Uydu İmgelerine Derin Öğrenme Tabanlı Süper Çözünürlük Yöntemlerinin Uygulanması

Askeri ve sivil hayatta önemli tüm uygulamalar için kullanılan görüntünün çözünürlüğünün yüksek olması çok önemlidir. Uydu imgeleri barındıran çalışmalarda süper çözünürlük ile iyileştirilmiş imgelerin kullanımı bina tespiti gibi uygulamalarda gereklidir. Düşük çözünürlüklü görüntünün giriş olarak verildiği süper çözünürlük algoritmalarında çeşitli iyileştirme adımları neticesinde çıktı olarak yüksek çözünürlüklü görüntü elde edilir. Bu çalışmada kullanıma açık uydu görüntülerinden alınan 6 sınıfa ayrılmış toplam 900 imge üzerinde, derin öğrenme tabanlı evrişimsel sinir ağları ile süper çözünürlük iyileştirilmesinin performansı analiz edilmiştir. Veri seti üzerinde derin öğrenme için test ve eğitim verileri ayrılmıştır. Verilere DenseNet201, SqueezeNet, Vgg16 olmak üzere toplam 3 derin öğrenme mimarisi ayrı ayrı uygulanmıştır. Süper çözünürlük adımı öncesinde ve sonrasında doğru sınıflandırılmış veri oranının kontrolü için evrişimsel sinir ağları uygulanmıştır. Sınıflandırma sonuçları karşılaştırılmış, sınıflandırma sonucunda evrişimsel sinir ağları öğrenme özellikleri süper çözünürlük sayesinde iyileştirilmiştir. Sınıflandırılma başarısı 6 sınıflandırılma mimarisi için en düşük %2,4 ve en yüksek %3,6 oranında arttırılmış olduğu kanıtlanmıştır.

Application of Deep Learning Based Super Resolution Methods To Satellite Images And Improvement Of Images

High resolution of the image used for all important applications in military and civil life is very important. In works with satellite images, the use of images enhanced with super resolution is necessary in applications such as building detection. In the super resolution algorithms where the low-resolution image is given as input, high resolution image is obtained as a result of various improvement steps. The performance of super resolution improvement with deep learning based convolutional neural networks on 900 images taken from available satellite images was analyzed. Test and training data are reserved for deep learning on the dataset. A total of 3 softmax functions (DenseNet201, SqueezeNet, Vgg16) were applied to the data separately. Evolutionary neural networks were applied to control the number of correctly classified data before and after the super resolution step. The classification results are compared and as a result of the classification, the learning properties of the convolutional neural networks are increased by super resolution. Classification success has proven to be increased by the lowest 2.4% and the highest 3.6% for the 6 classification architectures.

PDF

___

Bevilacqua, M., Roumy, A., Guillemot, C., Morel, M.L.A. 2012. Low- complexity single-image super-resolution based on nonnegative neighbor embedding, British Machine Vision Conference.
Chang, H., Yeung, D.Y., Xiong, Y. 2004. Super-resolution through neighbor embedding, IEEE Conference on Computer Vision and Pattern Recognition, 1-1.
Chao D., Chen C. L., Kaiming H., Xiaoou T. 2015. Image Super-Resolution Using Deep Convolutional Networks, IEEE Transactions on Pattern Analysis and Machine Intelligence, 38,2, 295-307.
Chopade P. B., Patil P. M. 2015. Image super resolution Scheme based on wavelet transform and its performance analysis, International Conference on Computing, Communication and Automation, 1182-1186.
Dai, D., Timofte, R., Van Gool, L. 2015. Jointly optimized regressors for image super-resolution, Eurographic, 95-104.
Freedman, G., Fattal, R. 2011. Image and video upscaling from local self-examples, ACM Transactions on Graphics, 30,2, 12.
Freeman, W.T., Pasztor, E.C., Carmichael, O.T. 2000. Learning low-level vision, International Journal of Computer Vision, 40,1, 25–47.
Glasner, D., Bagon, S., Irani, M. 2009. Super-resolution from a single image, IEEE International Conference on Computer Vision, 349–356.
Jia-Bin H., Abhishek S., and Narendra A., 2015, Single Image Super-Resolution from Transformed Self-Exemplars, Computer Vision and Pattern Recognition, 5197-5206.
Kim, K.I., Kwon, Y. 2010. Single-image super-resolution using sparse regression and natural image prior, IEEE Transactions on Pattern Analysis and Machine Intelligence, 32,6, 1127–1133.
Mateen M., Wen J., Song S.N., Huang Z. 2019. Fundus Image Classification Using VGG-19 Architecture with PCA and SVD, Symmetry.
Schulter, S., Leistner, C., Bischof, H. 2015. Fast and accurate image upscaling with super-resolution forests, IEEE Conference on Computer Vision and Pattern Recognition, 3791–3799.
Simonyan K., Zisserman A. 2014. Very deep convolutional networks for large-scale image recognition, arXiv, 1409.1556.
Timofte R., De Smet, Van Gool L. 2013. Anchored neighborhood regression for fast example-based super-resolution, IEEE In- ternational Conference on Computer Vision, 1920–1927.
Timofte, R., De Smet, V., Van Gool, L. 2014. A+: Adjusted anchored neighborhood regression for fast super-resolution, IEEE Asian Conference on Computer Vision, 111–126.
Yang, C.Y., Huang, J.B., Yang, M.H. 2010. Exploiting self-similarities for single frame super-resolution, IEEE Asian Conference on Computer Vision, 497–510.
Yang, C.Y., Ma, C., Yang, M.H. 2014. Single image super resolution: A benchmar, European Conference on Computer Vision, 372–386.
Yang, J., Lin, Z., Cohen, S. 2013. Fast image super-resolution based on in-place example regression, IEEE Conference on Computer Vision and Pattern Recognition, 1059–1066.
Yang, J., Wang, Z., Lin, Z., Cohen, S., Huang, T. 2012 Coupled dictionary training for image super-resolution, IEEE Transactions on Image Processing, 21,8, 3467–3478.
Yang, J., Wright, J., Huang, T., Ma, Y. 2008. Image super-resolution as sparse representation of raw image patches, IEEE Conference on Computer Vision and Pattern Recognition, 1–8.
Yang, J., Wright, J., Huang, T.S., Ma, Y. 2010. Image super-resolution via sparse representation, IEEE Transactions on Image Processing, 19,11, 2861–2873.
Zeyde, R., Elad, M., Protter, M. 2012. On single image scale-up using sparse-representations, Curves and Surfaces, 711–730.
Zhang J. 2019. A full convolutional network based on DenseNet for remote sensing scene classification, Mathematical Biosciences and Engineering, 6, 5, 3345–3367.
1-http://www.google.com/int/tr/earth, (01.04.2020)