Automatic Tool for Coral Reef Detection, Localization and Annotation

Kód výsledku v IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F49777513%3A23520%2F20%3A43960896" target="_blank" >RIV/49777513:23520/20:43960896 - isvavai.cz</a>

Výsledek na webu

<a href="http://www.kky.zcu.cz/cs/sw/corals" target="_blank" >http://www.kky.zcu.cz/cs/sw/corals</a>

DOI - Digital Object Identifier

—

Jazyk výsledku

angličtina

Název v původním jazyce

Automatic Tool for Coral Reef Detection, Localization and Annotation

Popis výsledku v původním jazyce

The created tool is designed to automatically detect and annotate various benthic substrate (Corals) types over image collections taken from multiple coral reefs as part of a coral reef monitoring project with the Marine Technology Research Unit at the University of Essex. Live corals are an important biological class that has a massive contribution to the ocean ecosystem biodiversity. Corals are a key habitat for thousands of marine species and provide an essential source of nutrition and yield for people in developing countries. Therefore, automatic monitoring of coral reefs condition plays a crucial part in understanding future threats and prioritizing conservation efforts. The system&apos;s performance was validated in the international ImageCLEFcoral competition, where achieve an impressive mAP@0.5 of 0.582 in localization and 0.678 in instance segmentation. The system is wrapped up around the Mask R-CNN, the state-of-the-art instance segmentation framework and the TensorFlow Object Detection API - a deep learning framework that allows fine-tuning the publicly available checkpoints. To increase the model performance, we extend the recent state-of-the-art Convolutional Neural Network (CNN) object detection framework (Mask R-CNN) with an additional known as well as some unique techniques, e.g., detection ensemble, test time data augmentations, accumulated gradient normalization, and pseudo-labelling. The &quot;tool&quot; works as an end-to-end system that takes an image as input and returns 2D locations and substrate type for each prediction. Both the pre-trained models and the Application Interface (API) were made OpenSource to support further research in this area. The software was created by researchers from the Faculty of Applied Sciences - University of West Bohemia (UWB) and the Institute of Information Theory and Automation - Czech Academy of Sciences. Lukas Picek from the UWB was supported by the UWB grant SGS-2019-027.

Název v anglickém jazyce

Automatic Tool for Coral Reef Detection, Localization and Annotation

Popis výsledku anglicky

The created tool is designed to automatically detect and annotate various benthic substrate (Corals) types over image collections taken from multiple coral reefs as part of a coral reef monitoring project with the Marine Technology Research Unit at the University of Essex. Live corals are an important biological class that has a massive contribution to the ocean ecosystem biodiversity. Corals are a key habitat for thousands of marine species and provide an essential source of nutrition and yield for people in developing countries. Therefore, automatic monitoring of coral reefs condition plays a crucial part in understanding future threats and prioritizing conservation efforts. The system&apos;s performance was validated in the international ImageCLEFcoral competition, where achieve an impressive mAP@0.5 of 0.582 in localization and 0.678 in instance segmentation. The system is wrapped up around the Mask R-CNN, the state-of-the-art instance segmentation framework and the TensorFlow Object Detection API - a deep learning framework that allows fine-tuning the publicly available checkpoints. To increase the model performance, we extend the recent state-of-the-art Convolutional Neural Network (CNN) object detection framework (Mask R-CNN) with an additional known as well as some unique techniques, e.g., detection ensemble, test time data augmentations, accumulated gradient normalization, and pseudo-labelling. The &quot;tool&quot; works as an end-to-end system that takes an image as input and returns 2D locations and substrate type for each prediction. Both the pre-trained models and the Application Interface (API) were made OpenSource to support further research in this area. The software was created by researchers from the Faculty of Applied Sciences - University of West Bohemia (UWB) and the Institute of Information Theory and Automation - Czech Academy of Sciences. Lukas Picek from the UWB was supported by the UWB grant SGS-2019-027.

Druh

R - Software

CEP obor

—

OECD FORD obor

20205 - Automation and control systems

Projekt

<a href="/cs/project/LO1506" target="_blank" >LO1506: Podpora udržitelnosti centra NTIS - Nové technologie pro informační společnost</a><br>

Návaznosti

P - Projekt vyzkumu a vyvoje financovany z verejnych zdroju (s odkazem do CEP)<br>S - Specificky vyzkum na vysokych skolach<br>I - Institucionalni podpora na dlouhodoby koncepcni rozvoj vyzkumne organizace

Rok uplatnění

2020

Kód důvěrnosti údajů

S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů

Interní identifikační kód produktu

ZCU/KKY/2020/015

Technické parametry

Tento softwarový nástroj používá pokročilé metody strojového učení. Pro bližší informace k technickým detailům a získání licence prosím kontaktujte: Lukáš Picek, KKY FAV Západočeská univerzita v Plzni, Technická 8, 301 00 Plzeň (IČ: 49777513), email: picekl@kky.zcu.cz, tel. 37763 2125.

Ekonomické parametry

Tento SW nástroj je volně dostupný pro využití nekomerčního charakteru. Vzhledem k významnosti výstupu v oblasti vzdělávání a souvisejícímu společenskému dopadu nelze zhodnotit ekonomické parametry SW výstupu - operativní výzkum ve veřejném zájmu.

IČO vlastníka výsledku

49777513

Název vlastníka

Západočeská univerzita v Plzni