以支援向量迴歸在水稻田產量高解析度光譜影像研究

A Support Vector Regression Approach on the Estimation of Paddy Rice of Image Classification

有鑑於水稻在人類生活中扮演著重要的角色，估計其產量仍是一項極具挑戰性的任務。傳統上，差異影像分類結合堤防是識別水田位置與分佈的主要方法。此方法可計算植被指標的變化。然而，使用迴歸方程式估計水稻產量的相對誤差往往相當大。本研究提出一種創新的方法，利用區域物件導向分類（ROC）技術與支援向量迴歸（SVR）產生一系列模擬的生產密度。與使用原始資料分析的平行研究相比，本研究分析了影響水稻核心影像波段因素的頻譜範圍。因此，可以產生生產誤差之減少。此方法有助於提高農業價值與效率。SVR的計算準確度約為4.2%，而RSES（粗集合開發平台系統）選取核心因子後約為3.4%。

In view of paddy rice plays an important role in human life, estimating its production remains a challenging task. Traditionally, difference image classification combined with levees has been a major approach for identifying the location and distribution of paddy fields. This method accounts for variations in vegetation indicators. However, the relative error in estimating paddy rice production using regression equations tends to be quite large. This study proposes an innovative method using the Regional Object-Oriented Classification (ROC) technique and Support Vector Regression (SVR) to generate a series of simulations for production density. The spectral range of core factors affecting paddy rice is analyzed, compared to the parallel study of using raw data analysis. Consequently, the prediction of production errors can be reduced. This method contributes to improving agricultural value and efficiency. The computational accuracy of SVR is approximately 4.2%, while the RSES(Rough Set Exploration System) core factor achieves about 3.4%.