phd-pesek-2022

Title

Possibilities of convolutional neural networks use for remote sensing image classification

Description

In recent years, the speed of technological progress in certain science fields is getting faster and faster. It is making it hard for other scientific areas to keep up with this tempo. One of the exemplary relationships is the link between artificial and convolutional neural network structures and the province of geomatics or remote sensing. New architectures of artificial neural network models are being published with an expeditious tempo and the common approach of the remote sensing researchers is to use the most recent structures, without the basic understanding of the background or relative performance. The goal of this thesis is to perform systematic research on the possibilities of use of chosen convolutional neural network architectures on various selected use cases from the field of remote sensing.

Student

Ing. Ondřej Pešek

Supervisors

Prof. Aleš Čepek, CSc.
Ing. Martin Landa, Ph.D.

Readers

Professional debate: Ing. Lukáš Brodský, Ph.D.
The final text: Would be nice to finish it firstly

Defence

Hopefully one day.

Text

Intermediate results

Using Convolutional Neural Networks for Cloud Detection on VENμS Images over Multiple Land-Cover Types

$Binary detection of clouds on a patch of an S05 VENμS tile from various architecture settings: (a\) RGB representation of a patch of an S05 VENμS tile; (b\) ground-truth mask; (c\) U-Net—Dropout 0 %; (d\) U-Net—Dropout 50%; (e\) SegNet—Dropout 0%; (f\) SegNet—Dropout 50%; (g\) DeeplabV3+ with ResNet-50—Dropout 0%; (h\) DeeplabV3+ with ResNet-50—Dropout 50%; (i\) DeeplabV3+ with ResNet-101—Dropout 0%; (j\) DeeplabV3+ with ResNet-101—Dropout 50%; (k\) DeeplabV3+ with ResNet-152—Dropout 0%; (l\) DeeplabV3+ with ResNet-152—Dropout 50%.$

ctu-geoforall-lab-projects / phd-pesek-2024

readme