稻米的插秧時間、種植面積與產量等，為政府制定農糧政策的重要參考指標，而使用多時期遙測影像可以用於偵測稻作插秧時間與辨識稻作分布。本研究利用SPOT影像的常態化植生指標(Normalized Difference Index，NDVI) 在時間序列上的變化，來偵測水稻的插秧日期與辨識二期稻作之種植位置。然而衛星的時間序列資料容易受到雲層與大氣狀態等因素所干擾，傳統上常使用小波分析濾除信號中高頻的雜訊，但使用小波轉換，必須選擇合適的小波函數，往往需要經驗與不斷的試驗，所以本研究提出經驗模態分解法（Empirical Mode Decomposition, EMD），濾除影響時序資料中的高頻雜訊，在時序雜訊濾除後，利用稻米時序變化資訊偵測稻作插秧日與分類稻米。本研究包含四個部分：(1) 從SPOT影像萃取水稻的NDVI；(2)分別使用經驗模態分解法與小波分解濾除NDVI時間序列的高頻雜訊；(3) 利用去除雜訊後的NDVI時間序列資料求取稻作的插秧時間；(4) 結合相關係數和符號檢定(Sign Test)的方法，利用時間序列資料偵測二期稻作之分布。實驗成果顯示，使用EMD的過濾成果偵測水稻田插秧時間與辨認二期稻作的種植範圍比使用小波分析成果更為穩定且準確。而且經過EMD濾除時序雜訊後分類二期稻作，三個測試區分類的整體精度皆在85%以上，且Kappa值皆大於70%，顯示EMD濾除雜訊後所進行的分類，可以提供不錯的分類成果。 關鍵字：SPOT、水稻、常態化植生指標、經驗模態分解法 The rice planting date and rice distribution are important information for the agriculture and food policy of the government. Traditionally, the investigation of the rice planting date and rice distribution cost a lot and is a time-consuming work. The use of SPOT NDVI time-series data for detecting the planting date and ditribution of the paddy rice is proposed in this study. However, the time-series satellite data are easily contaminated by noises such as cloud cover and atmospheric conditions. Conventionally, wavelet analysis is commonly used to reduce the high frequency noises in the time-series data. However, the selection of the proper mother wavelet and its parameters usually affects the success of reducing the noises of the time series signal. In this study, a method called Empirical Mode Decomposition (EMD) is proposed to reduce the high frequency noises in SPOT NDVI time-series data. Then by detecting and analyzing the local minimal points on the smooth NDVI time series profiles can provide the necessary information about the rice planting date. A classification method is also developed by using correlation and sign-test to discriminate double rice crops form NDVI filtering data. The method of this study basically includes four main parts: (1) constructing Normalized Difference Vegetation Index (NDVI) time-series data; (2) EMD is proposed to reduce the noise in a year-long SPOT images and retrieve the time variation of paddy rice; (3) detecting the local minimum points from NDVI time-series data of the rice pixels as rice planting dates; (4) using correlation and sign-test to discriminate rice crops. SPOT data in Chihshang, Taibao and Shinwu (2005) are used to test the proposed method. The experiment indicates EMD provides more stable results than wavelet analysis because EMD is insensitive to pre-determined parameters. Moreover, the results show that the time-series data filtered by using EMD method to estimate rice planting date and to detect double rice fields are more accuracy than that by using wavelet analysis. Keywords: SPOT, Paddy rice, NDVI, EMD
