Target detection by physical model observations in hyperspectral and LIDAR dataset

Hyperspectral target detection algorithms are the primary applications used in remote sensing. The various targets in the remote sensing images collected in hundreds of adjacent narrow spectral bands are found using target detection algorithms. In this thesis, target detection algorithms have been applied to fusion of LiDAR and hyperspectral dataset in which the shadow and open area targets are located.Target detection algorithms are applied to the radiance dataset, the reflectance dataset and the new reflectance dataset obtained from atmospheric correction and shadow correction in the physical model. Then, the target detection results are compared. Especially for the detection of targets located in the shadow area, sky view factor and shadow density values were determined over the LiDAR dataset of the studied area, and atmospheric transmittance, sun irradiance, sky radiance and path radiance values are obtained by MODTRAN software with the help of atmospheric data of dataset. The obtained data are put into physical model and atmospheric corrected and shadow corrected new reflectance dataset are obtained. Adaptive coherence estimator (ACE), spectral angle mapper (SAM) and matched filter (MF) are used in signature-based target detection algorithms. These algorithms are selected to analyze the effect of correction of hyperspectral data in the shadow area on target detection, in accordance with the target dataset. In the SHARE 2012 AVON campus dataset, twelve blue felt and ten red felt targets are located, eight blue and eight red felt targets have different shadow illumination levels. Signature-based algorithms, using open area target data as a reference signature are tested on the radiance dataset, the reflectance dataset and the new reflectance dataset obtained by atmospheric correction and shadow correction in the physical model. The results are analyzed over the areas under ROC curves and ROC curves. Target detection results are grouped in the results in order to see the effects of target illumination conditions, target background conditions and target colors on hyperspectral target detection. Fusion of LiDAR and hyperspectral dataset, the probability of detecting targets in the shadow and open area has increased from 70%-90% to 100%. In the applied algorithms, the ACE showed the best result, and the targets in the atmospheric and shadow corrected new reflectance dataset are detected with the lowest false positive rate according to the other reflectance and radiance dataset examined. In the atmospheric and shadow corrected new reflectance dataset in the physical model, SAM and MF algorithms showed higher performance than the target detection results in the given radiance and reflectance dataset and according to targets these performance has increased from 70% to 95% - 100%.