Scholarlink

<p>To classify the sedimentary types of the seafloor, image analyses were performed on the deep-sea camera (DSC) images, including semantic segmentation, clustering, and grouping. Segmentation for the classification of sedimentary types was conducted using the random forest-based classifier provided by the Labkit plugin in Fiji, and then K-means clustering and grouping using gray-level co-occurrence matrix (GLCM) were performed. Sedimentary types were grouped into three clusters, and the average pixel ratio of the ferromanganese crust was approximately 0.75 in the ferromanganese dominant (Mn-dominant) type, approximately 0.35 in the mixed type, and approximately 0.11 in the sediment-dominant (S-dominant) type. The GLCM feature analysis of patches corresponding to ferromanganese crusts and sediments within the image showed that they were distinguished at a dissimilarity value of approximately 2. In addition, representative images selected for the Mn-dominant type and S-dominant type were distinguished at a dissimilarity value of approximately 0.7. These results are expected to suggest that image classification using pixel ratios and GLCM features has the potential to reduce subjective decisions by researchers and provide clearer classification criteria. Moreover, it is expected that these semi-automated processes can reduce researchers’ efforts.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Semiautomated sedimentary type classification of seamounts in the Western Pacific using deep-sea camera images

  • 期刊论文

  • 发表日期：

    2025年6月11日

查看全文
  • Hyeonho An,
  • Yangwon Lee,
  • Seung Jin Yang,
  • Chailinn Park,
  • Youngtak Ko,
  • Kiho Yang,
  • Jaewoo Jung

摘要

To classify the sedimentary types of the seafloor, image analyses were performed on the deep-sea camera (DSC) images, including semantic segmentation, clustering, and grouping. Segmentation for the classification of sedimentary types was conducted using the random forest-based classifier provided by the Labkit plugin in Fiji, and then K-means clustering and grouping using gray-level co-occurrence matrix (GLCM) were performed. Sedimentary types were grouped into three clusters, and the average pixel ratio of the ferromanganese crust was approximately 0.75 in the ferromanganese dominant (Mn-dominant) type, approximately 0.35 in the mixed type, and approximately 0.11 in the sediment-dominant (S-dominant) type. The GLCM feature analysis of patches corresponding to ferromanganese crusts and sediments within the image showed that they were distinguished at a dissimilarity value of approximately 2. In addition, representative images selected for the Mn-dominant type and S-dominant type were distinguished at a dissimilarity value of approximately 0.7. These results are expected to suggest that image classification using pixel ratios and GLCM features has the potential to reduce subjective decisions by researchers and provide clearer classification criteria. Moreover, it is expected that these semi-automated processes can reduce researchers’ efforts.