Yüksek Çözünürlüklü Uzaktan Algılama Görüntülerinden Bina Segmentasyonu: Evrişimsel ve Dönüştürücü Mimarilerin Karşılaştırmalı Değerlendirmesi

Öz

Bina ayıklama çalışmalarında mimariler çoğunlukla farklı eğitim ve çıkarım kurguları altında karşılaştırılmakta, bu da bildirilen başarım değerlerini doğrudan kıyaslanamaz kılmaktadır. Bu çalışmada üç evrişimsel (U-Net, U-Net++, DeepLabV3+) ve bir dönüştürücü tabanlı (SegFormer) bölütleme mimarisi, WHU Bina Veri Kümesi'nin hava ortofoto alt kümesi üzerinde tek bir eğitim ve çıkarım protokolü altında değerlendirilmiştir. Evrişimsel modellerde kodlayıcı ResNet-34 olarak sabitlenerek kod çözücü tasarımının etkisi omurga kapasitesinden ayrıştırılmıştır. En yüksek başarımı %89,04 IoU ve %94,20 F1 ile U-Net++ elde etmiş, modeller arasındaki fark standart IoU ölçütünde 2,42 puanlık dar bir bantta kalmıştır. Buna karşılık yalnızca kenar bandındaki örtüşmeyi ölçen sınır IoU ölçütünde aynı yayılım 9,26 puana, yani yaklaşık 3,8 katına çıkmıştır. Bu bulgu, standart IoU'nun tek başına raporlanmasının mimariler arasındaki farkı sistematik biçimde küçük gösterdiğine işaret etmektedir. Üstel hareketli ortalama, sekiz yönlü test-zamanı artırma ve doğrulama kümesinde ayarlanan karar eşiğinden oluşan çıkarım protokolü 0,47–0,69 IoU puanı kazanç sağlamış, model sıralamasını değiştirmemiştir; bu büyüklük mimari kaynaklı toplam yayılımın yaklaşık dörtte birine karşılık gelmektedir. Hesaplama maliyeti tarafında parametre sayısının yanıltıcı bir gösterge olduğu görülmüş, U-Net++ ile U-Net arasındaki %6,7'lik parametre farkına karşılık %135,2'lik çarpma–toplama işlemi farkı ölçülmüştür. Sonuçlar, bina ayıklamada model seçiminin tek bir doğruluk değeri üzerinden yapılamayacağını, sınır konumunun kritik olduğu uygulamalarda sınır temelli ölçütlerin esas alınması gerektiğini göstermektedir.

Anahtar Kelimeler

• Bina çıkarımı
• Uzaktan algılama
• Hava ortofotosu
• Semantik bölütleme
• Derin öğrenme

Building Segmentation from High-Resolution Remote Sensing Imagery: A Comparative Evaluation of Convolutional and Transformer Architectures

Öz

Architectures for building extraction are usually compared under different training and inference set-ups, which makes the reported performance figures incommensurable. This study evaluates three convolutional segmentation architectures (U-Net, U-Net++, DeepLabV3+) and one transformer-based architecture (SegFormer) on the aerial orthophoto subset of the WHU Building Dataset under a single training and inference protocol. Fixing the encoder to ResNet-34 in the convolutional models separates the effect of decoder design from backbone capacity. U-Net++ achieved the highest performance at 89.04% IoU and 94.20% F1, and the difference between the models remained within a narrow range of 2.42 points on the standard IoU metric. On the boundary IoU metric, which measures overlap only within the edge band, the same spread rose to 9.26 points, roughly 3.8 times as wide. Reporting standard IoU alone therefore understates the difference between architectures systematically. An inference protocol combining an exponential moving average, eight-way test-time augmentation and a decision threshold tuned on the validation set yielded gains of 0.47 to 0.69 IoU points without changing the ranking of the models; that magnitude corresponds to roughly a quarter of the total spread attributable to architecture. On the cost side, parameter count proved a misleading indicator: a 6.7% difference in parameters between U-Net++ and U-Net corresponded to a 135.2% difference in multiply-accumulate operations. The results show that model selection in building extraction cannot rest on a single accuracy figure, and that boundary-based metrics should govern the choice in applications where the position of the boundary is critical.

Anahtar Kelimeler

• Building extraction
• Remote sensing
• Aerial orthophoto
• Semantic segmentation
• Deep learning

