Deep Learning Methods for Plankton Identification: A Bibliometric Analysis and General Review
Abstract
This study investigates trends in deep learning research on microscopic images to identify and classify plankton. We performed a bibliometric analysis using VOSviewer to display the visualizations. The first step is to extract the relevant journals from the Scopus database. Extraction from relevant journals found 42 articles, including 21 journals and 21 conferences. We analyzed publication trends and analyzes of co-authorship, co-occurrence, and citations. In addition, it examines the collaborative relationship between authors by country, the emergence of co-author keywords, and the citation of journal sources. We found that the researchers performed plankton classification tasks that achieved high accuracy (>90%). The plankton identification method uses a deep learning architecture based on the Convolutional Neural Network. Among the architectures are AlexNet, GoogleNet, InceptionV3, VGGNet (16.19), ResNet (50.101), DenseNet, NasNet, MobileNetV2, ResidualNet, Shufflenet, EfficienNet, and CapsNet. Some researchers combine CNN with SparseConvNets, LMP, SVM, MorphoCluster, and Quantification Algorithm (CC, AC, PCC, PAC, Hdy). Some methods also implement transfer learning, fine-tuning, and advanced features. In addition to the CNN approach, End-to-End Object Detection with transformers is one of the new approaches in the future to identify plankton.
BibTeX
@inproceedings{Ardhi2022DeepLearning,
issauthor = {Ardhi, Ovide Decroly Wisnu and Soeprobowati, Tri Retnaningsih and Adi, Kusworo and Prakasa, Esa and Rachman, Arief},
booktitle = {2022 1st International Conference on Smart Technology, Applied Informatics, and Engineering (APICS)},
title = {Deep Learning Methods for Plankton Identification: A Bibliometric Analysis and General Review},
year = {2022},
volume = {},
number = {},
pages = {96-101},
abstract = {This study investigates trends in deep learning research on microscopic images to identify and classify plankton. We performed a bibliometric analysis using VOSviewer to display the visualizations. The first step is to extract the relevant journals from the Scopus database. Extraction from relevant journals found 42 articles, including 21 journals and 21 conferences. We analyzed publication trends and analyzes of co-authorship, co-occurrence, and citations. In addition, it examines the collaborative relationship between authors by country, the emergence of co-author keywords, and the citation of journal sources. We found that the researchers performed plankton classification tasks that achieved high accuracy (>90%). The plankton identification method uses a deep learning architecture based on the Convolutional Neural Network. Among the architectures are AlexNet, GoogleNet, InceptionV3, VGGNet (16.19), ResNet (50.101), DenseNet, NasNet, MobileNetV2, ResidualNet, Shufflenet, EfficienNet, and CapsNet. Some researchers combine CNN with SparseConvNets, LMP, SVM, MorphoCluster, and Quantification Algorithm (CC, AC, PCC, PAC, Hdy). Some methods also implement transfer learning, fine-tuning, and advanced features. In addition to the CNN approach, End-to-End Object Detection with transformers is one of the new approaches in the future to identify plankton.},
keywords = {Support vector machines;Visualization;Bibliometrics;Transfer learning;Transformers;Market research;Picture archiving and communication systems;deep learning;plankton;identification;bibliometric;co-occurrence analysis},
doi = {10.1109/APICS56469.2022.9918707},
issn = {},
month = {Aug},
}