Hiperspektral görüntülerin graf tabanlı boyut indirgenerek sınıflandırılmasında parçacık sürü optimizasyonu yaklaşımı

Year 2024, Volume: 14 Issue: 4, 1219 - 1234, 15.12.2024

Betül Dolapcı , Caner Özcan , Emrah Özkaynak

https://doi.org/10.17714/gumusfenbil.1493821

Abstract

Hiperspektral görüntü verilerinden hem uzamsal hem spektral öznitelik çıkarımı ile analiz işlemi için gerekli detaylı bilgiler elde edilmektedir. Yüksek boyutlu görüntü verilerinin daha düşük boyutlu temsillerini sağlamada Laplacian Özharitaları (LÖ) ve Schrödinger Özharitaları (SÖ) graf tabanlı boyut azaltma algoritmalarının etkili olduğu bilinmektedir. Ancak bu yöntemler kapsamında kullanılan boyut azaltma parametresi değerinin literatürde sabit bir değer olarak kullanıldığı görülmektedir. Önerdiğimiz çalışma kapsamında bu parametre Parçacık Sürü Optimizasyounu (PSO) ile optimize edilmiştir. Öncelikle görüntüden Basit Doğrusal Yinelemeli Kümeleme (BDYK) algoritması ile kümelenmiş süperpikseller elde edilmiştir. Daha sonra süperpikseller graf veri yapısına dönüştürülüp girdi olarak LÖ ve SÖ algoritmalarına verilmiştir. Boyut azaltma işlemi sürecinde elde edilen farklı boyutlar için araya eklenen PSO algoritması ile en iyi özvektör değeri hesaplanmaktadır. En iyi özvektör değeri Indian Pines, Salinas ve Pavia Üniversitesi veri setleri için, 130, 120 ve 40 olarak hesaplanmıştır. Son aşamada optimizasyon tabanlı yöntemle elde edilen en iyi sonuçlar üzerinde Destek Vektör Makinesi (DVM) ile sınıflandırma işlemi gerçekleştirilmiştir. Tüm veri setleri için sınıflandırma doğruluklarının en iyi özvektör değeri ile arttırılması sağlanmıştır.

Keywords

Boyut indirgeme, DVM, Hiperspektral, PSO, Sınıflandırma

Particle swarm optimization approach for graph-based dimensionality reduction classification of hyperspectral images

Abstract

By extracting both spatial and spectral features from hyperspectral image data, detailed information required for the analysis process is obtained. It is important to provide lower dimensional representations of high-dimensional image data and Laplacian Eigenmaps (LÖ) and Schrödinger Eigenmaps (SÖ) graph-based dimension reduction algorithms are known to be effective for this. However, it is seen that the dimensionality reduction parameter value used in these methods is used as a fixed value in the literature. In our proposed work, this parameter is optimized with Particle Swarm Optimization (PSO). First, superpixels clustered by Simple Linear Iterative Clustering (LLICC) algorithm are obtained from the image. Then, the superpixels are transformed into a graph data structure and given as input to the LÖ and SÖ algorithms. The best eigenvector value is calculated with the PSO algorithm added for different dimensions obtained during the dimensionality reduction process. The best eigenvector values were calculated as 130, 120 and 40 for Indian Pines, Salinas and Pavia University datasets. In the last stage, classification process was performed with Support Vector Machine (SVM) on the best results obtained with the optimization-based method. Classification accuracies for all data sets were increased with the best eigenvector value.

