| Title |
Exploring convolutional architecture capabilities for image classification tasks with insufficient amount of data / |
| Authors |
Matsevytyi, Andrii ; Rudžionis, Vytautas Evaldas |
| Full Text |
|
| Is Part of |
IVUS 2024: Information society and university studies 2024: proceedings of the 29th international conference on information society and university studies, Kaunas, Lithuania, May 17, 2024.. Aachen : CEUR-WS. 2024, p. 99-108 |
| Keywords [eng] |
convolutional neural networks ; image classification ; kernel ; pooling ; accuracy metrics ; optimization ; low quality dataset ; small dataset |
| Abstract [eng] |
Nowadays Convolutional Neural Networks are used everywhere from facial recognition to malware detection and flat evaluation and are considered to bring significant changes to computer vision. They introduce solutions of such problems as insufficient and low-quality dataset. However, they tend to possess same problems as other Machine Learning and Deep Learning techniques. The paper considers and analyses the most commons methods for image classification, involving usage of feed-forward convolutional architecture. The object of the study is self- collected dataset, consisting of 7 classes, that provide of low-, middle- and highlevel features. The subject of the study is to explore the capabilities of CNNs key architecture blocks and their combinations. |
| Published |
Aachen : CEUR-WS |
| Type |
Conference paper |
| Language |
English |
| Publication date |
2024 |
| CC license |
|
