| Airborne hyperspectral remote sensing images have high spatial resolution and high spectral resolution.Through this technology,urban feature information can be obtained quickly,accurately and non-contactly.Studying the urban feature target recognition algorithm based on hyperspectral remote sensing images can not only improve a series of problems such as the inefficiency of traditional urban manual survey methods and errors caused by human subjective factors in manual surveys,but also improve the urban ecological environment and regulate urban management,Urban planning and construction to provide data foundation.In this regard,on the basis of investigating the current situation of hyperspectral remote sensing technology and urban applications,this paper first completes the pre-processing of hyperspectral remote sensing image data by performing normalization processing on selected data,PCA dimensionality reduction,training data set generation,and image slicing.Processing to provide a good data foundation for subsequent target recognition algorithms.This article studies the urban feature target recognition from the following aspects.First,the feature recognition technology based on spectral features uses Spectral Angle Mapping(SAM),Convolutional Neural Networks(CNN)and other algorithms to identify the spectral information of the features;secondly,The feature recognition technology based on spatial features uses the Speeded Up Robust Features(SURF)algorithm to recognize the spatial information of the features;finally,the feature recognition technology based on the combined features of the space spectrum combines the spectral and spatial features Object recognition.In this paper,overall accuracy(OA)and error rate(ER)are selected to evaluate the recognition results.Taking house identification as an example,the research results show that:The research results show that:1)In the research of spectral feature recognition,the CNN algorithm tried to use has higher overall accuracy(OA)and lower error rate(ER)compared with the traditional SAM algorithm.This result will be followed by machine-based Recognition of learned features and objects has a good beginning;2)In the study of spatial feature recognition,due to the irregularities of urban features such as houses,when only the spatial features of hyperspectral remote sensing images are used for target recognition,the result of artificial building target recognition will be incomplete,that is,in the process of during the house identification,the houses with a large difference from the appearance of the sample house will be identified.3)Target recognition based on the combination of space spectrum,by combining the recognition results of the spectral features and the spatial features,and removing the non-house information and other information obtained from the spectral feature recognition results while ensuring the overall accuracy(OA),In order to reduce the error rate(ER),to improve the effect of ground object recognition. |
