Please use this identifier to cite or link to this item:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/84376
Title: | 以無人機多光譜影像分析增進田間水稻穗外表型態於育種上的利用 Enhanced Paddy Rice Panicle Phenotyping for Breeding Application Based on UAV Multi-Spectral Image Analysis |
Authors: | 陳彥伯 Yen-Po Chen |
Advisor: | 盧虎生 Huu-Sheng Lur |
Keyword: | 穗型,多光譜,無人機,線性判別分析,KNN,XGBoost, Panicle type,multispectral images,UAV,LDA,KNN,XGBoost, |
Publication Year : | 2022 |
Degree: | 碩士 |
Abstract: | 水稻是台灣最主要的糧食作物,穗外表型對於提高產量和穀物品質的育種研究十分重要,傳統調查外表型的方法仰賴人工採樣,需要花費較多時間。本研究嘗試建立一個藉由無人機拍攝及多光譜影像分析來分辨田間水稻穗型的方法。本研究使用無人機拍攝多光譜影像,以QGIS提取多光譜影像中不同穗型的植株光譜平均數值,接著以逐步迴歸方法初步篩選對於分類較有貢獻的光譜與植生指數組合,並比較線性判別分析、KNN與XGBoost的分類結果。水稻穗型分為:0 (未抽穗);1 (直立);2 (半直立);3 (下垂);4 (極度下垂)。本研究找到四個對於穗型分類較有潛力的植生指數:Clgreen、Clrededge、NDVSI、GSAVI,並發現穗型與收穫時間有較高的關聯性。線性判別分析的分類結果顯示,準確率最高為 62.1%; KNN模型的分類結果顯示,在訓練集中準確率最高的為64.18%,在測試集中準確率最高的為61.94%; XGBoost模型的分類結果顯示,在訓練集中準確率最高的為62.13%,在測試集中準確率最高的為60.45%。研究結果顯示,藉由無人機拍攝及多光譜影像分析,可以分辨田間水稻穗型,用以判斷水稻成熟情形,有效減少育種研究時水稻性狀調查人力。 Rice (Oryza sativa L.) is the main food crop in Taiwan, and the panicle phenotype is essential for breeding research in terms of improving yield and grain quality. In this study, we propose a method to recognize the panicle type (type 0: not heading, type 1: erect, type 2: half erect, type 3: sagging, type 4: extremely sagging) by using multispectral images collected by unmanned aerial vehicles (UAV). We used UAV to acquire multispectral images from field and QGIS to extract the average values of plant spectra of different panicle types from the multi-spectral images, then used stepwise regression to initially screen the optimal combination between spectra bands and vegetation indices that are more beneficial for the following analysis, classification. We compared the classification results of linear discriminant analysis, KNN and XGBoost. Four vegetation indices for rice panicle types classification were found in stepwise regression Clgreen, Clrededge, NDVSI, GSAVI. We found high correlation between panicle type and harvest time. The highest classification accuracy of discriminant analysis was 62.1%. The similar results can be seen in KNN model (64.18% in the training set and 61.94% in the test set) and XGBoost (62.13% in the training set and 60.45% in the test set). The results show that we can recognize rice panicle type and growth stage with a UAV-based, multi-spectral airborne imagery analysis, which effectively save the human resources for paddy rice agronomic characteristics investigation of breeding research. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/84376 |
DOI: | 10.6342/NTU202200663 |
Fulltext Rights: | 同意授權(限校園內公開) |
metadata.dc.date.embargo-lift: | 2025-07-01 |
Appears in Collections: | 農藝學系 |
Files in This Item:
File | Size | Format | |
---|---|---|---|
ntu-110-2.pdf Restricted Access | 6.71 MB | Adobe PDF | View/Open |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.