Development of Hyperspectral Database and Web Based Classifying System for Rock Type Identification

In geological surveys, it is necessary to employ geological experts in determining rock samples correctly, this is however very costly and time consuming. Sinaice et al. (2020) demonstrated how employing hyperspectral data of igneous rocks and machine learning can be used in classifying rocks without any knowledge or background in geology. Although machine learning is able to achieve high accuracy with a sufficient amount of data (teaching data), there are no existing big-data-sets of rock hyperspectral data. In order to improve the accuracy and robustness of our machine learning model, we collected a large amount of spectral data of various types of rocks as an attempt to solve the machine learning bottleneck by creating web application that is able to share hyperspectral data among users. We created a web application that allows users to upload hyperspectral image data of rocks taken by various users, determine the rock type using our trained machine learning model, and subsequently browse the spectral database of the Mining Museum of Akita University. The machine learning model is capable of automatically improving its accuracy as data is uploaded by various users. Hence, each user is able to use the database and the determination function. By using this web application, it is possible to collect spectral data from a wide variety of rocks from various users of the web application, thereby improving the accuracy and robustness of rock type determination using hyperspectral data and machine learning, hence solving the aforementioned difficulties borne by researchers in previous studies.