Remote Sensing Extraction Method Of Color Steel Plate Information Based On Multi-feature Integration

Color steel plate buildings are mainly concentrated in urban villages,industrial parks and marginal areas within urban expansion,and the large gathering of color steel plate buildings will aggravate urban heat island and environmental pollution problems and make fire safety face greater challenges,and accurate extraction of color steel plate buildings information is the basis of relevant research and application.At present,remote sensing technology is one of the effective means to extract information of color steel plate buildings.Related researches have done a lot of research and analysis for remote sensing extraction of color steel plate buildings,however,due to different types and colors of color steel plate buildings are characterized differently on the image,which leads to the existing color steel plate buildings extraction model has weak generalization ability and accuracy is difficult to guarantee.Based on this,the thesis researches various color steel plate buildings extraction methods and constructs a new color steel plate buildings extraction model by collecting hyperspectral information of color steel plate buildings,taking into account the reflectance differences of different color steel plate buildings,and using GF-2 images and Zhuhai-1 hyperspectral images.The main work accomplished is as follows:(1)Spectral measurement and analysis of color steel plate buildingsThe paper has collected the hyperspectral information of blue,red and white colored color steel plate buildings with the use of the feature spectrometer,and analyzed the influence of the collection point location,the thickness of color steel coating and the orientation of color steel plate buildings on the reflectance.Based on GF-2 high-resolution images and Zhuhai-1hyperspectral(OHS)images,the spectral match between the spectral reflectance information of different color steel plate buildings and the measured spectral reflectance is studied,and the correction accuracy of the 6S model and FLAASH model on the reflectance of color steel plate buildings is compared and verified.The results show that the atmospherically corrected images of the 6S model can better reproduce the spectral information of white color steel plate buildings,and the atmospherically corrected images of the FLAASH model can better reproduce the spectral information of blue and red color steel plate buildings.(2)Multi-feature fusion based color steel plate buildings information extraction model constructionThe paper firstly fuses the panchromatic image of GF-2 with the OHS,and constructs different feature combination schemes based on the fused image.secondly,this paper uses PCA transform and Cfs Subset Eval evaluator-Best First search method for attribute selection(CBFS),and performs dimensionality reduction for each combination scheme,and finally performs Random Forest(RF)on the reduced results.The research results show that the CBFS/RF method has better integrity for large color steel plate buildings and can effectively improve the accuracy of color steel plate buildings extraction.(3)Construction of deep learning based model for extracting information of color steel plate buildingsBased on GF-2 satellite images,the paper expands the color steel plate buildings dataset for the problem of small number of samples in the existing color steel plate buildings deep learning dataset.The paper also selects Res Net50,U-net and Seg Net network models for the extraction of color steel plate buildings information and finds that both Res Net50 and U-net networks can extract the color steel plate buildings information more accurately,and the Res Net50 network has better effect on the extraction of fine color steel plate buildings.(4)Construction of color steel plate buildings information extraction model based on optimal integrationIn order to integrate the advantages of CBFS/RF method and deep learning,the paper selects voting method and scoring method to integrate CBFS/RF extraction results and deep learning extraction results,and constructs a new remote sensing extraction model for color steel plate buildings.The experimental results show that the extraction results of the integrated model improve the performance of each parameter compared with the pure objectoriented extraction and deep learning extraction results.In summary,both CBFS/RF based on multi-feature fusion and deep learning based on color steel plate buildings information extraction methods can extract more complete color steel plate buildings information,and the integrated model proposed on this basis can improve the extraction accuracy of color steel plate buildings information to a certain extent and provide data support for the subsequent related research.