The Feature Fusion Method Of Hyperspectral Imagery And LiDAR Data For Classification

As the best remote sensing data with potential for mining at present,hyperspectral imagery attracted wide attention and been applied in many fields such as agriculture,environment,ocean,resources,military and so on.Classification information of terrain features is an important basis for application in various fields.The quality of hyperspectral image classification results directly determines its role in these fields.Traditional hyperspectral image classification focuses on the mining of hyperspectral images themselves.Feature extraction,band selection and various fusion methods based on these indicate that the classification of hyperspectral images has made great achievements.However,hyperspectral image classification still faces many challenges.On the one hand,hyperspectral remote sensing sensors are susceptible to cloud influence,which makes the collected hyperspectral images produce shadow areas,which makes classification more difficult;on the other hand,limited training samples and high dimensions of images themselves make classification performance bottleneck.In order to solve these problems,it is necessary to introduce other remote sensing data to assist the classification of hyperspectral images.Based on the theory of remote sensing data fusion,this paper studies the fusion of hyperspectral images and LiDAR data to improve the accuracy of terrain classification,and proposes two different methods to achieve effective fusion and classification of heterogeneous data.Specifically,the feature fusion method based on multi-dimensional Gabor filter uses two-dimensional Gabor filter and three-dimensional Gabor filter to extract the texture features of LiDAR data and hyperspectral image respectively,and then classifies them under the guidance of superpixels;Based on the multi-modal feature fusion method,the extended multi-attribute profile and the histogram of oriented gradient are used to extract the morphological features and HOG features of LiDAR data and hyperspectral image respectively for decision fusion classification.Experiments show that the two kinds of feature fusion methods proposed in this paperplay a good role in the fusion of hyperspectral images and LiDAR data.Compared with several typical hyperspectral image classification methods,the MGabor-HL and MML-HL methods in this paper can effectively improve the accuracy of ground object recognition in the case of small samples.