Multi-class Weather Classification using EfficientNet-B4 with Attention

Weather classification has long been a crucial area of study for weather monitoring systems. However, it can be difficult to determine the weather from a single image because the weather is constantly changing due to a variety of factors. Despite investing a lot of time and money into manually extra...

Full description

Saved in:
Bibliographic Details
Main Authors: Yang, Anjie, Toe, Teoh Teik, Ran, Zihan, Xiao, Shuhan
Format: Conference Proceeding
Language:English
Subjects:
Attention Mechanism
CNN
Computational modeling
Computer vision
Convolutional neural networks
Costs
Data models
EfficientNet
Feature extraction
Machine Learning
Robustness
Weather Classification
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Weather classification has long been a crucial area of study for weather monitoring systems. However, it can be difficult to determine the weather from a single image because the weather is constantly changing due to a variety of factors. Despite investing a lot of time and money into manually extracting and changing the features of conventional models, researchers have had little success in achieving accuracy that is satisfactory. Recently, with the advancement of artificial intelligence in computer vision area, researchers have attempted to address the problem with new approaches, such as convolutional neural network (CNN). In this study, we built our classification model based on EfficientNet-B4, then improved the performance by adding Attention mechanism to it. In terms of accuracy and cost, our model performs better than the earlier models. Meanwhile, the model exhibits greater robustness in a variety of scenarios when using data augmentation.
ISSN:2327-0594
DOI:10.1109/ICSESS54813.2022.9930176