Multi-species Seagrass Detection Using Semi-supervised Learning

This paper introduces a semi-supervised two-stage framework that leverages a large collection of unlabelled seagrass data with the guidance of a small amount of labelled seagrass data. First, a state-of-the-art Convolutional Neural Network (CNN) named EfficientNet is trained on a small number of lab...

Full description

Saved in:
Bibliographic Details
Main Authors: Noman, Md Kislu, Islam, Syed Mohammed Shamsul, Abu-Khalaf, Jumana, Lavery, Paul
Format: Conference Proceeding
Language:English
Subjects:
Computer architecture
Convolutional neural networks
Deep Learning
EfficientNet
Image resolution
Multi-species
Sea-grass
Semi-supervised learning
Semisupervised learning
Training
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:This paper introduces a semi-supervised two-stage framework that leverages a large collection of unlabelled seagrass data with the guidance of a small amount of labelled seagrass data. First, a state-of-the-art Convolutional Neural Network (CNN) named EfficientNet is trained on a small number of labelled seagrass image patches. Next, the trained classifier generates pseudo labels for the unlabelled seagrass image patches. Finally, this classifier is retrained on the combination of labelled and pseudo labelled seagrass image patches. This approach achieves 98.0% multi-species detection accuracy on a newly developed seagrass dataset. Our approach on the publicly available 'DeepSeagrass' dataset achieves an overall accuracy of 91.3% and 93.0%, which outperforms the state-of-the-art multi-species seagrass detection accuracy of 88.2% and 92.4% for four and five classes, respectively.
ISSN:2151-2205
DOI:10.1109/IVCNZ54163.2021.9653222