使用雲林、彰化地區航空影像建立水稻判釋模型之研究;Establishing Rice Classification Model Using Aerial Images of Yunlin and Changhua Areas

NCU Institutional Repository > 資訊電機學院 > 軟體工程研究所 > 博碩士論文 > Item 987654321/95307

jsp.display-item.identifier=請使用永久網址來引用或連結此文件: http://ir.lib.ncu.edu.tw/handle/987654321/95307

题名:	使用雲林、彰化地區航空影像建立水稻判釋模型之研究;Establishing Rice Classification Model Using Aerial Images of Yunlin and Changhua Areas
作者:	高程昱;Kao, Cheng-Yu
贡献者:	軟體工程研究所
关键词:	坵塊分類;航攝影像;水稻判釋;parcel recognition;aerial images;rice classification
日期:	2024-08-22
上传时间:	2024-10-09 16:38:38 (UTC+8)
出版者:	國立中央大學
摘要:	稻米是台灣主要的糧食作物，在農產品中具有極高的產值。有效監控稻米有助於種植面積與產量的估算。傳統方法依賴於大量的人力調查，既耗時又昂貴。為了解決這一問題，現代技術如航空影像和機器學習被引入，以提高農業監控的效率和準確性。本研究利用農業工程研究中心（The Agricultural Engineering Research Center, AERC）提供的台灣航照影像，該影像由 DMC3 相機拍攝，無雲層遮蔽，且為標籤圖檔格式 (Tag Image File Format, TIFF)。此外，還使用了航照影像範圍內的農田地理資訊檔案。在過往需要花費長達五個月的時間來收集全臺灣的航照影像資料作為模型的訓練資料集，相當耗費時間及人力成本。因此，本研究的主要目標是僅使用雲林、彰化和嘉義地區資料集訓練稻米判釋模型，以此降低收集資料的時間成本，並且期望能達到與使用全台灣資料集訓練的模型有相同的準度及相似的 Kappa 系數。根據農業部統計，雲林、彰化和嘉義地區佔台灣稻米種植面積的47\%，且具有高度的作物多樣性，能代表台灣的農業景觀。收集這些地區的資料期間可以降低資料收集的時間成本，縮短至一個月，同時保持與使用全臺灣資料集相似的模型準確度。本研究驗證了在不降低判釋模型效能的前提下，減少資料收集成本的可行性，為政府機構在大規模農業監控和管理上提供了更經濟有效的解決方案。;Rice is a primary food crop in Taiwan, holding the highest production value among agricultural products. Effective monitoring of rice helps in estimating planting area and yield. Traditional methods rely on extensive manual surveys, which are time-consuming and costly. To address this issue, modern techniques such as aerial imaging and machine learning have been introduced to improve efficiency and accuracy in agricultural monitoring. This study utilizes TIFF(Tag Image File Format) format aerial images without cloud of Taiwan, provided by the AERC(Agricultural Engineering Research Center) and captured by the DMC3 camera. Additionally, geographic information files of the rice parcel within the scope of the aerial images are used. In the past, it took up to five months to collect data from all over Taiwan as a training data set for the model. The primary objective of this study is to determine whether a rice interpretation model trained using datasets exclusively from Yunlin, Changhua, and Chiayi regions can achieve a Kappa coefficient similar to a model trained with data from all over Taiwan. According to statistics from the Ministry of Agriculture, the Yunlin, Changhua and Chiayi regions account for 47\% of Taiwan′s rice planting area and have a high degree of crop diversity, representing Taiwan′s agricultural landscape. By focusing on these regions, the data collection period can be reduced to one month, significantly lowering the cost of data collection while maintaining model performance similar to that of using a nationwide dataset. The results of this study will provide insights into the feasibility of reducing data collection costs without compromising the performance of the interpretation model. This approach makes large-scale agricultural monitoring and management more feasible for government agencies.
显示于类别:	[軟體工程研究所 ] 博碩士論文

文件中的档案:

档案	描述	大小	格式	浏览次数
index.html		0Kb	HTML	38	检视/开启

在NCUIR中所有的数据项都受到原著作权保护.

社群 sharing

数据加载中.....