The endmember extraction algorithm, which selects a collection of pure signature spectra for different materials, plays an important role in hyperspectral unmixing. In this paper, the endmember extraction algorithm is described as a combinatorial optimization problem and a novel Mutation Operator Accelerated Quantum-Behaved Particle Swarm Optimization (MOAQPSO) algorithm is proposed. The proposed approach employs Quantum-Behaved Particle Swarm Optimization (QPSO) to find endmembers with good performances. To the best of our knowledge, this is the first time that QPSO has been introduced into hyperspectral endmember extraction. In order to follow the law of particle movement, a high-dimensional particle definition is proposed. In addition, in order to avoid falling into a local optimum, a mutation operation is used to increase the population diversity. The proposed MOAQPSO algorithm was evaluated on both synthetic and real hyperspectral data sets. The experimental results indicated that the proposed method obtained better results than other state-of-the-art algorithms, including Vertex Component Analysis (VCA), N-FINDR, and Discrete Particle Swarm Optimization (D-PSO).

Hyperspectral image classification(HSIC) based on deep learning has always been a research hot spot in the field of remote sensing. However, most of classification models extract relevant features based on fixed-scales convolution kernels, which ignores the complex features of hyperspectral images(HSIs) at different scales, and impairs the classification accuracy. To solve this problem, a multiscale densely connected attention network(MSDAN) is proposed for HSIC. Firstly, the model adopts three different scales modules with dense connection to enhance classification performance, strengthen feature reuse, prevent over fitting and gradient disappearance. Besides, in order to reduce model parameters and strengthen the extraction of spatial-spectral features, the traditional three-dimensional convolution is replaced by three-dimensional spectral convolution block and three-dimensional spatial convolution block. Furthermore, the spectral-spatial-channel attention is embedded into the end of each scale to enhance the favorable features for classification and further extract the discriminant features of the corresponding scale. Finally, the key feature extraction module is developed to extract multiscale fusion features to further enhance the classification performance of the network. The experimental results carried out on real HSIs show that the proposed MSDAN architecture has significant advantages compared with other most advanced methods.

Soil salinization leads to dehydration of plants, which seriously threatens ecologically sustainable development and food security guarantee. In the complex and diverse coastal wetland environment, the impervious surface and bare soil have similar spectral features with salinized soil, which make it difficult for traditional satellite data and algorithms to accurately and timely monitor the small surface features of salinization. This article presents a baseline-based soil salinity index (BSSI) for soil salinization monitoring using medium-resolution data. In BSSI, we construct a virtual salinization baseline by connecting the near-infrared (NIR) band and the short-wave infrared-2 (SWIR2) band to enhance the spectral feature of salinized soils which border on the impervious surface. In addition, we calculate the distance between the short-wave infrared-1 (SWIR1) band and the virtual salinization baseline as the BSSI, which can effectively improve the stability of salinity inversion for different soils. Through data comparison and model simulations, BSSI has shown advantages over a series of the traditional salinization spectral indices (SSIs). The results show that the saline soil extraction accuracy of BSSI exceeds 85% and the correlation coefficient of the BSSI and the degree of soil salinization exceeds 0.90. Since the related spectral bands, such as NIR, SWIR1, and SWIR2, are available on many existing satellite sensors such as Landsat TM/ETM+, OLI, and sentinel 2, the BSSI concept can be extended to establish long-term records for soil salinization monitoring.

Salt marsh vegetation in the Yellow River Delta, including Phragmites australis (P. australis), Suaeda salsa (S. salsa), and Tamarix chinensis (T. chinensis), is essential for the stability of wetland ecosystems. In recent years, salt marsh vegetation has experienced severe degradation, which is primarily due to invasive species and human activities. Therefore, the accurate monitoring of the spatial distribution of these vegetation types is critical for the ecological protection and restoration of the Yellow River Delta. This study proposes a multi-source remote sensing data fusion method based on Sentinel-1 and Sentinel-2 imagery, integrating the temporal characteristics of optical and SAR (synthetic aperture radar) data for the classification mapping of salt marsh vegetation in the Yellow River Delta. Phenological and polarization features were extracted to capture vegetation characteristics. A random forest algorithm was then applied to evaluate the impact of different feature combinations on classification accuracy. Combining optical and SAR time-series data significantly enhanced classification accuracy, particularly in differentiating P. australis, S. salsa, and T. chinensis. The integration of phenological features, polarization ratio, and polarization difference achieved a classification accuracy of 93.51% with a Kappa coefficient of 0.917, outperforming the use of individual data sources.

Anomaly target detection has been a hotspot of hyperspectral imagery (HSI) processing in recent decades. One of the key research points in HSI anomaly detection is accurate descriptions of background and anomaly targets. Considering this point, we propose a novel anomaly target detector in this paper. Improving upon the low rank and sparse matrix decomposition (LRaSMD) approach, the proposed method assumes that the low rank component can be described as parts-based representation. Parts refer to various ground objects in HSI. A new update rule of low rank component and sparse component is proposed. The proposed approach can be divided into three main steps: 1) Further refining the low rank component in the LRaSMD model as parts-based representation. Then, the HSI is decomposed as three parts: the product of basis matrix and coefficient matrix, sparse matrix, and noise; 2) The basis vectors matrix, coefficient matrix and sparse matrix are solved by the new update rules; 3) Since the anomaly targets exist in the sparse matrix, the sparse component is thus employed to detect the anomaly targets. Experiments implemented for five datasets demonstrate that the proposed algorithm achieved a better performance than the traditional algorithms.