在地表環境快速變遷與災害頻仍的背景下，遙測影像作為地表資訊擷取的重要資料來源，其應用價值日益凸顯。地表變遷偵測傳統上以先影像分類再進行變遷分析，而深度學習技術被廣泛應用於遙測影像分析領域，並在時空特徵自動學習與非線性地物變遷同步判釋上有優勢。本研究旨在探討深度學習模型於多類別地表變遷偵測任務中的應用潛能，進一步分析其於異質資料條件與資料不完整情境下之表現差異與適應能力。研究以雙時期影像為主要輸入架構，直接將變遷前(T1)與變遷後(T2)之影像資料餵入深度神經網路，進行端對端的分類與變遷偵測。相較於傳統以先分類再比對的方式，此方法能夠同步分析兩時期影像之語意與光譜差異，提升變遷區域的識別準確性，並有效獲取增減(gain and loss)以及不變區(no-change)等資訊。
研究問題聚焦於三個核心面向：首先，針對不同解析度與資料來源（如Landsat與Sentinel-2）之影像，探討深度學習模型在處理異質輸入時的表現差異與資料整合策略；其次，分析在無法取得完整雙時期影像時，採用已分類遮罩(mask)影像替代第一時期輸入的可行性；第三，評估透過統計光譜匹配技術進行異質資料一致性調整（如光譜分布校正與超解析處理）是否能提升模型預測能力與輸出一致性。本研究以Densely Attentive Refinement Network(DARNet)為基礎，提出改良之多類別變遷偵測架構。原始DARNet採用UNet為骨幹網路，搭配密集注意力模組(Dense Attention Module)進行細部語意強化。本研究則針對實務情境進行架構調整，一方面建立multiclass DARNet模型，以處理複雜的多類別地物變遷任務；另一方面，考量第一時刻影像以遮罩取代之情境，修改原始架構中T1與T2共享編碼器(encoder)參數的設計，使其能分別學習處理不同類型輸入的特徵表示。此舉不僅提升模型對輸入資料異質性的容忍度，也能在遮罩取代策略下保有辨識變遷的能力。
資料來源部分涵蓋三種主要應用場景。包含海岸線變遷、紅樹林變遷與崩塌變遷。為處理不同解析度影像之融合問題，研究中另導入統計匹配與超解析技術將Landsat影像轉換為與Sentinel-2解析度一致之10公尺空間解析度，提升雙源資料輸入的一致性與可比較性。模型評估方面，本研究採用整體準確率(Overall Accuracy, OA)、交集比(Intersection over Union, IoU)、F1 Score與Kappa係數作為評估指標，並特別針對變遷類別進行分析，強調模型於變遷區域的敏感性與辨識能力。研究結果顯示，多分類DARNet在多分類變遷場景中具有較佳的表現，尤其在面對細小區域的沙洲堆積、崩塌生成與紅樹林退化等變遷時，辨識力顯著優於其他基準模型。此外，改良後之DARNet在第一時刻影像缺失情境下，以遮罩替代輸入之策略能保有一定的預測能力，驗證該策略於資料缺失情境下之可行性。同時，研究亦證實經統計匹配與超解析處理後的影像，能有效提升Landsat與Sentinel-2異質影像間的光譜一致性，並進一步改善模型之泛化與精度表現。此結果顯示，在實務應用中，結合資料前處理與深度學習架構調整，能有效克服多源資料整合與歷史資料缺失等挑戰。

In the context of rapid environmental change and frequent natural disasters, remote sensing imagery has become a vital data source for capturing surface information. Traditional land change detection methods often rely on single-date classification followed by post-comparison, which may limit their accuracy in complex scenarios. Recent advances in deep learning have shown promising capabilities in learning spatial-temporal patterns and identifying nonlinear surface changes. This study aims to investigate the potential of deep learning models in multi-class land change detection, particularly under heterogeneous data conditions and incomplete temporal coverage. We adopt a bi-temporal framework using deep convolutional neural networks, directly inputting before-change (T1) and after-change (T2) satellite imagery for end-to-end classification. A standard multiclass DARNet model is first developed to support the classification of gain, loss, and no-change regions. Building upon this, an improved version of DARNet is proposed, featuring separate encoders for different input types, allowing for mask substitution when T1 imagery is unavailable. This design enhances model adaptability to real-world data availability constraints. Experiments cover coastal, mangrove, and landslide change cases, using multi-source imagery from Landsat and Sentinel-2. To address spatial and spectral discrepancies between sensors, statistical matching is employed to align spectral distributions and improve cross-sensor consistency. Evaluation metrics include Overall Accuracy, IoU, F1 Score, and Kappa coefficient, with a focus on change-specific performance.Results demonstrate that the improved DARNet outperforms baseline models in identifying fine-scale changes such as sandbar shifts and localized deforestation. Even under partial data conditions, the model maintains acceptable prediction accuracy with mask inputs. Statistical matching further enhances spectral consistency across heterogeneous datasets.
