Pasar al contenido principal

Resultados relacionados

Resultados totales (Incluyendo duplicados): 3
Encontrada(s) 1 página(s)

Repositori Institucional UIB
Publicación Artículo científico (JournalArticle)

Automatic deep learning-based pipeline for Mediterranean fish segmentation

  • Muntaner-Gonzalez, Caterina
  • Nadal-Martínez, Antonio
  • Martin-Abadal, Miguel
  • Gonzalez-Cid, Yolanda
[eng] Climate change and human activities are altering the Mediterranean marine biodiversity. Monitoring these alterations over time is crucial for assessing the health of coastal environments and preserving local species. However, this monitoring process is resource-intensive, requiring taxonomic experts and significant amounts of time. To address this, we present an automated pipeline that detects, classifies and segments 17 species of Mediterranean fish using YOLOv8, integrated into an underwater stereo vision system capable of real-time inference and selective data storage. The proposed model demonstrates strong performance in detecting, classifying, and segmenting 17 Mediterranean fish species, achieving an mAP50(B) of 0.886 and an mAP50(M) of 0.889.
Repositori Institucional UIB
Publicación Artículo científico (JournalArticle)

Autonomous underwater stereo vision system for non-invasive fish length estimation in marine environments

  • Muntaner-Gonzalez, Caterina
  • Martin-Abadal, Miguel
  • Gonzalez-Cid, Yolanda
[eng] Marine ecosystems are undergoing rapid change due to climate and human pressures, increasing the need for monitoring tools. Fish length data are particularly valuable for assessing population and ecosystem status, yet conventional measurement methods are often time-consuming and invasive. This study introduces a practical, non-invasive pipeline for automatic fish length estimation based on an underwater Stereo-Vision System (SVS) that provides synchronized image pairs for 3D reconstruction. Fish length is estimated by combining 2D segmentation masks with reconstructed 3D scene information, with a multi-object tracking module that enables track-based aggregation of frame-wise estimates to improve robustness. A YOLOv11-based segmentation model together with a BoT-SORT tracker is employed to detect, segment and track target species. The proposed method is evaluated on two datasets of increasing complexity, including controlled multi-species scenarios and live underwater footage in controlled conditions. It achieves a Mean Absolute Error (MAE) of 1.30 cm, a standard deviation of 1.68 cm, a Mean Absolute Percentage Error (MAPE) of 4.53%, and an overall success rate of 49.7%, while operating under a conservative filtering strategy based on geometric and track-level consistency constraints. The pipeline is further integrated into an autonomous SVS, enabling real-time on-board operation in marine environments. Results demonstrate that accurate and robust fish length estimation can be achieved under challenging underwater conditions, supporting the deployment of scalable, non-invasive monitoring solutions for ocean observation.
Repositori Institucional UIB
Publicación Artículo científico (JournalArticle)

A Deep Learning Approach to Estimate Halimeda incrassata Invasive Stage in the Mediterranean Sea

  • Muntaner-González, Caterina
  • Martín-Abadal, Miguel
  • González-Cid, Yolanda
[eng] Abstract: ersity in the Mediterranean Sea. Monitoring these changes over time is crucial for assessing the health of coastal environments and preserving local species. However, this monitoring process is resource-intensive, requiring taxonomic experts and significant amounts of time. Recently, deep learning approaches have attempted to automate the detection of certain seagrass species like Posidonia oceanica and Halophila ovalis from two different strategies: seagrass coverage estimation and detection. This work presents a novel approach to detect Halimeda incrassata and estimate its coverage, independently of the invasion stage of the algae. Two merging methods based on the combination of the outputs of an object detection network (YOLOv5) and a semantic segmentation network (U-net) are developed. The system achieves an F1-scoreof 84.2% and a Coverage Error of 5.9%, demonstrating its capability to accurately detect Halimeda incrassata and estimate its coverage independently of the invasion stage.