Research On Building Shadow Classification Method Based On Height Inversion

With the rapid development of digital city construction,it is particularly important to obtain building height information quickly and efficiently.Although the current domestic urban building height information is relatively complete,there is still a lack of information on the height of large regional building groups unknown abroad,and the problem of monitoring the change of clearance height around the airport based on high-score images still needs to be resolved.It is one of the important ways to get the height information of buildings by building shadow inversion.However,it is not possible to obtain the height information of all buildings only based on the geometric relationship of the sun,satellites,buildings and their shadows.In summary,in view of the problem that different types of building shadows require different methods for height inversion,this paper first extracts building shadows through UNet,and then proposes a shadow denoising method based on building polygons.Finally,according to certain rules,the noise-removed shadows of buildings are classified and tested with the data sets in Beijing and some parts of Xinjiang.The main contents are as follows:(1)Establish a small sample size data set for building shadow extraction,and BN and residual connections are added to the model to alleviate the over-fitting phenomenon of the UNet model and optimize the performance of the model.(2)According to the spatial position relationship of the sun,satellites,buildings and their shadows,the buildings and shadows under simple terrain and complex terrain are simulated and analyzed,and the judgment formulas for their spatial relationships are derived in detail.At the same time,the LCZ classification ideas are used to analyze them.It is applied to the classification of building groups and verified by examples.(3)A shadow denoising method with building polygons as additional conditions is proposed.This method solves the incomplete and inaccurate problem of traditional shadow denoising.It can not only remove the building shadow noise extracted by deep learning well,but also You can match the building with its shadow one by one.(4)Based on whether the building height can be retrieved by the fishing net method as the starting point,the building shadows are divided into?building shadows too small,?building shadows vertical adhesion,and ? buildings according to classification indicators such as area threshold,rectangularity,and boundary index.The shadows are projected to adjacent buildings,and the four types of building shadows applicable to the fishing net method are four types,which lay the foundation for more accurate building height information.The research shows that the accuracy of U-Net network extracted from buildings in Beijing and Xinjiang is 93.18% and 96.24%,respectively.The shadow classification method in this paper is applied to the shadow classification of buildings in Beijing and Xinjiang.The experimental results have an overall kappa coefficient of 0.7390 and a Macro?F1 value of0.7201,which shows the effectiveness of this method to a certain extent.Continuing to study5 experimental areas in Beijing and Xinjiang,the proportions of building heights that cannot be retrieved by the fishing net method are 36.59%,29.17%,23.53%,25.93% and 16.67%,respectively.This article provides a basis for calculating the height information of different types of building shadows using different methods.