3:30 PM - 3:45 PM
[SCG48-13] Depositional age of REY-rich mud around Minamitorishima Island using an image detection system for ichthyoliths based on deep learning models
Keywords:ichthyolith, biostratigraphy, deep learning, Mask R-CNN, EfficientNet-V2, REY-rich mud
In 2011, Kato et al. [1] discovered that deep-sea sediments in the Pacific Ocean are enriched in rare-earth elements and yttrium (REY), which can be a prominent resource for these industrially critical elements. More recently, extremely REY-rich mud with >5,000 ppm of total REY content was found in the Japanese exclusive economic zone (EEZ) around Minamitorishima Island, western North Pacific Ocean [2]. One of the key inforation to decipher the genesis of the promising resource is the depositional age. However, the absence of calcareous/siliceous microfossils has hampered a reliable age determination. An effective method to determine ages of such a “barren” pelagic clay is the ichthyolith biostratigraphy, which uses microfossils of fish teeth and denticles [3, 4].
A conventional observation method, however, requires manual searching, picking, observation, identification, and counting of ichthyoliths, which is a time-consuming process. Thus, the data of age-diagnostic ichthyoliths has been accumulated at a sluggish pace. To solve this problem, a deep-learning-based system to detect ichthyoliths from microscopic images has been established [5]. Although this system can detect fish teeth effectively from an image of sedimentary grains, it has not yet been verified whether this system is practically applicable in determining depositional ages of deep-sea sediments.
In this study, using the new deep-learning-based system, we detected 6284 ichthyoliths collected from 11 layers of a piston core MR14-E02 PC05 in the southern part of the Minamitorishima EEZ. Based on the detected images, we constrained the depositional ages of the REY-rich mud. Comparison of some corresponding layers between the cores in the same area demonstrated that, the newly constrained ages were consistent with those previously determined by the traditional method to observe ichthyoliths [4]. This result verifies that the detection system can be applicable for determining depositional ages of pelagic clay, which enables more efficient age-determination of REY-rich mud.
References: [1] Kato et al. (2011) Nature Geoscience 4, 535-539. [2] Iijima et al. (2016) Geochemical Journal 50, 557-573. [3] Doyle and Riedel (1985) Init. Repts. DSDP 86, 349-366. [4] Ohta et al. (2020) Scientific Reports, 10(1), 1-11. [5] Mimura et al., to be submitted.
A conventional observation method, however, requires manual searching, picking, observation, identification, and counting of ichthyoliths, which is a time-consuming process. Thus, the data of age-diagnostic ichthyoliths has been accumulated at a sluggish pace. To solve this problem, a deep-learning-based system to detect ichthyoliths from microscopic images has been established [5]. Although this system can detect fish teeth effectively from an image of sedimentary grains, it has not yet been verified whether this system is practically applicable in determining depositional ages of deep-sea sediments.
In this study, using the new deep-learning-based system, we detected 6284 ichthyoliths collected from 11 layers of a piston core MR14-E02 PC05 in the southern part of the Minamitorishima EEZ. Based on the detected images, we constrained the depositional ages of the REY-rich mud. Comparison of some corresponding layers between the cores in the same area demonstrated that, the newly constrained ages were consistent with those previously determined by the traditional method to observe ichthyoliths [4]. This result verifies that the detection system can be applicable for determining depositional ages of pelagic clay, which enables more efficient age-determination of REY-rich mud.
References: [1] Kato et al. (2011) Nature Geoscience 4, 535-539. [2] Iijima et al. (2016) Geochemical Journal 50, 557-573. [3] Doyle and Riedel (1985) Init. Repts. DSDP 86, 349-366. [4] Ohta et al. (2020) Scientific Reports, 10(1), 1-11. [5] Mimura et al., to be submitted.