Image classification is a repetitive, non-creative process and is well suited for automation. In the last decade, thanks to the advent of Big Data and GPU computing, Convolutional Neural Networks (CNNs) have become one of the most popular and powerful tools in such environment. Ensemble Learning (EL) is a fusion technique, with which multiclassifiers can be built to further enhance performance. We applied EL to classification problem in the field of microbiology, the classification of six species of foraminifera, expanding on a solution presented in an existing research paper. Utilizing multiple image pre-processing techniques (PCA, MDS, DCT-II), five CNNs (GoogLeNet, ResNet50, ResNet101, MobileNet, EfficientNetb0) and decision fusion of said networks, we managed to get results that surpass the ones of both human experts and other automated solutions.
La classificazione di immagini è un’attività non creativa e ripetitiva che si presta molto bene all’automazione. Nell’ultimo decennio, grazie all’avvento del Big Data e dello sviluppo tecnologico delle GPU, le reti neurali convoluzionali (CNN) sono diventate uno degli strumenti più popolari e performanti in questo ambito. L’Ensemble Learning (EL) è una tecnica di fusione tramite la quale è possibile creare multiclassificatori e migliorare ulteriormente le prestazioni. Abbiamo applicato EL ad un problema di classificazione nell’ambito della microbiologia, la classificazione di sei specie di foraminifera, espandendo la soluzione presentata da un paper di ricerca esistente. Utilizzando più metodi di pre-processing delle immagini (PCA, MDS, DCT-II), cinque CNN (GoogLeNet, ResNet50, ResNet101, MobileNet, EfficientNetb0) e la fusione a livello di decisione delle stesse reti, abbiamo ottenuti risultati che superano in maniera convincente sia quelli generati da esperti umani che da altre soluzioni automatizzate.
Convolutional Neural Network Ensemble for Foraminifera classification
VALLAZZA, MIRO
2021/2022
Abstract
Image classification is a repetitive, non-creative process and is well suited for automation. In the last decade, thanks to the advent of Big Data and GPU computing, Convolutional Neural Networks (CNNs) have become one of the most popular and powerful tools in such environment. Ensemble Learning (EL) is a fusion technique, with which multiclassifiers can be built to further enhance performance. We applied EL to classification problem in the field of microbiology, the classification of six species of foraminifera, expanding on a solution presented in an existing research paper. Utilizing multiple image pre-processing techniques (PCA, MDS, DCT-II), five CNNs (GoogLeNet, ResNet50, ResNet101, MobileNet, EfficientNetb0) and decision fusion of said networks, we managed to get results that surpass the ones of both human experts and other automated solutions.File | Dimensione | Formato | |
---|---|---|---|
Vallazza_Miro.pdf
accesso aperto
Dimensione
1.35 MB
Formato
Adobe PDF
|
1.35 MB | Adobe PDF | Visualizza/Apri |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/29285