ENGLISH

28卷/2期

28卷/2期

華藝線上圖書館

Pages:

63-81

論文名稱

自監督式深度學習影像匹配應用於福衛光學衛星影像幾何校正

Title

Self-supervised Deep-learning-based Image Matching for FORMOSAT Optical Satellite Image Orthorectification

作者

吳菉、張雅筑、林柏毅、林昭宏、曾義星、張立雨、張莉雪、李彥玲

Author

Lu Wu, Ya-Chu Chang, Bo-Yi Lin, Chao-Hung Lin, Yi-Hsing Tseng, Li-Yu Chang, Li-Hsueh Chang, Yen-Ling Lee

中文摘要

標準幾何校正流程在獲取控制點上花費大量人力及時間,為使衛星影像呈現精確的幾何成像,且提升獲取衛星影像之效率,本研究提出一新穎的自動化衛星影像幾何校正流程。藉由自監督深度學習影像匹配演算法及影像匹配策略,於衛星影像中自動化獲取更多的顯著特徵作為影像控制點,使得衛星影像幾何校正流程更穩健且便捷。實驗結果表明,自動化幾何校正流程不僅具穩定性且具適應性,幾何校正結果在福衛五號2米空間解析度下誤差約為2至4像元。

Abstract

The standard orthorectification process takes a lot of manpower and time to obtain control points. To correctly represent the image geometry on satellite images and improve the efficiency of satellite image orthorectification, a novel method for automatic satellite image orthorectification is proposed. In this study, a robust satellite image matching process is processed to obtain image control points, which adopted. Different from traditional labor-intensive methods, a novel image matching method is adopted to find image control points both on target images and an orthorectified reference image, which is adopted self-supervised deep learning image matching algorithm. This strategy makes the ortho-rectification process become automatic, robust, and attempts to distinguish more salient features than traditional methods in satellite images. The experimental results show that the automatic orthorectification process is not only stable but also adaptable. The quantity assessment is performed using root mean square error, and the accuracy of satellite image orthorectification result is 2 to 4 pixels under the 2-meter spatial resolution of FORMOSAT-5 images.

關鍵字

幾何校正、有理函數模型、深度學習、基於特徵的影像匹配、光學衛星影像

Keywords

Optical Satellite Image, Image Orthorectification, Rational Function Model, Deep Learning, Feature-Based Image Matching

附件檔名

華芸線上圖書館

https://www.airitilibrary.com/Publication/Index?DocID=10218661-N202306090016-00001

備註說明

N / A

Pages:

83-102

論文名稱

應用Taiwan Data Cube於多時期衛星影像之崩塌地分析

Title

Landslide Area Analysis with Temporal Satellite Images by Using Taiwan Data Cube

作者

黃鈺涵、曾義星

Author

Yu-Han Huang, Yi-Hsing Tseng

中文摘要

臺灣受地質和地理環境影響,崩塌災害頻繁發生,然而遙感探測的特點對於監測分析環境敏感地至關重要。本研究基於衛星影像時間序列概念,透過 Taiwan Data Cube 平台建立環境敏感地監測模型,進行常態化差異植生指標計算以及最大似然分類法,在時間面向中,可以觀察出該地環境的長期趨勢與變化;在空間面向中,則可以找出崩塌地識別之門檻值,藉由影像差分法判斷新生崩塌地的面積變化以及位置。本研究以六龜區及梅山明隧道作為試驗區,經由建立衛星影像時間序列,達到解析區域時空變化的目標,讓地理空間資訊應用更加全面。

Abstract

Due to geographical and geological factors, typhoons hit Taiwan frequently. It may cause serious disasters, so monitoring the condition of landslide area is critical. Based on the concept of satellite image time series (SITS), this study is to establish a long-term monitoring model by Normalized Difference Vegetation Index (NDVI) calculations and Maximum Likelihood Classification (MLC). For the temporal view, long-term changes in geologically sensitive area can be found out. For the spatial view, the interpretation standard can also be identified, and the location of the new landslide can also be distinguished by “Image Differencing”. This study selects “Liouguei District” and “Meishan open-cut tunnel” as the regions of interest. Through the establishment of satellite image time series, the goal of analyzing regional temporal and spatial changes is achieved, making the application of geospatial information more comprehensive.

