Thesis Details
Classification of Varying-Size Plankton Images with Convolutional Neural Network
This work considers techniques of automatic image analysis based on convolutional neural networks (CNN) focused on plankton classification. There is a large variation in the shapes and sizes of plankton images. This makes the classification for CNN based methods challenging since CNNs typically require a fixed input size. Naive methods utilize scaling of the images to a common size. However, this operation leads to the loss of small details that are necessary for correct classification. The aim of this work was to design and implement a CNN-based classifier of plankton images and explore possible methods that can deal with a variety of image sizes. Multiple methods such as patch cropping, utilization of a spatial pyramid pooling layer, inclusion of metadata and construction of multi-stream model were evaluated on a challenging dataset of phytoplankton images. With these methods an improvement of 1.0 point was achieved for the InceptionV3 architecture resulting in an accuracy of 96.2 %. The main contribution of this thesis is an improvement of multiple CNN plankton classifiers by successfully applying these methods.
Plankton, machine learning, convolutional neural network, classification, image processing, varying image size, computer vision
Beran Vítězslav, doc. Ing., Ph.D. (DCGM FIT BUT), člen
Burget Lukáš, doc. Ing., Ph.D. (DCGM FIT BUT), člen
Hradiš Michal, Ing., Ph.D. (DCGM FIT BUT), člen
Křivka Zbyněk, Ing., Ph.D. (DIFS FIT BUT), člen
Orság Filip, Ing., Ph.D. (DITS FIT BUT), člen
@mastersthesis{FITMT23107, author = "Jaroslav Bure\v{s}", type = "Master's thesis", title = "Classification of Varying-Size Plankton Images with Convolutional Neural Network", school = "Brno University of Technology, Faculty of Information Technology", year = 2020, location = "Brno, CZ", language = "english", url = "https://www.fit.vut.cz/study/thesis/23107/" }