本研究針對都市高解析度航空真正射影像中常見之陰影問題，提出一套結合多波段影像與深度學習之陰影偵測與去除方法。都市環境中因高樓建築產生之陰影，易造成地物判釋困難與分類誤差，進而影響後續空間分析與應用。因此，本研究整合可見光影像(RGB)、近紅外波段(NIR)與數值高度模型(DHM)，建構多波段輸入之深度學習模型，以提升陰影辨識之準確性，並進一步進行陰影補償與去除。
研究方法上，首先利用密點雲產製RGB與NIR真正射影像，並確保多源影像間之幾何一致性，接著建立陰影標註資料，採用以ResU-Net模型進行訓練。實驗以台北市信義區為主要訓練區域，並透過不同波段組合(RGB、RGB+NIR、RGB+NIR+DHM)進行模型效能比較。結果顯示，加入NIR與DHM後，IoU提升至0.9584，F1-Score 提升至0.9788，顯示多波段與高度資訊能有效提升陰影辨識能力。後續模型遷移至南投縣南崗工業區學習，評估模型於不同都市型態與地物組成下之適應能力，也在測試影像集達到IoU 0.8204、F1-Score 0.9014的成果。
進一步分析發現，RGB光譜資訊在複雜都市場景中易與高反射或低亮度地物產生混淆，而NIR可強化陰影與地物之光譜差異，DHM則提供高度資訊，使模型能有效區分陰影與建物或地表物件之空間關係。最後，本研究進行陰影去除作業，並透過影像評估指標評估補償效果，成功生成陰影影響降低之真正射影像。本研究證實，結合光譜與高度資訊之深度學習方法能有效提升都市航空真正射影像中陰影偵測與去除之準確性與穩定性，雖然在資料多樣性、模型泛化能力及陰影補償精細度方面仍有進一步提升空間，但對於高解析度遙測影像之地物判釋、城市分析及後續應用具有重要參考價值。

This study proposes a shadow detection and removal framework that integrates multispectral imagery and deep learning techniques to address the shadow problem commonly found in high-resolution urban aerial true orthoimages. Shadows cast by high-rise buildings in urban environments often lead to difficulties in object interpretation and classification errors, thereby affecting subsequent spatial analysis and applications. To overcome this issue, visible imagery (RGB), near-infrared (NIR) imagery, and a Digital Height Model (DHM) were integrated to construct a multi-band deep learning model for improving shadow detection accuracy and performing subsequent shadow compensation and removal.
In terms of methodology, RGB and NIR true orthoimages were first generated from dense point clouds while ensuring geometric consistency among multi-source datasets. A shadow-labeled dataset was then established, and a ResU-Net model was adopted for training. Experiments were conducted using Xinyi District, Taipei City, as the primary training area, and different input combinations, including RGB, RGB+NIR, and RGB+NIR+DHM, were evaluated and compared. The results indicate that incorporating NIR and DHM improved the Intersection over Union (IoU) to 0.9584 and the F1-score to 0.9788, demonstrating that multispectral and elevation information can effectively enhance shadow detection performance. The trained model was subsequently transferred to Nangang Industrial Park in Nantou County to evaluate its adaptability under different urban characteristics and land-cover compositions. The transferred model achieved an IoU of 0.8204 and an F1-score of 0.9014 on the testing dataset.
Further analysis revealed that RGB spectral information alone can be easily confused with highly reflective or low-intensity objects in complex urban environments. In contrast, NIR imagery enhances the spectral separability between shadows and surrounding objects, while DHM provides elevation information that enables the model to better distinguish spatial relationships between shadows and buildings or ground features. Finally, shadow removal was performed, and compensation effectiveness was evaluated using image quality assessment metrics, successfully generating true orthoimages with reduced shadow effects.
The results demonstrate that integrating spectral and elevation information within a deep learning framework can significantly improve the accuracy and robustness of shadow detection and removal in urban aerial true orthoimages. Although further improvements are still possible in terms of data diversity, model generalization capability, and shadow compensation refinement, the proposed approach provides valuable support for object interpretation, urban analysis, and subsequent applications using high-resolution remote sensing imagery.

