Evaluating UAS hyperspatial RGB imagery and textures for identifying beach areas along the south Texas Gulf Coast

Lihong Su, Harte Research Institute for Gulf of Mexico Studies at TAMU-CC, Corpus Christi, TX, United States and James C Gibeaut, Texas A & M University- Corpus Christi, Harte Research Institute for Gulf of Mexico Studies, Corpus Christi, TX, United States
Abstract:
Shoreline information is fundamental to extensive coastal researches. For practical purposes, the analysis of shoreline variability usually uses a group of shoreline indicators visibly discernible in coastal imagery, such as seaward dune vegetation line, wet/dry maxima line, and instantaneous water line. These indicators partition a beach into four areas from fore dune to sea: vegetation, dry sand or debris, wet sand and water. Conversely these shoreline indicators can be recognized when the four beach areas are identified. UAS remote sensing can acquire ultra-high spatial resolution imagery such as sub-decimeter pixel size, namely hyperspatial imagery, for coastal researches. UAS remote sensing provides a great opportunity to identify the four beach areas.

UAS imagery typically has low spectral resolution such as only three wide visible bands, namely red, green, and blue. Spatial features such as textures should play an important role in delineating these beach areas. Among numerous texture measures, the gray level co-occurrence matrices (GLCM) (Haralick et al. 1973) is the most used texture on remote sensing classification (Laliberte and Rango, 2009). The local binary pattern (LBP) is a relatively new texture measure (Ojala et al. 2002). It already obtained great success in computer vision and pattern recognition (Nanni et al. 2012). However, LBP use in remote sensing classification is relatively rare. The objectives of this investigation are to evaluate UAS hyperspatial imagery and texture measures and unsupervised classification techniques for identifying the beach areas along the south Texas Gulf coast.

The method consists of 4 steps: (1) color space transform from RGB to USGS Munsell HSV, and separation of land and sea by unsupervised classifications with hue and volume bands; (2) computation of GLCM and LBP textures on various window sizes and pixel sizes; (3) unsupervised classification with these texture factors; and (4) accuracy assessment.

Experiments were conducted with South Padre Island photos acquired by a Nikon D80 camera mounted on the US-16 UAV during March 2014. The preliminary results show that classification accuracy can reach 89.7% with LBP textures by green band. LBP textures perform better than GLCM textures. The experiments also demonstrate that the method works well.