[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-85024-en":3,"doc-seo-85024-105":29,"detail-sidebar-cat-0-en-105":91},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":20,"is_deleted":4,"is_public":20,"is_downloadable":20,"audit_status":20,"page_count":21,"language":22,"language_code":23,"site_id":24,"html_lang":23,"table_of_contents":25,"faqs":26,"seo_title":13,"seo_description":14,"update_tm":27,"read_time":28},85024,1374391975076,"Riley","https://ap-avatar.wpscdn.com/avatar/14000253ca4ec9f6853?x-image-process=image/resize,m_fixed,w_180,h_180&k=1783305029341752051",8,"Research & Report","A Self-Supervised Approach for Minimal-Annotation Hydroacoustic Data Exploration","Passive hydroacoustic monitoring produces large continuous recordings but remains underexploited because manual annotation is costly. Supervised detection performs well only with large labeled corpora, while weakly trained alternatives rely on strong prior knowledge and predefined targets. This work presents a self-supervised exploration pipeline for low-frequency data using a Masked AutoEncoder for patch representations, event-level embeddings, and dataset-scale clustering with UMAP and HDBSCAN. Evaluated on a multi-year near Mayotte dataset, it identifies hydroacoustic patterns with fast manual mapping, achieves classifier performance comparable to existing detectors, and recovers seasonal behaviors while discovering new signals.","arXiv :2607 .07733v 1 [ cs . SD] 7 Jul 2026  \nA Self-Supervised Approach for Minimal-Annotation Hydroacoustic Data Exploration  \nPierre-Yves Raumer, 1, 2, 3 Axel Marmoret,4 Dorian Cazau,3 Anatole Gros-Martial,5  \nRichard Dreo,6, 7 Ma¨elle Torterotot,3 Sara Bazin,8 Flore Samaran,3 and Jean-Yves Royer2  \n1 Laboratoire de G´eologie, Ecole Normale Sup´erieure/CNRS UMR 8538, PSL Research University, Paris 75005, France  \n2 Universit´e de Brest, CNRS, Ifremer, UMR6538 Geo-Ocean, 29280 Plouzan´e, France  \n3 Lab-STICC – UMR 6285 CNRS, ENSTA IP Paris, Brest,  \nFrance  \n4 IMT Atlantique, Lab-STICC, UMR 6285 CNRS, Brest,  \nFrance  \n5 Centre d’Etudes Biologiques de Chiz´e (CEBC), UMR 7372, CNRS-La Rochelle Universit´e, Villiers-en-Bois, France  \n6 Universit´e de Paris, Institut de physique du globe de Paris, CNRS, F-75005 Paris, France  \n7 SAS Boksound, Kernevez, 20590 Rosnoen, France.  \n8 Universit´e de Brest, CNRS, Ifremer, UMR6538 Geo-Ocean, IUEM,  \n29280 Plouzan´e, France  \nJASA  \n1 Passive hydroacoustic monitoring often generates large volumes of continuous record- 2 ings that are only partially exploited due to the cost of manual annotation. Su- 3 pervised detection methods perform well but require large labeled datasets, seldom  \n4 available for rare signals or understudied environments. This work proposes a self- 5 supervised exploration pipeline to address this limitation in low-frequency settings.  \n6 A Masked AutoEncoder (MAE) is pre-trained on a reconstruction pretext task, then  \n7 used to extract patch-level representations from spectrograms. Within each spec- 8 trogram, adjacent informative patches are aggregated into event-level embeddings, 9 enabling the disentanglement of overlapping events. These embeddings are then clus- 10 tered at the dataset scale using the dimension reduction algorithm UMAP and the  \n11 clustering algorithm HDBSCAN to identify hydroacoustic patterns. The pipeline was  \n12 applied to a multi-year hydroacoustic dataset collected near Mayotte Island, Indian  \n13 Ocean, containing marine mammal vocalizations, seismo-volcanic signals, and anthro- 14 pogenic noise. The 317 clusters were manually mapped to 15 hydroacoustic classes  \n15 or noise in less than one hour. The method was evaluated in two ways. Quantita- 16 tively, when used as a classifier, it achieved performance comparable to two existing  \n17 detectors. Qualitatively, it recovered known seasonal patterns of marine mammal  \n18 acoustic activity. It also identified patterns of previously unstudied signals, thereby  \n19 demonstrating its practical value.  \nJASA  \nI. INTRODUCTION  \nPassive hydroacoustic monitoring has become a key tool for observing the ocean at large spatial and temporal scales. Because low-frequency sound propagates efficiently underwater, hydrophone networks can record hydroacoustic activity over thousands of kilometers (e.g. Ingale et al. , 2021) . Such datasets provide valuable information on marine mammals (e.g. Torterotot et al. , 2020), geophysical activity (e.g. Raumer et al. , 2025), and anthropogenic noise (e.g. Haver et al. , 2023) . Driven by these motivations, long-term hydrophone observatories have been deployed worldwide, and collected large volumes of continuous recordings.  \nDespite this abundance of data, only a small fraction of hydroacoustic recordings has been explored. Existing methodologies generally suffer from a trade-off between human effort and detection flexibility. On one hand, manual annotation is time-consuming, userdependent, and scales poorly (e.g. Dubus et al. , 2023; Raumer et al. , 2024) . On the other hand, automatic detectors designed to assist these efforts either rely on large labeled datasets in the case of supervised training (e.g. Dubus et al. , 2024; Raumer et al. , 2024), or avoid the training phase at the cost of specificity, requiring high prior knowledge to target restricted and structured signals (e.g. Dr´eo et al. , 2025; Socheleau et al. , 2015; Torterotot et al. , 2019) . Cruc","cbCailZTQievmYXj","https://ap.wps.com/l/cbCailZTQievmYXj","pdf",21014937,1,40,"English","en",105,"# Introduction\n# Proposed Self-Supervised Exploration Pipeline\n## Representation Learning with MAE/ViT\n## Event-Level Embeddings and Clustering\n# Evaluation Results","[{\"question\":\"Why is passive hydroacoustic data difficult to explore with existing methods?