1.中文參考文獻
李欣宜，2019，「UAS 航測於地籍圖重測效率提升之研究－以苗栗縣竹南鎮鹽館前段為例」，嘉南藥理大學應用空間資訊系碩士論文：臺南。
邱式鴻、邱俊榮，2025，「都市區有人機傾斜攝影製作真正射影像精度評估與探討」，『國土測繪與空間資訊』，13(1)：109-130。
邱能賢，2020，「藉由深度學習去除相片中之陰影」，國立臺灣大學資訊工程學研究所碩士論文：臺北。
張志瑋、蔡展榮，2016，「空照影像陰影區塊之偵測和色彩修復之研究」，『航測及遙測學刊』，21(4)：239-256。
張譽鐙，2021，「基於深度學習與影像處理之陰影碎塊匹配」，國立臺北科技大學資訊工程系碩士論文：臺北。
賴忠佑，2017，「基於傾斜攝影製作真正射影像之最佳化處理」，國立高雄應用科技大學土木工程與防災科技研究所碩士論文：高雄。
謝皓展，2025，「以深度學習自動偵測航空真正射影像中的建物邊界像素」，國立政治大學地政學系碩士論文：臺北。

2.英文參考文獻
Adeline, K. R. M., Chen, M., Briottet, X., Pang, S. K., & Paparoditis, N. (2013). Shadow detection in very high spatial resolution aerial images: A comparative study. ISPRS Journal of Photogrammetry and Remote Sensing, 80, 21–38.
Adler-Golden, S. (2002). An Algorithm for De-Shadowing Spectral Imagery.
Akhtar, M., Murtza, I., Adnan, M., & Saadia, A. (2023). Cross-domain transfer learning for natural scene classification of remote-sensing imagery. Applied Sciences, 13(13), 7882.
Arévalo, V., González, J., & Ambrosio, G. (2008). Shadow detection in colour high‐resolution satellite images. International Journal of Remote Sensing, 29(7), 1945–1963.
Balduzzi, D., Frean, M., Leary, L., Lewis, J. P., Ma, K. W.-D., & McWilliams, B. (2018). The Shattered Gradients Problem: If resnets are the answer, then what is the question? (No. arXiv:1702.08591). arXiv.
Chai, D., Newsam, S., Zhang, H. K., Qiu, Y., & Huang, J. (2019). Cloud and cloud shadow detection in Landsat imagery based on deep convolutional neural networks. Remote Sensing of Environment, 225, 307–316.
Chen, Y., Briese, C., Karel, W., & Pfeifer, N. (2014). True orthophoto generation using multi-view aerial images. The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, XL–3, 67–71.
Dabove, P., Daud, M., & Olivotto, L. (2024). Revolutionizing urban mapping: Deep learning and data fusion strategies for accurate building footprint segmentation. Scientific Reports, 14(1), 13510.
Dong, X., Cao, J., & Zhao, W. (2024). A review of research on remote sensing images shadow detection and application to building extraction. European Journal of Remote Sensing, 57(1), 2293163.
Dozier, J., 1989 ―Spectral signature of alpine snow cover from the Landsat Thematic Mapper‖, Remote Sensing Environment, Vol. 28, pp. 9-22.
Dumoulin, V., & Visin, F. (2016). A guide to convolution arithmetic for deep learning. arXiv preprint arXiv:1603.07285.
Firmenichy, D., Brown, M., & Susstrunk, S. (2011). Multispectral interest points for RGB-NIR image registration. 2011 18th IEEE International Conference on Image Processing, 181–184.
Glorot, X., Bordes, A., & Bengio, Y. (n.d.). Deep Sparse Rectifier Neural Networks.
Guangyao, D., Huili, G., Wenji, Z., Xinming T., and Beibei, C., 2013 ―An index-based shadow extraction approach on high-resolution images‖, International Symposium on Satellite Mapping Technology and Application, pp. 19–26.
Guo, Y., Stolyar, A. L., & Walid, A. (2018). Shadow-Routing Based Dynamic Algorithms for Virtual Machine Placement in a Network Cloud. IEEE Transactions on Cloud Computing, 6(1), 209–220.
Guoqing Zhou, Weirong Chen, Kelmelis, J. A., & Deyan Zhang. (2005). A comprehensive study on urban true orthorectification. IEEE Transactions on Geoscience and Remote Sensing, 43(9), 2138–2147.
Hall, K. (n.d.). Organization: Old Dominion University 5238 Hampton Road Norfolk, VA 23529.
Han, H., Han, C., Lan, T., Huang, L., Hu, C., & Xue, X. (2020). Automatic Shadow Detection for Multispectral Satellite Remote Sensing Images in Invariant Color Spaces. Applied Sciences, 10(18), 6467.
He, K., Zhang, X., Ren, S., & Sun, J. (2016). Deep Residual Learning for Image Recognition. 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 770–778.
Heiskanen, J., Kajuutti, K., Jackson, M., Elvehoy, H., and Pellikka, P., 2002 ―Assessment of glaciological parameters using Landsat satellite data in Svartisen, Northern Norway‖, Proceedings of European Association of Remote sensing Laboratories (EARSel) Workshop on Observing Our Cryosphere from Space: Techniques and Methods for Monitoring Snow and Ice with Regard to Climate Change. Bern Switzerland, 11–13 March, pp. 34–42.
