Landfast ice polynyas are important features near many northern coastal communities, and their automated detection from synthetic aperture radar (SAR) imagery is positioned to support on-ice travel safety under changing Arctic sea ice conditions. This research leveraged original datasets of over 5,000 Sentinel-1 SAR observations of wintertime polynyas mapped near the Canadian communities of Sanikiluaq and Nain to investigate deep learning-based landfast ice polynya detection. The Faster-RCNN object detection network was optimized for polynya detection through several modifications to network design elements and training strategies. Resulting detection models generalized well between regions and accurately detected polynyas with local backscatter contrasts above 5 dB, e.g. achieving 90% target recall at 24% precision. Polynyas smaller than 500 meters and with local backscatter contrasts less than 3 dB, constituting approximately half of all observations, were frequently missed. Precision scores below 30% were consistently incurred in attempts to optimize recall. Results highlight challenges to the consistent performance of single-image detectors due to variable and frequently weak polynya signatures in dual-polarized Sentinel-1 SAR backscatter. Future investigations into multi-temporal, multi-frequency, and/or higher-resolution SAR imagery could further support the delivery of robust hazard detection systems relevant to community sea ice safety and monitoring. Les polynies de glace c & ocirc;ti & egrave;re constituent un & eacute;l & eacute;ment important de la formation g & eacute;ographique & agrave; proximit & eacute; de nombreuses communaut & eacute;s c & ocirc;ti & egrave;res dans l'Arctique. Leur d & eacute;tection automatis & eacute;e & agrave; partir d'images radar & agrave; synth & egrave;se d'ouverture (SAR) soutiendrait la s & eacute;curit & eacute; des d & eacute;placements sur la glace dans un contexte temporellement variable des conditions de la banquise arctique. Cette recherche a exploit & eacute; des donn & eacute;es originales comprenant plus de 5 000 observations SAR Sentinel-1 de polynies hivernales cartographi & eacute;es pr & egrave;s des communaut & eacute;s canadiennes de Sanikiluaq et de Nain, afin d'& eacute;tudier la d & eacute;tection des polynies de glace c & ocirc;ti & egrave;re par apprentissage profond. Le r & eacute;seau de d & eacute;tection d'objets Faster-RCNN a & eacute;t & eacute; optimis & eacute; pour la d & eacute;tection des polynies gr & acirc;ce & agrave; plusieurs modifications apport & eacute;es & agrave; sa conception et & agrave; ses strat & eacute;gies d'entra & icirc;nement. Les mod & egrave;les de d & eacute;tection obtenus ont montr & eacute; une bonne g & eacute;n & eacute;ralisation entre les r & eacute;gions et ont d & eacute;tect & eacute; avec pr & eacute;cision les polynies pr & eacute;sentant des contrastes de r & eacute;trodiffusion locaux sup & eacute;rieurs & agrave; 5 dB, atteignant par exemple un rappel de 90 % pour la classe cible, associ & eacute;e & agrave; une pr & eacute;cision de 24 %. Les polynies de moins de 500 m & egrave;tres et pr & eacute;sentant des contrastes de r & eacute;trodiffusion locaux inf & eacute;rieurs & agrave; 3 dB, qui repr & eacute;sentent environ la moiti & eacute; des observations, ont souvent & eacute;t & eacute; manqu & eacute;es. Des scores de pr & eacute;cision inf & eacute;rieurs & agrave; 30 % ont & eacute;t & eacute; fr & eacute;quemment obtenus lors des tentatives d'optimization du rappel. Les r & eacute;sultats mettent en & eacute;vidence les difficult & eacute;s rencontr & eacute;es pour assurer des r & eacute;sultats uniformes par des images uniques, en raison de la variabilit & eacute; et de la faible intensit & eacute; fr & eacute;quente des signatures des polynies par r & eacute;trodiffusion SAR & agrave; double polarization de Sentinel-1. De plus amples recherches sur l'imagerie SAR multi-temporelle, multi-fr & eacute;quence et/ou & agrave; plus haute r & eacute;solution contribueront & agrave; la mise en place de syst & egrave;mes de d & eacute;tection robustes des polynies et, par cons & eacute;quent, & agrave; la s & eacute;curit & eacute; et & agrave; la surveillance des glaces de mer.
原文链接
分享
引用
