فهرست نشریات علمی دسترسی آزاد ایران

Hyperspectral image (HSI) classification is one of the most important processes on these images, which artificial intelligence (AI) techniques have recently achieved significant success in this process. Data representation using a low-dimensional subspace is critical in classification task of HSI. By employing the Low-Rank Representation (LLR) approach, low-dimensional representations from data can be effectively extracted. Since this method neglects local information, the extracted features are not sufficiently rich and informative for classification. This paper proposes a machine learning method for hyperspectral image classification, which involves employing the Structure of the Data Regularized LLR with Dictionary Learning (SDLRRDL) model. Our AI-based model, SDLRRDL, presents an approach for learning data structures through a low-rank and sparse representation. Also, to leverage structural data information, a penalty is added to the low-rank representation model. The method can create similar features for data of the same class by combining image class signature and spectral-spatial information. Moreover, image samples are represented through a linear combination of dictionary atoms. Rich and informative features are extracted through the trained dictionary utilizing a training data set that better matches the training content. Then, extracted features are classified using the support vector machine with high accuracy. Simulation results demonstrate that the proposed method has a superior classification accuracy compared to state-of-the-art methods on three popular HSI datasets. The proposed method improves the classification accuracy of the state-of-the-art methods more than 2.13, 0.2, and 0.6 percent on the Indian Pines, Pavia University, and Salinas datasets, respectively.