Hossin, M., & Sulaiman, M. N. (2015). A review on evaluation metrics for data classification evaluations. International Journal of Data Mining & Knowledge Management Process, 5(2), 1–11.
Hu, X., Zhu, L., Fu, C.-W., Qin, J., & Heng, P.-A. (2018). Direction-Aware Spatial Context Features for Shadow Detection. 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, 7454–7462.
KantSingh, K., Pal, K., & J. Nigam, M. (2012). Shadow Detection and Removal from Remote Sensing Images using NDI and Morphological Operators. International Journal of Computer Applications, 42(10), 37–40.
Khan, S. H., Bennamoun, M., Sohel, F., & Togneri, R. (2016). Automatic Shadow Detection and Removal from a Single Image. IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI), 38(3), 431–446
Li, Z., Shen, H., Li, H., Xia, G., Gamba, P., & Zhang, L. (2017). Multi-feature combined cloud and cloud shadow detection in GaoFen-1 wide field of view imagery. Remote Sensing of Environment, 191, 342–358.
Long, J., Shelhamer, E., & Darrell, T. (2015). Fully Convolutional Networks for Semantic Segmentation.
Loshchilov, I., & Hutter, F. (2019). Decoupled Weight Decay Regularization (No. arXiv:1711.05101). arXiv.
Luo, D., Zhang, H. K., De Lemos, H., Li, J., & Roy, D. P. (2025). A globally applicable deep learning model for Sentinel-2 cloud and shadow detection. Science of Remote Sensing, 12, 100278.
Luo, S., Li, H., & Shen, H. (2020). Deeply supervised convolutional neural network for shadow detection based on a novel aerial shadow imagery dataset. ISPRS Journal of Photogrammetry and Remote Sensing, 167, 443–457.
Luo, S., Shen, H., Li, H., & Chen, Y. (2019). Shadow removal based on separated illumination correction for urban aerial remote sensing images. Signal Processing, 165, 197–208.
M, H., & M.N, S. (2015). A Review on Evaluation Metrics for Data Classification Evaluations. International Journal of Data Mining & Knowledge Management Process, 5(2), 01–11.
Mostafa, Y. (2017). A Review on Various Shadow Detection and Compensation Techniques in Remote Sensing Images. Canadian Journal of Remote Sensing, 43(6), 545–562.
Otsu, N. (1979). A Tlreshold Selection Method from Gray-Level Histograms.
Polidorio, A. M., Flores, F. C., Imai, N. N., Tommaselli, A. M. G., & Franco, C. (2003). Automatic shadow segmentation in aerial color images. 16th Brazilian Symposium on Computer Graphics and Image Processing (SIBGRAPI 2003), 270–277.
Ronneberger, O., Fischer, P., & Brox, T. (2015). U-Net: Convolutional Networks for Biomedical Image Segmentation (No. arXiv:1505.04597). arXiv.
Rufenacht, D., Fredembach, C., & Susstrunk, S. (2014). Automatic and Accurate Shadow Detection Using Near-Infrared Information. IEEE Transactions on Pattern Analysis and Machine Intelligence, 36(8), 1672–1678.
Salamati, N., Fredembach, C., & Süsstrunk, S. (2009). Material Classification Using Color and NIR Images. Color and Imaging Conference, 17(1), 216–222.
Shariat, M., Azizi, A., & Saadatseresht, M. (2008). ANALYSIS AND THE SOLUTIONS FOR GENERATING A TRUE DIGITAL ORTHO PHOTO IN CLOSE RANGE PHOTOGRAMMETRY.
Shor, Y., & Lischinski, D. (2008). The Shadow Meets the Mask: Pyramid‐Based Shadow Removal. Computer Graphics Forum, 27(2), 577–586.
Sudre, C. H., Li, W., Vercauteren, T., Ourselin, S., & Jorge Cardoso, M. (2017). Generalised Dice Overlap as a Deep Learning Loss Function for Highly Unbalanced Segmentations. In M. J. Cardoso, T. Arbel, G. Carneiro, T. Syeda-Mahmood, J. M. R. S. Tavares, M. Moradi, A. Bradley, H. Greenspan, J. P. Papa, A. Madabhushi, J. C. Nascimento, J. S. Cardoso, V. Belagiannis, & Z. Lu (Eds.), Deep Learning in Medical Image Analysis and Multimodal Learning for Clinical Decision Support (Vol. 10553, pp. 240–248). Springer International Publishing.
