資源・素材 & EARTH 2017(札幌)

講演情報(2017年8月24日付)

若手ポスター発表(Poster:MMIJ Students and Young Researchers)

資源開発技術(Mining technologies)

2017年9月26日(火) 15:30 〜 17:30 ポスター会場2 N301 (N棟3階/Fl.3.,Build. N)

15:30 〜 17:30

[PY2-89] DEVELOPMENT OF A DIFFERENTIATION AND IDENTIFICATION SYSTEM FOR IGNEOUS ROCKS USING HYPER-SPECTRAL IMAGES AND A CONVOLUTIONAL NEURAL NETWORK (CNN) SYSTEM

○Brian Bino SINAICE1, Youhei Kawamura1, Takeshi Shibuya2, Jo Sasaki1, Hibiki Yoshimoto1, Yutaka Ito1, Shinji Utsuki3 (1. Akita University, 2. University of Tsukuba, 3. Hazama Ando Corporation)

キーワード:Hyper spectral Image analysis, Convolution Neural Network

Rock identification has a tendency of posing a challenge to most especially when a certain species of rock may or may not contain a mineral constituent typically associated with it. It is for this reason that this study aims at combining two modern technologies to interpret and identify rocks; ‘Hyper-spectral image analysis’ and ‘Convolution Neural Network (CNN) Deep Learning’ as a way of differentiating one rock from the next, simply by quantifying how these rocks respond across the electromagnetic spectrum and finally running the data in a CNN software. Hyper-spectral image analysis involves the understanding of image data generated by a hyper-spectral camera; which in this study was used to acquire spectral images of six different kinds of igneous rock samples from ‘Akita mining museum’, and splitting each rock’s image pixel by pixel then quantifying how each of these pixels responds across the electromagnetic spectrum. The second step was to run numerical data acquired from the hyper-spectral camera in a CNN software, this is a deep learning computer system which allows one to first feed and run known data (learning data), which in this case was the previously extracted hyper-spectral strengths of each rock. The system calculates and quantifies similarities and differences of the rocks, and the results would in future aid in identification of the rock should the same rock(s) be encountered yet again. Test data was then run in order to test the viability of the system, results show a positive outcome as the test data shows higher accuracy percentages, which means the system recognized the rock which might have previously been a mystery. In conclusion, it can be said that the use of hyper-spectral images and CNN deep learning are a viable way in which rocks can be identified even without prior knowledge of the rock.

講演PDFファイルダウンロードパスワード認証

講演集に収録された講演PDFファイルのダウンロードにはパスワードが必要です。

現在有効なパスワードは、[資源・素材学会会員専用パスワード]です。
※[資源・素材学会会員専用パスワード]は【会員マイページ】にてご確認ください。(毎年1月に変更いたします。)

[資源・素材学会会員専用パスワード]を入力してください

パスワード