農業影像二元分類：坵塊分離的檢測;Binary Classification of Agricultural Images: Detection of Parcel Separation

NCUIR > College of Electrical Engineering & Computer Science > Graduate Institute of Computer Science and Information Engineering > Electronic Thesis & Dissertation > Item 987654321/93402

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/93402

Title:	農業影像二元分類：坵塊分離的檢測;Binary Classification of Agricultural Images: Detection of Parcel Separation
Authors:	張淯淞;Chang, Yu-Sung
Contributors:	資訊工程學系
Keywords:	坵塊變動檢測;坵塊分類;VGG16-UNet;NetVLAD;CBAM;Parcel Change Detection;Parcel Classification;VGG16-UNet;NetVLAD;CBAM
Date:	2023-08-03
Issue Date:	2024-09-19 16:57:15 (UTC+8)
Publisher:	國立中央大學
Abstract:	每年政府會使用含有農作物標記的坵塊向量圖（簡稱坵塊圖），對坵塊的資訊進行管理。但坵塊可能會隨著時間有所變化，有部份的坵塊會在新年度發生分離的情況，這時就需要對該坵塊所對應的坵塊向量進行修改。不過，出動人力查找坵塊分離是一件曠日費時的任務，因此使用 AI 分類器自動找出新年度航照圖中的坵塊分離儼然是一個重要的議題。過去坵塊變異的檢測研究多是使用影像相似度比對的方式來了解同一個坵塊是否有變化，而本研究則是提出一套產生 “仿坵塊分離使用情形” 合成資料的方法，與兩個判斷坵塊是否分離的影像二分類器。我們使用 “仿坵塊分離使用情形” 的合成資料訓練我們提出的影像二分類器，使其學會坵塊分離的影像樣態，並且使用訓練好的分類器來檢測真實航照中的坵塊分離。此外，本研究的貢獻在於透過實驗，我們驗證了真實測試資料集當中存在的幾種特殊的坵塊影像，會對我們所提出之影像二分類器造成判釋結果的影響。這個發現可提供後續研究在準備模型訓練資料集上的指引，未來的研究能遵循這個指引，針對特殊的坵塊影像作資料集的準備，並進一步探索判釋特殊坵塊影像的方法。;Every year the government uses a parcel vector(aka polygon) shapefile with crop labels to manage the information of parcels. However since parcels may change over time, some parcels will be separated in the new year, and the parcel vector corresponding to the parcel needs to be modified. Finding the parcel separations manually is a time-consuming task, so it is an important issue to use AI classifiers to automatically detect the separation of parcel in the new year’s aerial images. In the past, most researches on the detection of parcel change used image comparisons to check whether the same parcel changes. In this paper, we propose a way to synthesize the ”Parcel separation pseudo data” and two binary classifier for predicting Parcel separation. We use the ”Parcel separation pseudo data” to train the several proposed binary classifiers, and use the trained classifier to detect Parcel separation in real aerial image. Besides, the contribution of this study also include that there are some kinds of special parcel image in the real test dataset affect the accuracy of our proposed classifier. This finding can provide a valuable guidance for future research in preparing training datasets. Future research can follow this guideline to prepare datasets for special parcel images, and further explore the method of interpreting special parcel images.
Appears in Collections:	[Graduate Institute of Computer Science and Information Engineering] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	44	View/Open

社群 sharing

Loading...