Teke, M., Başeski, E., Ok, A., Yüksel, B., & Senaras, C. (2011). Multi-spectral False Color Shadow Detection (Vol. 6952, pp. 109–119).
Tian, J., & Tang, Y. (2011). Linearity of each channel pixel values from a surface in and out of shadows and its applications. CVPR 2011, 985–992.
Tsai, V. J. (2006). A comparative study on shadow compensation of color aerial images in invariant color models. IEEE Transactions on Geoscience and Remote Sensing, 44(6), 1661–1671.
Wang, J., Li, X., & Yang, J. (2018). Stacked Conditional Generative Adversarial Networks for Jointly Learning Shadow Detection and Shadow Removal. 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, 1788–1797.
Watanabe, S., Miyajima, K., Inoue, U., & Nakamura, T. (2004). Detecting changes of buildings using stereo matching from aerial images taken at different positions and times. Electronics and Communications in Japan (Part III: Fundamental Electronic Science), 87(6), 55–65.
Weiss, K., Khoshgoftaar, T. M., & Wang, D. (2016). A survey of transfer learning. Journal of Big data, 3(1), 9.
Wu, S.-T., Hsieh, Y.-T., Chen, C.-T., & Chen, J.-C. (2014). A Comparison of 4 Shadow Compensation Techniques for Land Cover Classification of Shaded Areas from High Radiometric Resolution Aerial Images. Canadian Journal of Remote Sensing, 40(4), 315–326.
Xiao, C., Liu, M., Xiao, D., Dong, Z., & Ma, K.-L. (2014). Fast Closed-Form Matting Using a Hierarchical Data Structure. IEEE Transactions on Circuits and Systems for Video Technology, 24(1), 49–62.
Yang, Q., Tan, K.-H., & Ahuja, N. (2012). Shadow Removal Using Bilateral Filtering. IEEE Transactions on Image Processing, 21(10), 4361–4368.
Yang, Y., Ran, S., Gao, X., Wang, M., & Li, X. (2020). An Automatic Shadow Compensation Method via a New Model Combined Wallis Filter with LCC Model in High Resolution Remote Sensing Images. Applied Sciences, 10(17), 5799.
Yosinski, J., Clune, J., Bengio, Y., & Lipson, H. (2014). How transferable are features in deep neural networks?. Advances in neural information processing systems, 27.
Zhang, H., Sun, K., & Li, W. (2014). Object-Oriented Shadow Detection and Removal From Urban High-Resolution Remote Sensing Images. IEEE Transactions on Geoscience and Remote Sensing, 52(11), 6972–6982.
Zhang, L., Zhang, Q., Wu, Y., Zhang, Y., Xiang, S., Xie, D., & Wang, Z. (2024). A Full-Scale Shadow Detection Network Based on Multiple Attention Mechanisms for Remote-Sensing Images. Remote Sensing, 16(24), 4789.
Zhang, L., Zhang, Q., & Xiao, C. (2015). Shadow Remover: Image Shadow Removal Based on Illumination Recovering Optimization. IEEE Transactions on Image Processing, 24(11), 4623–4636.
Zhang, Y., Chen, G., Vukomanovic, J., Singh, K. K., Liu, Y., Holden, S., & Meentemeyer, R. K. (2020). Recurrent Shadow Attention Model (RSAM) for shadow removal in high-resolution urban land-cover mapping. Remote Sensing of Environment, 247, 111945.
Zhou, G., Chen, W., Kelmelis, J. A., & Zhang, D. (2005). A comprehensive study on urban true orthorectification. IEEE Transactions on Geoscience and Remote sensing, 43(9), 2138-2147.Zhou, T., Fu, H., Sun, C., & Wang, S. (2021). Shadow Detection and Compensation from Remote Sensing Images under Complex Urban Conditions. Remote Sensing, 13(4), 699.
Zhu, J., Guo, Y., Sun, G., Yang, L., Deng, M., & Chen, J. (2023). Unsupervised domain adaptation semantic segmentation of high-resolution remote sensing imagery with invariant domain-level prototype memory. IEEE Transactions on Geoscience and Remote Sensing, 61, 1-18.
Zhu, L., Deng, Z., Hu, X., Fu, C.-W., Xu, X., Qin, J., & Heng, P.-A. (2018). Bidirectional Feature Pyramid Network with Recurrent Attention Residual Modules for Shadow Detection. In V. Ferrari, M. Hebert, C. Sminchisescu, & Y. Weiss (Eds.), Computer Vision – ECCV 2018 (Vol. 11210, pp. 122–137). Springer International Publishing.