Innovative approach for automatic land cover information extraction from LiDAR data

Nagwa Taha Hamdy El-Ashmawy

doi:10.32920/ryerson.14648820

Abstract

An airborne laser scanning (ALS) system with LiDAR (Light Detection and Ranging) technology is a highly precise and accurate 3D point data acquisition technique. LiDAR technology has been extensively used in digital surface/terrain modelling (DSM/DTM), and related applications such as 3D city modelling and building extraction. The capability of LiDAR systems to record the intensity of the return laser pulse backscattered energy in addition to the range data has motivated researchers to investigate the use of LiDAR intensity data for extracting land cover information. The main goal of this research is to maximize the benefits of the use of LiDAR data independently of any external source of data for automatically extracting accurate land cover information. Several new approaches are introduced in this research: a) classifying and filling the LiDAR intensity point cloud to produce a land cover image, b) combing multiple classified data of multiple LiDAR data-strips, c) statistical analysis segmentation technique that uses the concept of the kurtosis change curve algorithm for automatic classification of LiDAR data, and d) accelerating the classification process of large datasets by partitioning the large datasets into small, manageable datasets. Applying the traditional image classification techniques on LiDAR elevation and intensity data exclusively is included. Pixel-based, object-based, and point-based classification logics are conducted, and their results are compared to reference data. The results indicated that LiDAR data (range and intensity) can independently be used in land cover classification. By applying traditional pixel-based, supervised image classification techniques, the classification results show that auxiliary layers, which are extracted from range and intensity data, can be used for land cover classification. However, applying the supervised classification techniques on the LiDAR point cloud data without converting the data into images (Point-based logic) produced more accurate land cover classification results. The experiments on the proposed classification approach using the statistical analysis segmentation technique (based on the concept of the kurtosis change curve algorithm) show that it can be used to classify LiDAR data for land cover mapping.

Highlights

The concepts of data collection and information extraction have changed since the 1970s
Three auxiliary layers were extracted within the data preparation stage: the texture of the intensity, the normalized digital surface model Normalized DSM (NDSM), and the slope of the elevation models
The Normalized DSM (NDSM), the slope of the elevation layers, and the texture of the intensity were used as auxiliary layers that were derived from the light detection and ranging (LiDAR) elevation and intensity data

Summary

Introduction

The concepts of data collection and information extraction have changed since the 1970s. Releasing remote sensing data for civilian applications has challenged researchers to develop new techniques for image interpretation and information extraction. One of the techniques that have been used for information extraction is classification. Various approaches and algorithms have been developed for classification depending on data characteristics. Whenever a new technology for data acquisition becomes available, researchers investigate the suitability of using existing classification techniques with this data type, and/or develop new classification techniques that are more appropriate for the new data type. Several supervised and unsupervised classification algorithms have been developed to classify image data, especially for images captured by optical satellite sensors

Objectives

Methods

Results

Conclusion