Development of correction chain for multispectral airborne video camera data for natural recource assesment

Abstract

Land cover and its changes can be monitored using remote sensing data. This is traditionally conducted using satellite data, but with the recent advantages in computer and sensor technology, airborne remote sensing data have become a useful tool in projects dealing with natural resource inventories and environmental analysis. There are several kind of sensors constructed for acquiring airborne remote sensing data, however, but one of the most cost-effective sensors considered are video sensors.

To be able to use video data in such a projects, radiometric calibration and corrections for spatial and spectral distortions are evident, as distortions in image geometry and radiometry hamper the use of the raw data. When producing a correction chain involving corrections for sensor and system induced effects, atmosphere related effects, topographic effects and effects related to illumination, view angle and target, and when performing these corrections for the data, a fast and reliable tool for small-scale environmental monitoring and natural resource analysis is developed. When these corrections are performed for single and consecutive video frames before mosaicking procedures, the data can be used for large-scale monitoring applications.

In this study a correction chain is developed for digital video data acquired with an uncalibrated four-band video camera system constructed at the University of Munich, Institute for Geography. The technology of the video camera system was old-fashioned but the data was usable for the development of the correction chain and for natural resource assessment. The study area was alpine foreland between Munich and the German Alps.

In order to test the necessity of the environmental corrections overall, and the efficiency of the corrections developed, a digital land cover classification over mountainous test site Hirschberg was performed. The classification was performed for data without correction for bidirectional effects and without illumination compensation, with correction for bidirectional effects, with illumination compensation and with both corrections using same training areas and similar maximum likelihood classification algorithm. As the classification accuracy was used as a measure of the data quality and the usefulness of the correction method, a development of an accuracy assessment method was essential.

The accuracy for the classification with uncorrected data was 64.3 %, for the data with correction for bidirectional effects only 69.9 %, for the data with illumination compensation 62.0 % and for the data with both environmental corrections 85.8 %. The good accuracy with both environmental corrections can be argued firstly with the well developed corrections, and secondly with the small test site and the small number of classes.

Because the individual corrections reduce the distortions on the video data and the use of the whole correction chain increases classification accuracy significantly, the correction methods can be considered necessary and useful. The correction chain can also be applied to the data acquired using other optical airborne sensors such as digital cameras or imaging spectrometers with slight modifications, because specific sensor performance and geometry of each sensor have to be taken into account.