Keywords

Dimension reduction, SVM, Hyperspectral, PSO, Classificiation

References

• Achanta, R., Shaji, A., Smith, K., Lucchi, A., Fua, P. & Süsstrunk, S. (2012). SLIC superpixels compared to state-of-the-art superpixel methods. IEEE Transactions on Pattern Analysis and Machine Intelligence, 34(11), 2274-2282. https://doi: 10.1109/TPAMI.2012.120.
• Acosta, I., C., C., Khodadadzadeh, M., Tolosana-Delgado, R. & Gloaguen, R. (2020). Drill-Core hyperspectral and geochemical data integration in a superpixel-based machine learning framework. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 13, 4414-4228. https://doi: 10.1109/JSTARS.2020.3011221.
• Alasvand, Z., Naderan, M. & Akbarizadeh, (2017). Superpixel-based feature learning for joint sparse representation of hyperspectral images. 2017 3rd International Conference on Pattern Recognition and Image Analysis (IPRIA), 156-159. https://doi: 10.1109/PRIA.2017.7983037.
• Belkin, M., & P. Niyogi. (2003). Laplacian eigenmaps for dimensionality reduction and data representation. Neural Computation 15(6): 1373–1396. https://doi:10.1162/089976603321780317.
• Bernabe, S., P. Reddy Marpu, A. Plaza, M. Dalla Mura, & J. Atli Benediktsson (2014). Spectral–Spatial classification of multispectral images using kernel feature space representation. IEEE Geoscience and Remote Sensing Letters 11 (1): 288–292. https://doi:10.1109/LGRS.2013.2256336.
• Cahill, N., D., W. Czaja, D. & W., Messinger, (2014). Schroedinger eigenmaps with nondiagonal potentials for spatial-spectral clustering of hyperspectral imagery. Algorithms and Technologies for Multispectral, Hyperspectral, and Ultraspectral Imagery XX, Vol. 9088, International Society for Optics and Photonics, p. 908804.
• Cahill, N. D., S. E. Chew, & P. S. Wenger (2015). Spatial-Spectral dimensionality reduction of hyperspectral imagery with partial knowledge of class labels. proceedings SPIE Defense & Security: Algorithms and Technologies for Multispectral, Hyperspectral, and Ultraspectral Imagery XXI 94720: 1–14. https://doi:10.1117/12.2177139.
• Czaja, W., & M. Ehler (2013). Schroedinger eigenmaps for the analysis of biomedical data. IEEE Transactions on Pattern Analysis and Machine Intelligence 35: 1274–1280. https://doi:10.1109/TPAMI.2012.270.
• Datta, D., Mallick, P., K., Bhoi, A., K., Ijaz, M., F., Shafi, J. & Choi, J (2022). Hyperspectral image classification: potentials, challenges, and future directions. Advanced Computational Intelligence Algorithms for Signal and Image Processing, https://doi.org/10.1155/2022/3854635.
• Fejjari, A., Saheb Ettabaa, K. & Korbaa, O. Spatial spectral schroedinger eigenmaps approach based on spectral angle distance for hyperspectral imagery classification. Indian Soc Remote Sens 49, 2689–2700 (2021). https://doi.org/10.1007/s12524-021-01417-3.
• Gao F, Wang Q, Dong J & Xu Q. Spectral and spatial classification of hyperspectral images based on random multi-graphs. Remote Sensing, https://doi.org/10.3390/rs10081271.
• Ghasrodashti, E., K., Helfroush, M., S. & Danyali, H. (2017). A wavelet-based classification of hyperspectral images using Schroedinger eigenmaps. International Journal of Remote Sensing, https://doi:10.1080/01431161.2017.1302108.
• GIC-Grupo De Inteligencia Computacional. (2021, 12 Temmuz). https://www.ehu.eus/ccwintco/index.php/Hyperspectral_Remote_Sensing_Scenes
• Gurram, P. & Kwon, H.: Support-vector-based hyperspectral anomaly detection using optimized kernel parameters. IEEE Geosci. Remote Sens. Lett. 8(6), 1060–1064 (2011). https://doi.org/10.1109/LGRS.2011.2155030.
• He, L., J. Li, C. Liu & S. Li, Recent advances on spectral–spatial hyperspectral image classification: an overview and new guidelines, IEEE Transactions on Geoscience and Remote Sensing, https://doi: 10.1109/TGRS.2017.2765364.
• Jia, S., Zhang, Z., Zhang, M., Xu, M., Huang, Q., Zhou, J. & Jia, X. (2021) Multiple feature-based superpixel-level decision fusion for hyperspectral and LiDAR data classification, IEEE Transactions on Geoscience and Remote Sensing, 2021, https://doi: 10.1109/TGRS.2020.2996599.
• Kennedy, S., M,. W. Williamson, J. D. Roth & J. W. Scrofani (2020). Cluster-Based spectral-spatial segmentation of hyperspectral imagery, IEEE Access, https://doi: 10.1109/ACCESS.2020.3011668.
• Kim, D.H. & L.H. Finkel, Hyperspectral image processing using locally linear embedding, in: Neural Engineering, 2003. Conference Proceedings. First International IEEE EMBS Conference on, IEEE, 2003, pp. 316–319.
• Özdemir, A. & Polat K. (2020). Deep learning applications for hyperspectral imaging: a systematic review. Journal of the Institute of Electronics and Computer, https://doi.org/10.33969/JIEC.2020.21004.
• Özer, F. & Özkaya, U. (2017). Süperpiksel algoritmalarının gürültülü imgeler için bölütleme performansının incelenmesi, Akıllı Sistemlerde Yenilikler ve Uygulamaları Konferansı 2017, Antalya, Turkey.
• Suresh, S. & Lal, S. (2019). A metaheuristic framework based automated spatial-spectral graph for land cover classification from multispectral and hyperspectral satellite images. Infrared Physics & Technology, https://doi.org/10.1016/j.infrared.2019.103172.
• Verdoja, F. & Grangetto, M. (2020). Graph Laplacian for image anomaly detection. Machine Vision and Applications, https://doi.org/10.1007/s00138-020-01059-4.
• Üstüner, M. (2023). Çekirdek tabanlı aşırı öğrenme makinesi ile hiperspektral görüntü sınıflandırma. Turkish Journal of Remote Sensing and GIS, 4(2), 198-212. https://doi.org/10.48123/rsgis.1237772.
• Wang, L., Peng, J. & Sun, W. Spatial–Spectral squeeze-and-excitation residual network for hyperspectral image classification. Remote Sens. 2019, 11, 884.
• Zhang, X., Selene, E., Chew, Zhenlin, Xu & Nathan D. Cahill, SLIC superpixels for efficient graph-based dimensionality reduction of hyperspectral imagery, Proc. SPIE 9472, Algorithms and Technologies for Multispectral, Hyperspectral, and Ultraspectral Imagery XXI, 947209 (21 May 2015), https://doi.org/10.1117/12.2176911.
• Zhang, X., Y. Liang & N. Cahill, Using superpixels to improve the efficiency of Laplacian Eigenmap based methods for target detection in hyperspectral imagery, 2016 IEEE International Geoscience and Remote Sensing Symposium (IGARSS), Beijing, China, 2016, pp. 5876-5879, https://doi: 10.1109/IGARSS.2016.7730535.
• Zhao, Y. & Yan, F. (2021). Hyperspectral image classification based on sparse superpixel graph. Remote Sensing, https://doi.org/10.3390/rs13183592.