Kaynakça

1. [1] Ji S, Wei S, Lu M. Fully Convolutional Networks for Multisource Building Extraction From an Open Aerial and Satellite Imagery Data Set. IEEE Transactions on Geoscience and Remote Sensing, 57(574-586), (2019).
2. [2] Sirmacek B, Unsalan C. Building detection from aerial images using invariant color features and shadow information. 2008 23rd International Symposium on Computer and Information Sciences, 1-5, (2008).
3. [3] Liow Y-T, Pavlidis T. Use of shadows for extracting buildings in aerial images. Computer Vision, Graphics, and Image Processing, 49(242-277), (1990).
4. [4] Long J, Shelhamer E, Darrell T. Fully Convolutional Networks for Semantic Segmentation. 3431-3440, (2015).
5. [5] Ronneberger O, Fischer P, Brox T. U-Net: Convolutional Networks for Biomedical Image Segmentation. Medical Image Computing and Computer-Assisted Intervention – MICCAI 2015, Springer International Publishing, Cham, 234-241, (2015).
6. [6] Zhou Z, Rahman Siddiquee MM, Tajbakhsh N, Liang J. UNet++: A Nested U-Net Architecture for Medical Image Segmentation. Deep Learning in Medical Image Analysis and Multimodal Learning for Clinical Decision Support, Springer International Publishing, Cham, 3-11, (2018).
7. [7] Chen L-C, Zhu Y, Papandreou G, Schroff F, Adam H. Encoder-Decoder with Atrous Separable Convolution for Semantic Image Segmentation. Computer Vision – ECCV 2018, Springer International Publishing, Cham, 833-851, (2018).
8. [8] Liu Z, Lin Y, Cao Y, Hu H, Wei Y, Zhang Z, et al. Swin Transformer: Hierarchical Vision Transformer using Shifted Windows. 2021 IEEE/CVF International Conference on Computer Vision (ICCV), 9992-10002, (2021).

9. [9] Xie E, Wang W, Yu Z, Anandkumar A, Alvarez JM, Luo P. SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers. Advances in Neural Information Processing Systems, Curran Associates, Inc., 12077-12090, (2021).
10. [10] Wang H, Miao F. Building extraction from remote sensing images using deep residual U-Net. European Journal of Remote Sensing, 55(71-85), (2022).
11. [11] Liu H, Luo J, Huang B, Hu X, Sun Y, Yang Y, et al. DE-Net: Deep Encoding Network for Building Extraction from High-Resolution Remote Sensing Imagery. Remote Sensing, 11(2380), (2019).
12. [12] Lin J, Jing W, Song H, Chen G. ESFNet: Efficient Network for Building Extraction From High-Resolution Aerial Images. IEEE Access, 7(54285-54294), (2019).
13. [13] Yu M, Chen X, Zhang W, Liu Y. AGs-Unet: Building Extraction Model for High Resolution Remote Sensing Images Based on Attention Gates U Network. Sensors, 22(2932), (2022).
14. [14] Yang G, Zhang Q, Zhang G. EANet: Edge-Aware Network for the Extraction of Buildings from Aerial Images. Remote Sensing, 12(2161), (2020).
15. [15] Wang Y, Zhao Q, Wu Y, Tian W, Zhang G. SCA-Net: Multiscale Contextual Information Network for Building Extraction Based on High-Resolution Remote Sensing Images. Remote Sensing, 15(4466), (2023).
16. [16] Wang Y, Wang S, Dou A. A Dual-Branch Fusion Network Based on Reconstructed Transformer for Building Extraction in Remote Sensing Imagery. Sensors, 24(365), (2024).
17. [17] Yuan W, Xu W. MSST-Net: A Multi-Scale Adaptive Network for Building Extraction from Remote Sensing Images Based on Swin Transformer. Remote Sensing, 13(4743), (2021).
18. [18] Song J, Zhu A-X, Zhu Y. Transformer-Based Semantic Segmentation for Extraction of Building Footprints from Very-High-Resolution Images. Sensors, 23(5166), (2023).
19. [19] Hao M, Chen S, Lin H, Zhang H, Zheng N. A prior knowledge guided deep learning method for building extraction from high-resolution remote sensing images. Urban Informatics, 3(6), (2024).
20. [20] Chueprasert T, Udomchaiporn A, Intagosum S. Comparative Analysis of Deep Learning Models for Building Extraction from High-resolution Satellite Imagery. CURRENT APPLIED SCIENCE AND TECHNOLOGY, e0260846-e0260846, (2025).
21. [21] Üzen H, Türkoğlu M, Hanbay D. Result Weighting-Based Resnet Feature Pyramid Network Architecture for Surface Defect Detection. Gazi Üniversitesi Fen Bilimleri Dergisi Part C: Tasarım ve Teknoloji, 9(760-772), (2021).
22. [22] Aslan A, Yazicioglu C. Building Extraction from Remote Sensing Imagery: Quantifying Label Efficiency and the Effect of Transfer Learning on Annotation Cost. Konya, Türkiye, 422-435, (2026).
23. [23] He K, Zhang X, Ren S, Sun J. Deep Residual Learning for Image Recognition. 770-778, (2016).
24. [24] Yang D, Gao X, Yang Y, Guo K, Han K, Xu L. Advances and Future Prospects in Building Extraction From High-Resolution Remote Sensing Images. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 18(6994-7016), (2025).
25. [25] Chen M, Wu J, Liu L, Zhao W, Tian F, Shen Q, et al. DR-Net: An Improved Network for Building Extraction from High Resolution Remote Sensing Image. Remote Sensing, 13(294), (2021).