification are different,and the two algorithms only use the spectral characteristics of the image to classify the image,and the waste of the rich information resources of the remote sensing image is relatively low.Therefore,an improved algorithm based on spectrum-based wavelet analysis is proposed.Two-dimensional wavelet transform is applied to the sample image,and the characteristics of the sample image of different types of objects are enhanced to extract the texture energy characteristics of the samples.As a K-means algorithm Initial clustering center,and then K-means algorithm classification.Secondly,four kinds of classification algorithms,such as ISODATA algorithm,K-means algorithm,maximum likelihood method and K-means improved algorithm,which is characterized by texture feature,are used to classify the seven types of objects in the study area.Finally,the classification accuracy of the four algorithms is evaluated objectively by the classification accuracy evaluation index confusion matrix and Kappa coefficient.The experimental results show that:(1)In the improved algorithm,the two-dimensional discrete wavelet transform of the original image is weakened by the noise caused by the soil brightness,and the difference of the edge of the different features is obtained.(2)The K-means optimization algorithm,which is supplemented by the texture feature,avoids the problem of randomly selecting the initial value,and realizes the fusion of the unsupervised algorithm and the supervised algorithm.(3)The overall accuracy of the classification is 77.67%of the ISODATA algorithm,73.94%of the K-means algorithm,and 82.92%of the maximum likelihood method,which is 92.22%.In order to obtain better classification effect,we should consider the image attribute and the target of interest when applying and improving the classification of the image to be classified according to the different attributes.Difficulty and complexity,computational efficiency and impact factors,as well as application background and needs.