機器學習搭配影像增強應用於自動化岩心影像辨識之研究

Abstract

地質科學領域的發展基本始於岩性的劃分。人們為探索地底下的奧秘而開始鑽取岩心，一旦開始劃分岩性，就會在岩性判釋上面臨質和量之間的權衡。傳統的岩心判釋方法依賴於觀察者們對岩心的描述及經驗的積累，然而這些判斷過程並不易於再現及理解。作為一種新興技術，機器學習擁有強大的學習能力，潛力十足，可以開發自動化的岩性辨識，以助於日後的地質科學研究。 本研究使用來自德國東弗里西亞瓦登海（Norderney Island）的高解析度岩心影像，涵蓋了多種沉積物相，這些沉積物相已通過傳統分析識別：冰碛岩、風積/沖積、土壤、泥炭、潟湖、泥坪、砂灘、河道填充和海灘前砂地。主要目的是利用機器學習來模仿沉積學家的觀察行為，開發一種影像機器學習分類模型。 本研究的最終目標是建立一個影像的自動化岩性辨識模型，並以影像增強為輔。期許模型不僅能在此研究區域有良好的岩性預測，更可以在未來應用於其他地區的岩心影像，擴展影像機器學習系統的功能。 總而言之，希冀自動化岩性辨識模型能利用機器學習的底蘊，並成為區域地質研究的基石，並在前所未有的深探中發揮關鍵的作用。

The development of the geoscience field is based on the delineation of sediment facies, but there is a trade-off between quality and quantity. The conventional method of facies discrimination relies on observer-dependent sedimentological descriptions, which are not always easy to reproduce and understand. As a new technology, machine learning provides an important possibility to develop a new type of automatic sediment facies classification that can aid in conducting geoscience research with more consistent quality. This approach uses high-resolution photos of sediment cores from Norderney Island in the East Frisian Wadden Sea, Germany, covering various sediment facies that have been identified through conventional facies analysis, such as moraine, eolian/fluvial, soil, peat, lagoon, sand flat, channel fill, and beach-foreshore. The main objective is to use machine learning to mimic the observational behavior of sedimentologists and develop an image-based machine learning classification model. The ultimate goal is to establish a model that can be applied to sediment cores in other regions and expand the capabilities of the image-based machine learning system with image enhancement. This model is also expected to be a cornerstone for regional geological mapping and play a crucial role in exploring unprecedented depths.