Shoreline Based Feature Extraction and Optimal Feature Selection for Segmenting Airborne LiDAR Intensity Images

Modern airborne laser swath mapping (ALSM) systems measure both elevation and reflection intensity of the terrain. However, this intensity has been under utilized as a feature for image classification because it does not represent true terrain radiance. In areas with minimal topographic relief, such as beaches, we show that segmenting intensity images rather than elevation images has great potential for scene analysis. Several intensity-based features are extracted from ALSM data collected along a beach and partitioned into three classes to detect the water line. Class-conditional probability density functions are estimated for each feature to asses which are most informative. Results indicate significant class separation using centroidal features. Their classification performance is evaluated using a naive Bayes classifier and the area under receiver operating characteristic curves. The method presented provides a novel feature extraction and a systematic feature selection procedure for high-resolution ALSM intensity data.

1. INTRODUCTION

Airborne laser swath mapping (ALSM), often referred to as light detection and ranging (LiDAR), enables sub-meter spatial sampling of topography. An ALSM system pulses a near-infrared laser from an airborne platform to the ground several thousand times per second. By precisely determining the position and attitude of the aircraft as well as the angle at which each laser pulse leaves the aircraft, the direction of the laser pulse toward the ground can be accurately calculated. Combining this information with the recorded return time of the reflected pulse allows for a three-dimensional point sampling of the ground and landcover [1]. From the resulting ALSM point data, digital images are created in which the pixel value corresponds to the mean topographic elevation of the points within that pixel. These images often exhibit rms elevation errors of less than 10cm over minimally-vegetated surfaces, such as beaches. For most ALSM systems, the points also have associated intensity values, which can be similarly interpolated into an intensity image. To date, the intensity values have rarely been utilized by earth scientists, largely because they are only recorded as relative values. Thus, they are unitless and do not represent absolute surface radiance.