Research On High Recognition Compound Derivative Features Of Landcover Accurate Classification Method Based On LiDAR Data

The airborne Li DAR(Light Detection And Ranging,laser scanning ranging)system adopts a non-contact active measurement method,which can quickly obtain spatial information of ground objects,and is an important source of feature data in the field of remote sensing ground object classification.With the continuous expansion of the application field of airborne Li DAR,complex changes in the material and structure of the ground features,changing imaging characteristics,changes in the imaging environment with climate and seasons,and the limitations of the Li DAR system in collecting data,the original feature space is predetermined.It is difficult to meet the high-precision classification requirements in a complex environment with its classification features,which severely restricts the realization of tasks such as threedimensional precise positioning,urban planning,and precise detection of objects.This paper conducts in-depth research and discussion on how to combine the influencing factors that restrict the classification accuracy to construct a new high-identification compound derivative feature that is beneficial to the improvement of the classification accuracy.The main contents are as follows:(1)The construction of the feature space set of Li DAR system data and the analysis of the feature’s recognition performance: extract complementary Li DAR point cloud spatial features and remote sensing image spectrum features to construct a feature space set;study the characteristics of each feature in the Li DAR feature space Obtain methods and physical meanings,analyze and compare the classification and identification capabilities of roads,buildings,trees,grass and other typical features of each feature;describe the physical characteristics of the features from the two perspectives of the similarity and complexity of the features themselves and the original features Analyze the apparent characteristics of different ground objects in Li DAR data sources,and study the factors that restrict the accuracy of ground object classification.(2)Construction of adaptive derived feature function based on XGBoost algorithm: use XGBoost algorithm to calculate attribute importance in a single decision tree by improving the performance measurement of each attribute split point,and extract feature importance for a single feature;statistics The importance of each feature variable is adaptively selected to participate in the effective feature classification;in the transfer learning communication signal transmission,the non-uniform quantization method of the analog sampling signal uses the logarithmic characteristic of A-law compression to perform the selected feature Coordinated combination;constructing an adaptive derivative feature function for four typical features(buildings,trees,grass,roads),improving the feature recognition ability of features,and realizing accurate classification of features.(3)Accurate classification of ground objects based on the characteristics of G-RNDVI(Composite Normalized Difference Vegetation Index): Using the difference in reflectivity of artificial structures such as buildings,roads,and vegetation on the green band spectrum and red band spectrum,the NDVI(Normalized Difference Vegetation Index)and GNDVI(Green Normalized Difference Vegetation Index)are weighted by two different vegetation indexes to construct a composite normalized difference vegetation index feature with high identification ability;the study varies with the weighting coefficient,The effectiveness of the newly constructed composite derived features;combined with the fuzzy DS evidence synthesis theory classification algorithm,it can solve the problem of the reduction of classification accuracy due to the limitations of the original features in complex environment scenarios,especially for the existing confusion areas,and achieve the accuracy of Li DAR data Classification of features.