Overall, this study provides a practical and adaptable deep learning solution for multi-class land change detection, addressing real-world challenges such as missing data and sensor heterogeneity, and supporting applications in environmental monitoring and disaster.

Aman, M. A., & Chu, H. J. (2023). Long-term river extent dynamics and transition detection using remote sensing: Case studies of Mekong and Ganga River. Science of the Total Environment, 876, 162774.
Baloloy, A. B., Blanco, A. C., Ana, R. R. C. S., & Nadaoka, K. (2020). Development and application of a new mangrove vegetation index (MVI) for rapid and accurate mangrove mapping. ISPRS Journal of Photogrammetry and Remote Sensing, 166, 95-117.
Bishop-Taylor, R., Nanson, R., Sagar, S., & Lymburner, L. (2021). Mapping Australia's dynamic coastline at mean sea level using three decades of Landsat imagery. Remote Sensing of Environment, 267, 112734.
Bishop-Taylor, R., Sagar, S., Lymburner, L., Alam, I., & Sixsmith, J. (2019). Sub-pixel waterline extraction: Characterising accuracy and sensitivity to indices and spectra. Remote Sensing, 11(24), 2984.
Blaschke, T. (2010). Object based image analysis for remote sensing. ISPRS journal of photogrammetry and remote sensing, 65(1), 2-16.
Bunting, P., Rosenqvist, A., Hilarides, L., Lucas, R. M., Thomas, N., Tadono, T., ... & Rebelo, L. M. (2022). Global mangrove extent change 1996–2020: global mangrove watch version 3.0. Remote Sensing, 14(15), 3657.
Canty, M. J., & Nielsen, A. A. (2006). Visualization and unsupervised classification of changes in multispectral satellite imagery. International Journal of Remote Sensing, 27(18), 3961-3975.
Chen, H., Xu, X., & Pu, F. (2024). Src-net: Bi-temporal spatial relationship concerned network for change detection. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing.
Feyisa, G. L., Meilby, H., Fensholt, R., & Proud, S. R. (2014). Automated Water Extraction Index: A new technique for surface water mapping using Landsat imagery. Remote sensing of environment, 140, 23-35.
Jamaluddin, I., Thaipisutikul, T., Chen, Y. N., Chuang, C. H., & Hu, C. L. (2021). MDPrePost-Net: A Spatial-Spectral-Temporal Fully Convolutional Network for Mapping of Mangrove Degradation Affected by Hurricane Irma 2017 Using Sentinel-2 Data. Remote Sensing, 13(24), 5042.
Konapala, G., Kumar, S. V., & Ahmad, S. K. (2021). Exploring Sentinel-1 and Sentinel-2 diversity for flood inundation mapping using deep learning. ISPRS Journal of Photogrammetry and Remote Sensing, 180, 163-173.
Li, W., & Gong, P. (2016). Continuous monitoring of coastline dynamics in western Florida with a 30-year time series of Landsat imagery. Remote Sensing of Environment, 179, 196-209.
Li, Z., Yan, C., Sun, Y., & Xin, Q. (2022). A densely attentive refinement network for change detection based on very-high-resolution bitemporal remote sensing images. IEEE Transactions on Geoscience and Remote Sensing, 60, 1-18.
Liao, C., Hu, H., Yuan, X., Li, H., Liu, C., Liu, C., ... & Zhu, Q. (2023). BCE-Net: Reliable building footprints change extraction based on historical map and up-to-date images using contrastive learning. ISPRS Journal of Photogrammetry and Remote Sensing, 201, 138-152.
Liao, P. S., Chen, T. S., & Chung, P. C. (2001). A fast algorithm for multilevel thresholding. J. Inf. Sci. Eng., 17(5), 713-727.
