1. School of Information and Software Engineering, East China Jiaotong University, Nanchang 330013, China
2. Key Laboratory of Poyang Lake Wetland and Watershed Research (Jiangxi Normal University), Ministry of Education, Nanchang 330022, China
Abstract：Hyperspectral data collects essential and detailed spectral responses from ground objects through hundreds of contiguous narrow spectral wavelength bands and is widely used for vegetation fine classification. However, classification accuracy is not often satisfactory in a cost-effective way when using all original hyperspectral information (HSI) for practical applications because of its strong correlation and redundantness. Therefore, feature wavelength/band selection is crucial and difficult for HSI applications. Previous band selection methods have some drawbacks, such as low computation efficiency, lack of interpretability, being trapped in local optimization, and so on. Our study focuses on the hyperspectral feature band selection for the vegetation species fine classification of Poyang Lake wetland in continuous extreme drought conditions. Hyperspectral reflectance data of 10 plant species, such as Green polygonum, Artemisia Selengensis, Astragalus sinicus, Rorippa globose, Rumex trisetifer Stokes, Sonoma alopecurus, Phalaris arundinacea, Carexcinerascens, Miscanthus sacchariflorus and Phragmites australis collected by SVC spectrometer (SVC HR1024) is used in this work. We introduce the Quantum Genetic Algorithm (QGA), which is combined with Spectral Angle Mapper-based k-Nearest Neighbors classifier (KNN-SAM), and propose a new feature band selection algorithm, i.e., QGA-KNN-SAM, to select feature wavelength. Then, we use the K-Medoide clustering algorithm to determine the feature band interval. In our experiment, the classification performance of the proposed QGA-KNN-SAM is compared with the traditional GA-KNN-SAM algorithm. QGA-KNN-SAM generates an average classification accuracy value of 95%, higher than GA-KNN-SAM (90%). Moreover, QGA-KNN-SAM generates the feature bands range between 589~634.4 nm, which is relatively more concentrated than achieved by GA-KNN-SAM (1 107.6~1 205 nm). A wavelength band that reflects the surface hydrological characteristics and vegetation should be considered in the fine classification of wetland vegetation, which is different from the fine classification of traditional vegetation. Compared with the band distribution of commonly used multispectral and hyperspectral satellite images, it is found that the QGA-KNN-SAM algorithm selects feature bands with better directionality and interpretability. This algorithm improves the computational efficiency and interpretability of band selection and compensates for the lack of the QGA method in band selection research, providing methodological and theoretical support for similar studies.
[1] Song G, Wang Q. Ecological Informatics, 2023, 76: 102141.
[2] Vaiphasa C, Skidmore A K, Boer W F，et al. ISPRS Journal of Photogrammetry and Remote Sensing, 2007, 62(3): 225.
[3] Ullah S, Shakir M, Iqbal M S, et al. Ecological Informatics, 2021, 64: 101362.
[4] Varade D, Maurya A K, Dikshit O. International Journal of Remote Sensing, 2019, 40(20): 8031.
[5] Guo B F, Gunn S R, Damper R I, et al. IEEE Geoscience and Remote Sensing Letters, 2006, 3(4): 522.
[6] Li H C, Chang C I, Wang L, et al. Constrained Multipler Band Selection for Hyperspectral Imagery, IEEE International Geoscience and Remote Sensing Symposium, IEEE, 2016.
[7] Yarra C, Nagesh S, Deshmukh O D, et al. Journal of the Acoustical Society of America, 2019, 146(3): 1615.
[8] Yuan Y, Lin J, Wang Q. IEEE Transactions on Geoscience and Remote Sensing, 2016, 54(3)：1431.
[9] Cao X, Wu B, Tao D, et al. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2016, 9(9): 4352.
[10] Li S, Qiu J, Yang X, et al. Engineering Applications of Artificial Intelligence, 2014, 27: 241.
[11] Feng J, Chen J, Sun Q, et al. IEEE Transactions on Cybernetics, 2020, 51(9): 4414.
[12] He K, Sun W, Yang G, et al. IEEE Transactions on Geoscience and Remote Sensing, 2022, 60: 3169018.
[13] Wang J, Tang C, Liu X, et al. IEEE Transactions on Cybernetics, 2023, 53(8): 5250.
[14] He C, Zhang Y, Gong D, et al. IEEE Transactions on Evolutionary Computation, 2022, 26(6): 1566.
[15] ZHANG Liu, YE Nan, MA Ling-ling, et al（张 刘，叶 楠，马灵玲，等）. Spectroscopy and Spectral Analysis (光谱学与光谱分析), 2021, 41(10)：3194.
[16] Han K H, Kim J H. IEEE Transactions on Evolutionary Computation, 2002, 6(6): 580.
[17] Zhang B, Chen S, Gao S, et al. IEEE Transactions on Instrumentation and Measurement, 2023, 72: 1000112.
[18] Gharehchopogh F S. Artificial Intelligence Review, 2023, 56(6): 5479.
[19] Kruse F A, Lefkoff A B, Boardman J W, et al. Remote Sensing of Environment, 1993, 44(23): 145.
[20] Liu D, Han L. International Journal of Pattern Recognition and Artificial Intelligence, 2022, 36(1): 2254002.
[21] Agrawal N, Govil H, Mishra G, et al. Remote Sensing, 2023, 15(12): 3133.
[22] Alam T, Raza Z. Journal of Systems and Software, 2018, 135: 126.
[23] Huang D, Wang C D, Wu J S, et al. IEEE Transactions on Knowledge and Data Engineering, 2020, 32(6): 1212.
