Regarding to the limitations and benefits of remote sensing sensors, fusion of remote sensing data from multiple sensors is effective at land cover classification. All these data have different characteristics, e.g., different spatial and spectral resolutions, different angle of view, and different abilities and disabilities. For many applications, the information provided by individual sensors is incomplete, inconsistent, or imprecise. Fusion of information from different sensors can produce a better understanding of the observed site, which is not possible with single sensor. Particularly, Light Detection And Ranging (LiDAR) provides accurate height information for objects on the earth, which makes LiDAR become more and more popular in terrain and land surveying. On the other hand, hyperspectral imaging is a relatively new technique in remote sensing that acquires hundreds of images corresponding to different spectral channels. The rich spectral information of HS data increases the capability to distinguish different physical materials, leading to the potential of a more accurate image classification. As hyperspectral and LIDAR data provide complementary information (spectral reflectance, and vertical structure, respectively), a promising and challenging approach is to fuse these data in the information extraction procedure.
This paper presents a multiple fuzzy classifier system (Multiple Classifier System or MCS) for fusions of hyperspectral and LiDAR data based on Decision Template (DT). After feature extraction on each data, the classification was performed by fuzzy K-Nearest Neighbor (KNN) on hyperspectral and LiDAR data separately. In a multiple fuzzy decision system, a set of decisions is first produced and then combined by a specific fusion method. The output of the fuzzy classifiers that provide the class belongingness of an input pattern to different classes is arranged in a matrix form defined as decision profile (DP) matrix. Then, a fuzzy decision fusion method (Decision Tempate) is utilized to fuse the results of fuzzy KNNs on hyperspectral and LiDAR data. In order to assess the fuzzy MCS proposed method, a crisp MCS based on (Support Vector Machine) SVM as crisp classifier and Naive Bayes (NB) as crisp classifier fusion method is applied on hyperspectral and LiDAR data. The experiments were executed on a hyperspectral image and a LiDAR derived Digital Surface Model (DSM); both with spatial resolution of 2.5 m. The dataset have captured over the University of Houston campus and the neighbouring urban area by the NSF-funded Centre for Airborne Laser Mapping (NCALM). Also hyperspectral image has 144 spectral bands in 380 nm to 1050 nm region. Training and testing samples were selected from different areas of the images. They are spatially disjointed.
Fuzzy MCS on hyperspectral and LiDAR data provide interesting conclusions on the effectiveness and potentialities of the joint use of these two data. Overall accuracies of fuzzy classifiers on LiDAR and hyperspectral data are %75 and %88 respectively. Fusion of these two fuzzy classifiers produced %96 as overall accuracy. Second scenario for joint use of hyperspectral and LiDAR data is fusion of these two data through a crisp decision fusion system. The results show that fuzzy classifier provided higher accuracies than crisp classification based on SVM for both data. In the presence of  mixed  coverage  pixels  in  remote  sensing  data, crisp  classifiers  may  produce  errors  while  fuzzy  classifiers  are not  affected  by  such  errors  and  in  principle  can  produce  a  classification  that  is  more  accurate  than  any  crisp  classifier.  Also, fusion of ensemble of fuzzy classifiers based on Decision Template method produced more accuracy than fusion of crisp SVMs based on Bayesian Theory.