Remote Sensing Image Classification Based On Machine Learning

With the rapid development of aerospace technology and sensor technology,the data of remote sensing image is increasingly diversified,database is also on the rise and the surface features is complicated,how to classify effectively and precisely sensing image has been a important content to be researched.For the fast growth of artificial intelligence,classification algorithms in machine learning is gradually becoming a effective way to deal with remote sensing images.To improve the classification precision of remote sensing images,this paper proposed,on the basis of machine learning,a effective and simple classifier and a two-level classifier.The main part and innovations are as follows:(1)Introduces systematically research status of remote sensing image,briefly elaborates some supervised classification algorithms and unsupervised classification algorithms and some classifier in use;concludes briefly the research contents and primary process.(2)Aiming to the characters of remote sensing images easily influenced by air absorption and scattering,sensor calibration,terrain and so on,which lead to distortion of image,this paper exploited quadratic polynomial model to correct geometrically remote sensing image,adopted bilinear interpolation to resample to prerevise image and conducted atmospheric correction,which eliminated effectively distortion effect from sensor and the like.Meanwhile,the influence of scattering particle wad removed,laying foundation for subsequent classification.(3)Neural network model has such characters as fault toleration,strong ability to learn,but it will be very time-consuming to get satisfactory classification outcome.However,extreme learning machine model is kind of neural network model with simple construct,and can recognize,rapidly and effectively,samples.This paper constructed a method of remote sensing image classification based on hyper kernel function,this method used the global characters and local characters of hyper kernel function and combined with local region information of remote sensing image improving classification precision.(4)This paper proposed a kind of two-level classifier based on the spectral and spatial information.Combing with spectrum and space structure,this method firstly used spectral angle matching method as front classifier,which extracted ground with obvious spectral information character and great difference;then utilized the tensor construct information of remote data to get support tensor machine as second level classification.Classifying selected region of interest not only improve classification precision but visual effect is improved greatly.