由於貴會林務局農林航空測量所之航照影像，主要以可視化為主，為了減少航照相幅間差異對深度學習AI判釋精度影響，以高解析度衛星照片之地表反射率為基準，針對航照影像進行校正後再由貴會資訊中心開發的深度學習AI辨釋模型應用。本專案規畫以未經鑲嵌的正射航攝影像，藉由Pleiades衛照參考影像，建立航照影像與地表反射率間，各波段(紅光、綠光、藍光以及近紅外)通用的轉換式，並以此轉換式應用至航照影像自動化校正程序，以符合農航所新購航遙測飛機啟用後之影像校正量能需求。並運用資料立方(Data Cube)技術管理校正影像，提高校正影像後續應用效能，支援AI作物判釋模組存取與運算。 ;In order to get accurate result from the Artificial Intelligence (AI) application on agricultural purpose, it is essential to calibrate the radiations (surface reflectivity) before taking aerial images (DMC) as dataset. Our team take high resolution Pleiades satellite images to calibrate orthographic DMC images including red band, green band, blue band and near infrared band. By the results of radiative calibration, we developed a universal formula for transformation between digital number of DMC images and surface reflectivity in each band. With the formula, we can establish an auto-progresses of radiative calibration of the DMC image. Additionally, We also apply the data cube platform to integrate and manage the radiative calibrated DMC images for the application of AI.
