10:00 AM - 10:15 AM
[MIS21-05] Initial application of automatic feature recognition algorithm to enhance geological survey method
Keywords:field geology, outcrop, image analysis
The purpose of this research is to establish the in-situ geological survey method by replacing the traditional outcrop description and classification process, which is strongly relying on the experts’ knowledge and experience, with computational processing using an automatic recognition algorithm for lithologic features. The first phase of this research is to improve the efficiency of geological surveys on the Earth and to increase the possibility of making new discoveries. In the second phase, key technologies will be developed for automatic selection of sites for detailed scientific data acquisition in space exploration of underground cavities on the Moon and Mars, and for data compression and high-efficiency transfer using edge computing, in order to realize an efficient autonomous exploration system.
Important strategy of this research is that multidisciplinary experts are involved, including the experts in the space exploration, image recognition algorithms, and volcanology. Just an application of existing image recognition algorithms that have reached maturity in the fields such medical imaging to geological survey has the potential to create new value. Conversely, overcoming issues that have not been a problem in fields such as medical imaging could lead to breakthrough in the image recognition algorithms.
Developed automated feature recognition algorithms will be applied not only to geological survey on the Earth but also to the autonomous space exploration system, where it is difficult to obtain the enough amount of the training data, so that it is important to realize the wide range of application as much as possible. If the results of automatic feature recognition are in good agreement with the results of expert recognition, detailed investigation to analyze the variables used by the algorithm, such as image elements, luminance, color information, texture features, and logic used to evaluate final scientific features, and analyze whether the recognition logic is consistent with that of the expert. For these reasons, it is our policy not to use machine learning, which requires a large amount of training data and makes it difficult to systematically understand the internal logic of the feature recognition algorithm.
Most of the current feature recognition algorithm is consists of the image processing methods developed in the previous studies for the fields of medical images and satellite images. The image elements those are used to describe the scientific features, are segments obtained by subdividing the entire image into small section based on luminance and color information of rocks and boundary lines of each stratum, and feature lines which are the ridges of irregularities that appear on the surface due to lave flow during rock formation.
Referring to previous studies in the fields of medical images and satellite images, the image elements used are segments obtained by subdividing the entire image into small sections based on luminance and color information of rocks and the boundary lines of each stratum, and feature lines, which are the ridges of irregularities that appear on the outer surface due to lava flow during rock formation, such as lava. After image preprocessing for robust segmentation, image segmentation is carried out, followed by cluster classification using color information and average values of texture features for each segment, and then the neighbor segments are merged depending on the segment colors and the statistics such as GLCM if needed, finally lithological features such as layer thickness distribution and the presence of unconformities are evaluated.
As an initial application, the extraction of the geological bedding structures for the photo image of the outcrop of the airfall tephra beds exposing on the “Chisou Disetsudanmen” outcrop in Izu-Oshima as an example of the simple stratified beds without major folding and unconformity. Not only clear boundaries, but also unclear boundaries of scoria layers could be recognized by image enhancement and edge extraction using luminance gradients with pre-defined direction of interest and detection threshold.
In the future, the robustness of this automatic feature recognition algorithm for various geological formations will be validated, and the efficient method that can be used to analyze a wide range of geological formations based on many optical images based on a uniform formula and constants will be realized.
Important strategy of this research is that multidisciplinary experts are involved, including the experts in the space exploration, image recognition algorithms, and volcanology. Just an application of existing image recognition algorithms that have reached maturity in the fields such medical imaging to geological survey has the potential to create new value. Conversely, overcoming issues that have not been a problem in fields such as medical imaging could lead to breakthrough in the image recognition algorithms.
Developed automated feature recognition algorithms will be applied not only to geological survey on the Earth but also to the autonomous space exploration system, where it is difficult to obtain the enough amount of the training data, so that it is important to realize the wide range of application as much as possible. If the results of automatic feature recognition are in good agreement with the results of expert recognition, detailed investigation to analyze the variables used by the algorithm, such as image elements, luminance, color information, texture features, and logic used to evaluate final scientific features, and analyze whether the recognition logic is consistent with that of the expert. For these reasons, it is our policy not to use machine learning, which requires a large amount of training data and makes it difficult to systematically understand the internal logic of the feature recognition algorithm.
Most of the current feature recognition algorithm is consists of the image processing methods developed in the previous studies for the fields of medical images and satellite images. The image elements those are used to describe the scientific features, are segments obtained by subdividing the entire image into small section based on luminance and color information of rocks and boundary lines of each stratum, and feature lines which are the ridges of irregularities that appear on the surface due to lave flow during rock formation.
Referring to previous studies in the fields of medical images and satellite images, the image elements used are segments obtained by subdividing the entire image into small sections based on luminance and color information of rocks and the boundary lines of each stratum, and feature lines, which are the ridges of irregularities that appear on the outer surface due to lava flow during rock formation, such as lava. After image preprocessing for robust segmentation, image segmentation is carried out, followed by cluster classification using color information and average values of texture features for each segment, and then the neighbor segments are merged depending on the segment colors and the statistics such as GLCM if needed, finally lithological features such as layer thickness distribution and the presence of unconformities are evaluated.
As an initial application, the extraction of the geological bedding structures for the photo image of the outcrop of the airfall tephra beds exposing on the “Chisou Disetsudanmen” outcrop in Izu-Oshima as an example of the simple stratified beds without major folding and unconformity. Not only clear boundaries, but also unclear boundaries of scoria layers could be recognized by image enhancement and edge extraction using luminance gradients with pre-defined direction of interest and detection threshold.
In the future, the robustness of this automatic feature recognition algorithm for various geological formations will be validated, and the efficient method that can be used to analyze a wide range of geological formations based on many optical images based on a uniform formula and constants will be realized.