Loading…
Influence of Different Activation Functions on Deep Learning Models in Indoor Scene Images Classification
The success of deep learning in the field of computer vision and object recognition has made significant breakthroughs, especially in improving recognition accuracy. The scene recognition algorithms have been evolved over the years because of the developments in machine learning and deep convolution...
Saved in:
Published in: | Pattern recognition and image analysis 2022-03, Vol.32 (1), p.78-88 |
---|---|
Main Authors: | , |
Format: | Article |
Language: | English |
Subjects: | |
Citations: | Items that this one cites Items that cite this one |
Online Access: | Get full text |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Summary: | The success of deep learning in the field of computer vision and object recognition has made significant breakthroughs, especially in improving recognition accuracy. The scene recognition algorithms have been evolved over the years because of the developments in machine learning and deep convolution neural networks (DCNNs). In this paper, the classification of indoor scenes using three deep learning models, namely, ResNet, MobileNet, and EfficientNet is attempted. The influence of activation functions on classification accuracy is being explored. Three activation functions, namely, tanh, ReLU, and sigmoid, are deployed in the work. The MIT-67 indoor dataset is split into scenes with and without people to test its effect on the accuracy of classification. The novelty of the work includes splitting the dataset, based on the spatial layout and segregating, into two groups, namely, with people and without people. Amongst the three pre-trained models, EfficientNet has given good results. |
---|---|
ISSN: | 1054-6618 1555-6212 |
DOI: | 10.1134/S1054661821040039 |