建立水稻田產量估算的創新方法：結合蒙地卡羅和物件區塊化的演算法

The Study of Estimation Production on Paddy Rice New Approach: Integration on Monte Carlo simulation and Regional Object-based Classification

萬絢(Shiuan Wan) ； 鄭育欣(Yu-Hsin Cheng) ； 李哲源(Che-yuan Li)

《航測及遙測學刊》 24卷3期 (2019/09) Pp. 183-196

https://doi.org/10.6574/JPRS.201909_24(3).0004

摘要

鑑於水稻為國內最重要的糧食作物，然推估其產量是困難工作，傳統研究水稻產量方式為，以坵塊在不同耕地之多時期之一系列影像圖片，利用各坵塊內之光譜植生指標隨時間的變化，進以影像差分分類法，先評估種植範圍，再結合地真實際產量，過去方法僅以迴歸方式進行產量推估，但比對不同年期之推估產量與實際產量，出現誤差大問題，對產量推估成果相當有限。本研究與中興測量合作開發，透過高光譜影像和正射影像先進行稻田影像區塊化作業（Regional Object-based Classification，ROC），針對每一個稻田區塊再利用亂數基礎分類法（entropy-based classification，EBC）合併蒙地卡羅法（Monte Carol method）的整合技術分析影像內容，計算每一個區塊的生產密度，推估稻田合理產量，對農業產量的推估精確度有研究貢獻。應用ROC＋蒙地卡羅法推估值誤差約在14.5%，標準差在1.24ton／ha。

關鍵字

影像判釋 ； 影像區塊化 ； 蒙地卡羅法

並列摘要

Whereas the paddy rice has played an important role on the life of human being, however, the estimation of its production is a difficult task. In tradition, the difference image classification on image data combine with levees is a major solution for their location and distribution. It includes the variation of each indicators of vegetation. On the other hand, the amounts of the relative error on the production of the paddy rice are quite large based on the developed regression equation. This study cooperated with Chun Hsing Survey applying the hyperspectral imagery data. The orthophotos are first used to perform the ROC (Regional Object-based Classification) and the entropy based classification (EBC) for classifying each rice levee of field. The Monte Carlo method to analyze the image content, calculate the production density of each block, estimate the reasonable yield of rice fields, and make an absolute contribution to the accuracy of agricultural yield estimation. The ROC+ Monte Carlo method is used to estimate the error of about 14.5% and the standard deviation is 1.24 ton/ha.

並列關鍵字

Image classification ； Regional Object-based Classification ； Monte Carlo method