\",\"answer\":\"Manual annotation is expensive and does not scale, while supervised detectors require large labeled datasets and weak alternatives need strong prior knowledge targeting predefined signal types.\"},{\"question\":\"What is the core mechanism of the proposed approach?\",\"answer\":\"A Masked AutoEncoder pre-trains a Vision Transformer to learn patch-level representations from spectrograms, aggregates adjacent informative patches into event-level embeddings, then performs dataset-scale clustering.\"},{\"question\":\"How is the method evaluated and what does it achieve?\",\"answer\":\"On a multi-year low-frequency dataset near Mayotte, the approach produces many clusters that can be manually mapped to hydroacoustic classes quickly, yields classifier performance comparable to two existing detectors, and recovers seasonal acoustic patterns while identifying previously unstudied signals.\"}]",1784200455,101,{"code":4,"msg":30,"data":31},"ok",{"site_id":24,"language":23,"slug":32,"title":13,"keywords":33,"description":14,"schema_data":34,"social_meta":86,"head_meta":88,"extra_data":90,"updated_unix":27},"a-self-supervised-approach-for-minimal-annotation-hydroacoustic-data-exploration","",{"@graph":35,"@context":85},[36,53,68],{"@type":37,"itemListElement":38},"BreadcrumbList",[39,43,47,50],{"item":40,"name":41,"@type":42,"position":20},"https://docshare.wps.com","Home","ListItem",{"item":44,"name":45,"@type":42,"position":46},"https://docshare.wps.com/document/","Document",2,{"item":48,"name":12,"@type":42,"position":49},"https://docshare.wps.com/document/research-report/",3,{"item":51,"name":13,"@type":42,"position":52},"https://docshare.wps.com/document/a-self-supervised-approach-for-minimal-annotation-hydroacoustic-data-exploration/85024/",4,{"url":51,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":23,"description":14,"dateModified":61,"datePublished":62,"encodingFormat":60,"isAccessibleForFree":63,"interactionStatistic":64},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":40,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-17","2026-07-16",true,{"@type":65,"interactionType":66,"userInteractionCount":20},"InteractionCounter",{"@type":67},"ViewAction",{"@type":69,"mainEntity":70},"FAQPage",[71,77,81],{"name":72,"@type":73,"acceptedAnswer":74},"Why is passive hydroacoustic data difficult to explore with existing methods?","Question",{"text":75,"@type":76},"Manual annotation is expensive and does not scale, while supervised detectors require large labeled datasets and weak alternatives need strong prior knowledge targeting predefined signal types.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"What is the core mechanism of the proposed approach?",{"text":80,"@type":76},"A Masked AutoEncoder pre-trains a Vision Transformer to learn patch-level representations from spectrograms, aggregates adjacent informative patches into event-level embeddings, then performs dataset-scale clustering.",{"name":82,"@type":73,"acceptedAnswer":83},"How is the method evaluated and what does it achieve?",{"text":84,"@type":76},"On a multi-year low-frequency dataset near Mayotte, the approach produces many clusters that can be manually mapped to hydroacoustic classes quickly, yields classifier performance comparable to two existing detectors, and recovers seasonal acoustic patterns while identifying previously unstudied signals.","https://schema.org",{"og:url":51,"og:type":87,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":89,"canonical":51},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":92},[93,97,101,105,110,115,119,122,127,130,134],{"id":20,"doc_module":4,"doc_module_name":45,"category_name":94,"show_sort_weight":95,"slug":96},"Story & Novel",90,"story-novel",{"id":46,"doc_module":4,"doc_module_name":45,"category_name":98,"show_sort_weight":99,"slug":100},"Literature",80,"literature",{"id":52,"doc_module":4,"doc_module_name":45,"category_name":102,"show_sort_weight":103,"slug":104},"Exam",70,"exam",{"id":106,"doc_module":4,"doc_module_name":45,"category_name":107,"show_sort_weight":108,"slug":109},5,"Comic",60,"comic",{"id":111,"doc_module":4,"doc_module_name":45,"category_name":112,"show_sort_weight":113,"slug":114},6,"Technology",50,"technology",{"id":116,"doc_module":4,"doc_module_name":45,"category_name":117,"show_sort_weight":21,"slug":118},7,"Healthcare","healthcare",{"id":11,"doc_module":4,"doc_module_name":45,"category_name":12,"show_sort_weight":120,"slug":121},30,"research-report",{"id":123,"doc_module":4,"doc_module_name":45,"category_name":124,"show_sort_weight":125,"slug":126},9,"Religion & Spirituality",20,"religion-spirituality",{"id":125,"doc_module":4,"doc_module_name":45,"category_name":128,"show_sort_weight":125,"slug":129},"World Cup","world-cup",{"id":131,"doc_module":4,"doc_module_name":45,"category_name":132,"show_sort_weight":131,"slug":133},10,"Lifestyle","lifestyle",{"id":135,"doc_module":4,"doc_module_name":45,"category_name":136,"show_sort_weight":106,"slug":137},19,"General","general"]