Lin, C. H., & Bioucas-Dias, J. M. (2019). An explicit and scene-adapted definition of convex self-similarity prior with application to unsupervised Sentinel-2 super-resolution. IEEE Transactions on Geoscience and Remote Sensing, 58(5), 3352-3365.
Lin, C. H., & Bioucas-Dias, J. M. (2020). Nonnegative blind source separation for ill-conditioned mixtures via John ellipsoid. IEEE Transactions on Neural Networks and Learning Systems, 32(5), 2209-2223.
Liu, Q., Liu, L., & Wang, Y. (2017). Unsupervised change detection for multispectral remote sensing images using random walks. Remote Sensing, 9(5), 438.
McFeeters, S. K. (1996). The use of the Normalized Difference Water Index (NDWI) in the delineation of open water features. International journal of remote sensing, 17(7), 1425-1432.
Rouse Jr, J. W., Haas, R. H., Deering, D. W., Schell, J. A., & Harlan, J. C. (1974). Monitoring the vernal advancement and retrogradation (green wave effect) of natural vegetation (No. E75-10354).
Roy, M., Ghosh, S., & Ghosh, A. (2014). A novel approach for change detection of remotely sensed images using semi-supervised multiple classifier system. Information Sciences, 269, 35-47.
Sambandham, V. T., Kirchheim, K., Ortmeier, F., & Mukhopadhaya, S. (2024). Deep learning-based harmonization and super-resolution of Landsat-8 and Sentinel-2 images. ISPRS Journal of Photogrammetry and Remote Sensing, 212, 274-288.
Sarker, I. H. (2021). Machine learning: Algorithms, real-world applications and research directions. SN computer science, 2(3), 160.
Verbesselt, J., Hyndman, R., Newnham, G., & Culvenor, D. (2010). Detecting trend and seasonal changes in satellite image time series. Remote sensing of Environment, 114(1), 106-115.
Volpi, M., Tuia, D., Bovolo, F., Kanevski, M., & Bruzzone, L. (2013). Supervised change detection in VHR images using contextual information and support vector machines. International Journal of Applied Earth Observation and Geoinformation, 20, 77-85.
Xu, H. (2006). Modification of normalised difference water index (NDWI) to enhance open water features in remotely sensed imagery. International journal of remote sensing, 27(14), 3025-3033.
Yuan, Y., Lv, H., & Lu, X. (2015). Semi-supervised change detection method for multi-temporal hyperspectral images. Neurocomputing, 148, 363-375.
Zhang, C., Yue, P., Tapete, D., Jiang, L., Shangguan, B., Huang, L., & Liu, G. (2020). A deeply supervised image fusion network for change detection in high resolution bi-temporal remote sensing images. ISPRS Journal of Photogrammetry and Remote Sensing, 166, 183-200.
Zhang, X., Yu, W., Pun, M. O., & Shi, W. (2023). Cross-domain landslide mapping from large-scale remote sensing images using prototype-guided domain-aware progressive representation learning. ISPRS Journal of Photogrammetry and Remote Sensing, 197, 1-17.
王嘉陽. (2021). 基於物件之隨機森林法於 UAV 航照影像之土地利用分類及變遷偵測. 國立臺灣大學土木工程學系學位論文, 2021, 1-73.
吳俊德. (1992). 應用衛星遙測於土地變遷探測方法之研究. 國立中央大學大氣物理研究所學位論文, 1-70
吳俊毅. (2017). 應用整合位移偵測演算法於臺灣西部河道變化之研究. 交通大學土木工程系所學位論文, 1-120.
張世駿, 江介倫, 吳進龍, 鄭克聲, & 陳增壽. (2006). 衛星遙測影像應用於地表變遷偵測之研究. 農業工程學報, 52(3), 21-31.
潘麗慧, 羅南璋, & 黃凱易. (2006). 多期衛星影像相對輻射糾正於臺中縣烏石坑地區變遷偵測之探討. 林業研究季刊, 28(1), 29-38.
鄧淑萍. (2010). 假設檢定及衛星遙測影像應用於地表覆蓋變遷偵測之研究. 國立臺灣大學森林環境暨資源學系學位論文, 2010, 1-122.
謝依達, 鍾玉龍, 廖晟淞, 余曜光, 鄧國禎, & 吳守從. (2011). 以變遷偵測技術探討高解析力數值航攝影像於森林火災自動製圖之應用. 航測及遙測學刊, 16(1), 11-22.