CNN for Computer Vision tasks
Object is a neural networks in the field of computer vision and data analysis. The article explores the fundamental principles and aspects underlying the functioning of neural networks. Among the identified limitations are the complexity of tuning hyperparameters and computational costs associated...
Saved in:
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Igor Sikorsky Kyiv Polytechnic Institute
2024-03-01
|
| Series: | Adaptivni Sistemi Avtomatičnogo Upravlinnâ |
| Subjects: | |
| Online Access: | https://asac.kpi.ua/article/view/302422 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Object is a neural networks in the field of computer vision and data analysis. The article explores the fundamental principles and aspects underlying the functioning of neural networks. Among the identified limitations are the complexity of tuning hyperparameters and computational costs associated with increasing the network's depth and the size of training data. The aim of the study is to analyze contemporary solutions related to convolutional neural networks (CNN) to select an optimal topology that enhances addressing typical computer vision tasks. The article covers essential methods such as convolutional, pooling, fully connected, and inception layers. Different types of convolutional operations (expanded, partial, strided) and various CNN models (LeNet, AlexNet, VGGNet, GoogLeNet, ResNet) are also examined. After reaching the goal or objectives, it is suggested to employ various methods such as dimensionality reduction, batch normalization, data augmentation, utilization of advanced optimization methods, diverse activation functions, as well as combinations of these approaches.
Ref. 10, pic. 4, tabl. 1
|
|---|---|
| ISSN: | 1560-8956 2522-9575 |