數位多頻譜影像技術之研究

Abstract

本論文研究為數位多頻譜影像技術之研究，針對過去實驗室已成功開發的數位多頻譜影像系統進行改良。首先，為了提高系統的穩定性，我們針對演算法進行修正，提出權重主成分分析法，選擇性地控制資料庫內各個頻譜資訊對於組成主成分的影響程度，因此可重建出更接近真實的頻譜。接著，為了改善系統的封閉性，我們提出新的實驗架構，模擬一開放式的環境，透過模擬環境光源來探討如何降低環境光源可能會造成的影響；同時建立色彩校正數學模型，並以實驗來驗證此模型之正確性。最後，我們將多頻譜影像技術應用於蘋果腐敗之檢測，並建立一套能辨別蘋果好壞的系統。

In this study, we focus on improving the multi-spectral imaging system which has been successfully developed. Firstly, in order to improve the stability of the system, we apply a new technique called weighted principal component analysis (wPCA). Implementation of this method provides the ability in selection of extracted principal eigenvectors. The result shows the reconstructed spectra based on wPCA are significantly improved in comparison to those obtained from the standard PCA. Then, we propose a new experimental framework to simulate an open environment. We establish a mathematical model by giving a calibration target and controllable ambient light. The experimental result shows that the average colorimetric errors and the RMS errors are decreased by 66.54% and 47.45%. Lastly, we utilize the multi-spectral imaging technique to detect bruises on apples and also build a system that can differentiate between good apples and bad apples. The identification rate is up to 87%.