WEKO3
アイテム
Study on the Development of a Rock Identification System Using Artificial Intelligence and Spectral Imaging
https://doi.org/10.20569/00006112
https://doi.org/10.20569/000061123828d6f3-366e-442f-ba9b-66ef5856973c
| 名前 / ファイル | ライセンス | アクション |
|---|---|---|
内容要旨及び審査結果要旨 (198.4 kB)
|
|
|
|
本文 (8.6 MB)
|
|
| Item type | 学位論文 / Thesis or Dissertation(1) | |||||||
|---|---|---|---|---|---|---|---|---|
| 公開日 | 2022-12-06 | |||||||
| タイトル | ||||||||
| タイトル | Study on the Development of a Rock Identification System Using Artificial Intelligence and Spectral Imaging | |||||||
| 言語 | ||||||||
| 言語 | eng | |||||||
| 資源タイプ | ||||||||
| 資源タイプ識別子 | http://purl.org/coar/resource_type/c_db06 | |||||||
| 資源タイプ | doctoral thesis | |||||||
| ID登録 | ||||||||
| ID登録 | 10.20569/00006112 | |||||||
| ID登録タイプ | JaLC | |||||||
| アクセス権 | ||||||||
| アクセス権 | open access | |||||||
| アクセス権URI | http://purl.org/coar/access_right/c_abf2 | |||||||
| 別タイトル | ||||||||
| その他のタイトル | 人工知能とスペクトルイメージング法を用いた岩石識別システムの開発に関する研究 | |||||||
| 作成者 |
SINAICE, BRIAN BINO
× SINAICE, BRIAN BINO
|
|||||||
| 内容記述（抄録） | ||||||||
| 内容記述タイプ | Other | |||||||
| 内容記述 | It is imperative that one acknowledges that we now live in a digital age where almost every aspect of our daily lives produces an immense amount of data, with this, one may say the 21st century and those to come are and will be data-driven. With a closer look into the mining industry, trends show a steady but inevitable shift from traditional to modern digital methods. One may attribute this shift to a number of reasons such as the shortage of specialists in various fields within the mining industry, be it exploration, monitoring, maintenance, and processing amongst others. Also, there is a lessening desire to conduct on-sight investigations as health and safety regulations stiffen along with the necessity to distance human life from occupational mishaps. Lastly, traditional methods of delineating rocks and minerals may at times be thought of as subjective, which is an inherent part of procedures conducted by humans. For this reason, this study aims to provide solutions to these problems through the employment of artificially intelligent (AI) algorithms, which are known to be objective in their analysis. Coupled with spectral imaging techniques in the discrimination of rocks and minerals, the study was split into three main chapters, each covering the achievable potential advantages of employing AI-based classifications, hence solving the aforementioned problems. The first aim was to combine two technologies to classify rocks; hyperspectral imaging, and artificial intelligence in the form of a one-dimensional deep learning convolution neural network (1D DL CNN). In order to classify rocks, visual imagery data generated using a hyperspectral camera was quantified in terms of how each captured image pixel responds across the electromagnetic spectrum, obtaining hyperspectral signatures of that particular rock. The second step involved running the hyperspectral signature data in a 1D CNN, application of this highly capable DL technique allows one to perform classification procedures with minimal error. The output results showed that the 1D CNN was capable of performing rock classifications via hyperspectral signatures as supported by the optimised (across three models) average per class prediction precision of 91.2% for all eight igneous rock lithologies employed in this part of the study. Having realized the time, computational and data storage costs attained in solving the first aim, the second aim was thus an attempt to develop a method by which rock classification may be performed without these costs. This was achieved by integrating Hyperspectral Imaging, Neighbourhood Component Analysis (NCA) and Machine Learning (ML) as a combined system that can classify rocks. Modestly put, hyperspectral imaging gathers electromagnetic signatures of the rocks in hundreds of spectral bands. However, this data suffers from what is termed the ‘dimensionality curse’, which led to the employment of NCA as a dimensionality reduction technique. NCA, in turn, highlights the most discriminant feature bands, the number of which is dependent on the intended application(s) of the system. The study’s envisioned application was rock classification via specialized multispectral bands. Hence, the study performed a 204-band hyperspectral to 5-band multispectral reduction, the reason being, that current production drones are limited to five multispectral band sensors. Based on these bands, the study applied ML to identify and classify rocks, thereby supporting the study’s hypothesis, reducing computational costs, and attaining an average optimized ML classification accuracy of 91.2%. The third aim was to investigate the potential use of drones in mining environments as a way in which data pertaining to the state of a site may be remotely collected. This aim proposes a combined system that employs a six bands multispectral image capturing camera mounted on an Unmanned Aerial Vehicle (UAV) drone, Spectral Angle Mapping (SAM), as well as AI. Unlike in the second aim, the 6 multispectral bands are factory pre-set, hence the employment of SAM to aid in pinpointing the spectra of sought after magnetite iron sands. Depth possessing multispectral data was captured at different flight elevations in an attempt to find the best elevation where remote identification via the UAV drone was possible. Data was analysed via SAM to deduce the cosine similarity thresholds at each elevation. Using these thresholds, AI algorithms were trained and tested to find the best performing model at classifying magnetite iron sand. Considering the post-flight logs, the spatial area coverage of 338 m2, a global classification accuracy of 99.7%, as well the per-class precision of 99.4%, the 20 m flight elevation outputs presented the best performance ratios overall. In conclusion, the study emphasizes that the employment of AI-based spectral imaging methods in various aspects of the mining industry is necessary to ensure the continuation of a robust and future-proof rock and mineral classification practice. Lastly, data-driven classification practices are sustainable, easily optimizable, repeatable, and objective in their outputs, deeming the proposed systems viable for current and future industrial applications. |
|||||||
| 著者版フラグ | ||||||||
| 出版タイプ | VoR | |||||||
| 出版タイプResource | http://purl.org/coar/version/c_970fb48d4fbd8a85 | |||||||
| 書誌情報 | 発行日 2022-09-29 | |||||||
| 出版者 | ||||||||
| 出版者 | 秋田大学 | |||||||
| 学位名 | ||||||||
| 学位名 | 博士（工学） | |||||||
| 学位授与機関 | ||||||||
| 学位授与機関識別子Scheme | kakenhi | |||||||
| 学位授与機関識別子 | 11401 | |||||||
| 学位授与機関名 | 秋田大学 | |||||||
| 学位授与年月日 | ||||||||
| 学位授与年月日 | 2022-09-29 | |||||||
| 学位授与番号 | ||||||||
| 学位授与番号 | 甲第1440号 | |||||||
