Eetu Puttonen, Teemu Hakala, Olli Nevalainen, Sanna Kaasalainen, Anssi Krooks, Mika Karjalainen, Kati Anttila
Optical Engineering, Vol. 54, Issue 01, 013105, (January 2015) https://doi.org/10.1117/1.OE.54.1.013105
TOPICS: Vegetation, Target detection, Clouds, LIDAR, Camouflage, RGB color model, Laser scanners, Hyperspectral target detection, Reflectivity, Optical engineering
Laser scanning systems that simultaneously measure multiple wavelength reflectances integrate the strengths of active spectral imaging and accurate range measuring. The Finnish Geodetic Institute hyperspectral lidar system is one of these. The system was tested in an outdoor experiment for detecting man-made targets from natural ones based on their spectral response. The targets were three camouflage nets with different structures and coloring. Their spectral responses were compared against those of a Silver birch (Betula pendula), Scots pine shoots (Pinus sylvestris L.), and a goat willow (Salix caprea). Responses from an aggregate clay block and a plastic chair were used as man-made comparison targets. The novelty component of the experiment was the 26-h-long measurement that covered both day and night times. The targets were classified with 80.9% overall accuracy in a dataset collected during dark. Reflectances of four wavelengths located around the 700 nm, the so-called red edge, were used as classification features. The addition of spatial aggregation within a 5-cm neighborhood improved the accuracy to 92.3%. Similar results were obtained using a set of four vegetation indices (78.9% and 91.0%, respectively). The temporal variation of vegetation classes was detected to differ from those in man-made classes.