TGTakayuki Genda

Takayuki Genda

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Takayuki Genda has a total of 12 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Applying Deep-Learning-Based Sonar Image Analysis to Exploration of Hydrothermal Activities

    by Kazuhide Mimura, Junji Kaneko, Tatsuo Nozaki, Takafumi Kasaya, Yohei Sugimoto, Takayuki Genda, Kentaro Nakamura

    Seafloor hydrothermal activity provides significant mineral resources for various metals and serves as an important medium for elucidating the origin of life and material cycling between the ocean and solid Earth. Previous studies have demonstrated that multi-beam echo sounders (MBES) are effective for exploring undiscovered hydrothermal activity. However, conventional methods require human observers to continuously monitor MBES data converted to images, often for extended periods overnight. We have previously proposed an automated observation of MBES images using object detection, a type of deep-learning technique. In this study, we conducted an MBES survey at Higashi-Aogashima Knoll Caldera near the Izu Bonin arc in the western North Pacific Ocean and applied this AI based method for practical exploration. From approximately 25,000 images collected over about 25 h of surveying, the model identified the prominent area potentially associated with hydrothermal activity. Notably, in one of the prominent areas, a new occurrence of hydrothermal vent was confirmed through a subsequent remotely operated vehicle (ROV) survey.

  • Open Access

    Precise Detection of Hydrothermal Emission Signatures from Autonomous Underwater Vehicle Multi-Beam Echo Sounder Images by Deep Learning

    by Takayuki Genda, Kazuhide Mimura, Kentaro Nakamura, Takeshi Nakatani, Kazuya Kitada, Kazutaka Yasukawa, Yasuhiro Kato

    Seafloor massive sulfide deposits are attracting attention as new sources of industrially important metals. Methods for detecting these deposits using a multi-beam echo sounder (MBES) was previously demonstrated, with recent interest in automating this detection using object detection models. However, conventional models often produce numerous false positives due to the sparse occurrence of hydrothermal signals during actual explorations. Here, we report that by incorporating a simple timeseries analysis into the outputs of conventional object detection models, we have, for the first time, successfully achieved highprecision detection of hydrothermal signatures from MBES images acquired by autonomous underwater vehicles (AUVs). An object detection model, YOLOv8, was utilized to detect hydrothermal signals from time-series MBES data obtained by AUVs. A 5-term moving average was applied to the confidence scores, followed by binary classification to determine the presence of hydrothermal signals in each frame. Cross-validation using data from 13 dives showed an improvement in precision from 0.425-0.582 at the frame and 0.142-0.541 at the hydrothermal event levels. Additionally, an evaluation method considering practical aspects of hydrothermal exploration, such as deployment frequency and discovery rates, was developed. This method more accurately assesses the practical utility of hydrothermal exploration. The results indicate that the proposed algorithm, leveraging time-series information through a moving average, is potentially effective for practical hydrothermal exploration.