深度學習與太空科技應用

臺北科技大學 115-1 學期課程，課號 366257。

課程大綱

This course teaches how deep learning is applied to satellite and space-technology data, from the measurement physics of a satellite image to a model running on an onboard computer. Students learn the data contract of Earth-observation datasets (chips, labels, honest splits, metrics), convolutional and transformer models, geospatial foundation models (Prithvi-EO-2.0, Clay, TerraMind), model compression for payload hardware (pruning, quantisation, distillation) and onboard autonomy. Every technique is introduced through a real mission or product (ESA Φsat-1/2, Planet Pelican, FORMOSAT-8A, JPL Dynamic Targeting). Taught in English with open data (Sentinel-1/2, Landsat, Planet open datasets) and open models; the final project is an individual model demonstrated live on an ASV Lab edge board. 本課程講授深度學習於衛星與太空科技資料之應用：從衛星影像的量測物理出發，到模型於機上電腦實際執行。內容涵蓋地球觀測資料集契約（切片、標註、誠實的資料分割與評估指標）、卷積網路與 Transformer、地理空間基礎模型（Prithvi-EO-2.0、Clay、TerraMind）、酬載硬體之模型壓縮（剪枝、量化、蒸餾）與機上自主。每項技術皆以真實任務或產品引入（ESA Φsat-1/2、Planet Pelican、福衛八號、JPL Dynamic Targeting）。全英語授課，使用開放資料與開放模型；期末為個人專題，於 ASV Lab 邊緣運算板上實機展示。

課程進度

W1 9/7 — Course introduction; remote-sensing story; space technology fundamentals (orbits, sensors, GSD). HW1: one-page EO problem statement. 課程介紹；遙測與太空科技基礎。 W2 9/14 — AI applications in space; satellite data as measurement: data cube, bands, GeoTIFF, radiometry. Lab 0. 太空 AI 應用；衛星資料即量測。 W3 9/21 — Classical image processing; the dataset contract: chips, labels, splits, metrics, mission cost. Lab 1. 傳統影像處理；資料集契約。 W4 9/28 — Teachers’ Day, no class. 教師節放假。 W5 10/5 — Object detection and tracking on aerial/UAV imagery (recorded lecture + TA workshop, UAVDT). Lab 2. 航空/無人機影像物件偵測與追蹤。 W6 10/12 — Neural networks and CNNs for Earth observation; U-Net; convolution quiz. 神經網路與卷積網路；U-Net。 W7 10/19 — Efficient CNNs for satellite payloads: depthwise-separable convolution, regularisation, residual and attention blocks, IoU losses; missions Φsat-2 and Pelican. Project topic approval (individual). 衛星酬載之高效 CNN；專題題目核准（個人）。 W8 10/26 — Retrocession Day makeup, no class (data collection week). 光復節補假（資料蒐集週）。 W9 11/2 — Transformers and geospatial foundation models (Prithvi-EO-2.0, Clay, TerraMind). Lab 4: fine-tune Prithvi on Sen1Floods11. Transformer 與地理空間基礎模型。 W10 11/9 — Anniversary makeup, no class. 校慶補假。 W11 11/16 — Adaptation and honest evaluation: probing, LoRA, full fine-tuning; held-out geography, time, sensor; midterm review. Lab 5. 模型調適與誠實評估；期中複習。 W12 11/23 — Midterm exam (written, sessions 1–8). 期中考（筆試）。 W13 11/30 — Generative AI for Earth observation: super-resolution, cloud removal, diffusion, validation. Trained model due. 生成式 AI 於地球觀測；完成模型訓練。 W14 12/7 — Onboard AI: ground vs flight, pruning, quantisation, distillation, ONNX/OpenVINO export, latency on the edge board. Lab 6. 機上 AI：模型壓縮與邊緣板延遲量測。 W15 12/14 — Onboard autonomy and mission agents: dynamic targeting, Φsat-2, fault detection, agentic mission planning. Demo dry run. 機上自主與任務代理；預演。 W16 12/21 — Final project demonstration on ASV Lab edge boards and oral presentation of the report; instructor comments returned (replaces the final exam). 期末專題實機展示與口頭報告，教師回饋（取代期末考）。

評量方式

出席、課堂參與及討論 / Attendance, Class Participation, and Discussion: 10% 作業與實作練習 / Assignments and Practical Exercises: 20% 期中考試 / Midterm Examination: 20% 專題提案與進度報告 / Project Proposal and Progress Report: 10% 期末專題報告 / Final Project Report: 25% 期末口頭簡報 / Final Presentation: 15%

教材

【遵守智慧財產權觀念，請使用正版教科書，不得使用非法影印教科書】 使用外文原文書：是 教師自編講義、課堂簡報及線上教學資源 Instructor-prepared lecture notes, presentation slides, and online learning materials. Ian Goodfellow, Yoshua Bengio, and Aaron Courville, Deep Learning, MIT Press, 2016. François Chollet, Deep Learning with Python, 2nd Edition, Manning Publications, 2021. Richard Szeliski, Computer Vision: Algorithms and Applications, 2nd Edition, Springer, 2022. John A. Richards, Remote Sensing Digital Image Analysis, 6th Edition, Springer, 2022. Thomas M. Lillesand, Ralph W. Kiefer, and Jonathan W. Chipman, Remote Sensing and Image Interpretation, 7th Edition, Wiley, 2015. NASA、ESA、USGS、TASA及其他太空與遙測機構所提供的開放資料、技術文件與案例研究。 Open datasets, technical documentation, and case studies provided by NASA, ESA, USGS, TASA, and other space and remote-sensing organizations. 課程相關之最新期刊論文、國際研討會論文、開源程式碼及衛星影像資料集。 Selected journal articles, conference papers, open-source software, and satellite-imagery datasets related to the course.