Spectral-spatial information extraction and classification of mangrove species using joint sparse representation

Classification of mangrove species is very important for monitoring and protecting the coastal ecosystem. In this paper, we present a new spectral-spatial classifier that uses multi-spectral image captured by the ZY-3 satellite to distinguish seven mangrove species in the Beihai ecological monitoring area, Guangxi, China. In order to extract the spatial information, a correlative filter is designed to incorporate neighborhood correlative information before classification. Moreover, a feature optimization algorithm based on dictionary learning is applied to reduce the noise and improve the discrimination of sample features. Finally, a classification method using joint sparse representation is proposed to extract the mangrove region and recognize seven mangrove species. The classification results show that the major species in the study area are Aegiceras corniculatum and Avicenna marina that conform to field investigations. The overall accuracy reaches 95.62% and the kappa coefficient achieves the value of 0.9380. Hence, the accuracy and efficiency of our proposed method are demonstrated in mangrove species classification.