關鍵字

Taiwan Data Cube、衛星影像時間序列、常態化差異植生指標、最大似然分類

Keywords

Taiwan Data Cube (TWDC), Satellite Image Time Series (SITS), Normalized Difference Vegetation Index (NDVI), Maximum Likelihood Classification (MLC)

附件檔名

華芸線上圖書館

https://www.airitilibrary.com/Publication/Index?DocID=10218661-N202306090016-00002

備註說明

N / A

Pages:

103-123

論文名稱

應用無人空中載具影像調查外來入侵植物小花蔓澤蘭

Title

Using Unmanned Aerial Vehicle Images to Survey Invasive Plants of Mikania micrantha

作者

江秉鴻、陳建璋、魏浚紘

Author

Ping-Hung Chiang, Jan-Chang Chen, Chun-Hung Wei

中文摘要

近30年來我國遭受小花蔓澤蘭 (Mikania micrantha) 之入侵,為有效且快速了解其分佈狀況,本研究透過遙感探測 (Remote Sensing, RS),以無人空中載具 (Unmanned Aerial Vehicle, UAV)影像進行像元影像分析 (Pixel-based Image Analysis) 及物件導向影像分析(Object-based Image Analysis),以人工判釋建立小花蔓澤蘭UAV影像判釋準則,透過該準則進行監督性分類。由不同空間解析度分類結果可得知,最佳分類空間解析度約為10-15 cm,且須於花季進行拍攝。若空間解析度小於10 cm,易形成椒鹽效應。建議先經過物件導向處理再進行監督性分類,或使高解析度 (<10 cm) 影像融合為一新低解析度 (10-15 cm) 影像進行分類。綜合上述以RGB波段UAV調查小花蔓澤蘭分佈狀態是可行的。

Abstract

Over the past 30 years, Taiwan has been invaded by Mikania micrantha. In order to effectively and quickly understand its distribution, this study conducted pixel-based image analysis and object-based image analysis with unmanned aerial vehicle images through remote sensing detection, and detected Mikania micrantha with artificial interpretation. Criteria for image interpretation, according to which the supervised classification is carried out. According to the classification results of different spatial resolutions, the best classification spatial resolution is about 10-15 cm and the images must be taken during the flowering season. If the spatial resolution is less than 10 cm, the salt and pepper effect is easy to form. It is recommended to perform supervised classification after object based image analysis or to fuse high resolution images (<10 cm) into a new low resolution image (10-15 cm) for classification. On this basis, the distribution of Mikania micrantha can be studied with RGB UAVs.

關鍵字

外來入侵種、小花蔓澤蘭、遙感探測、無人空中載具

Keywords

Invasive Alien Plants, Mikania micrantha, Remote Sensing, Unmanned Aerial Vehicle

附件檔名

華芸線上圖書館

https://www.airitilibrary.com/Publication/Index?DocID=10218661-N202306090016-00003

備註說明

N / A

Pages:

125-139

論文名稱

應用卷積神經網絡於自動化森林覆蓋型辨識工作

Title

The Application of Automatic Forest Cover Detection Tool with Convolutional Neural Network

作者

王禹翔、吳笙緯、魏擇壹、鄭錦桐、鍾智昕、吳淑華、鄧國楨、黃宗仁

Author

Yu-Hsiang Wang, Sheng-Wei Wu, Ze-Ui Wei, Ching-Tung Cheng, Chih-Hsin Chung, Shwu-Hwa Wu, Juo-Chen Teng, Tsung-Jen Huang

中文摘要

臺灣森林佔全島面積六成,高解析航照是世界各國量化森林資源的重要途徑,影像辨識與圖資編修的需求源源不絕。本研究透過卷積神經網絡 (Convolutional Neural Network, CNN) 建立自動化森林覆蓋型辨識模型,於3,248幅DMC影像中,成功使用16種網格特徵設計單一樹種分類模型,實現19種臺灣常見造林樹種之判釋,研究結果獲得74.6%的整體精度評價,並透過可解釋性的特徵分析發現近紅外光波段、高程與坡向有較明顯的貢獻價值。面對國內的森林資源調查資料與林型圖資清查,本研究開發了「森林覆蓋型辨識工具」,產出的向量圖資可供影像判釋人員使用,以期提升檢訂調查圖層編修之品質。

Abstract

Forests in Taiwan cover approximately 60% of the island. High-resolution aerial imagery is an important data source for quantifying forest resources in countries around the world, so there is a continuous demand for image recognition and map editing. In this study, an automated forest cover classification model was developed using a Convolutional Neural Network (CNN). A single classification model for tree species was designed using 16 grid features, and successfully identified 19 common afforestation tree species in Taiwan from 3,248 DMC images. The research achieved an overall accuracy of 74.6% and feature analysis revealed that near-infrared band, digital elevation model, and aspect had significant contributions. In order to support forest resource investigation and map editing in Taiwan, a "Forest Cover Recognition Tool" was also developed. The vector data produced by this tool can be used by image interpreters to improve the quality of map editing and revision.

關鍵字

卷積神經網絡、航攝影像辨識、森林資源調查、檢訂調查作業

Keywords

Convolutional Neural Network, Aerial Image Classification, Forest Resources Inventory, Revision Investigation

附件檔名

華芸線上圖書館

https://www.airitilibrary.com/Publication/Index?DocID=10218661-N202306090016-00004

備註說明

N / A

更多活動學刊