Page 174 - Kỷ yếu hội thảo khoa học lần thứ 12 - Công nghệ thông tin và Ứng dụng trong các lĩnh vực (CITA 2023)

158

Apply CNN-XGBoost into Weather Image Recognition

Tran Quy Nam and Phi Cong Huy

Posts and Telecommunications Institute of Technology, Hanoi, Vietnam

Abstract. This study implements some hybrid deep learning network models
for weather image classification. This study proposes to apply a hybrid model,
namely CNN-XGBoost model to test its performance, in comparison with other
simple Convolutional Neural Network (CNN) model with softmax and in
addition with the other hybrid models, namely CNN-SVC, CNN-Decision Tree,
CNN-AdaBoost, Multi-layer Perceptron Classifier which are all applied into the
same problem of weather image classification. The models apply an identical
test dataset which is a set of 11 different image classes that are collected from
different resources of weather images with various kinds of weather
phenomena. The test results show that the CNN-XGBoost gives the best results,
which is suitable for application in evaluating weather images. The aim of this
study is to check whether what kind of hybrid deep learning has the best
performance in the problem of weather image classification, not focus on
accuracy improvement of the deep learning models in classification problem.

Keywords: weather, image, CNN, XGBoost.

1 Introduction

The weather is changing nowadays which has a large impact on human life and socio-
economic development of many countries in the world. The correct recognition of
weather phenomenon is one of important factors to support our lives and nature
development. There are some ways to recognize weather phenomenon, such as
measurement of temperature, atmosphere, observational data collected by Doppler
radar, weather satellites, and other instruments such as weather balloon to measure
atmospheric parameters... The weather models use some mathematical and statistical
equations, along with new and past weather data, to provide informative guidance. In
computer science, the development of computer vision system has achieved great
success in many areas, such as image processing with high accuracy has already got
many applications in surveillance, navigation, driver assistance system
The automatic methodology of weather image classification through AI (Artificial
Intelligence) technology can help people to achieve sustainable development. The
accurate processing and identification of weather images taken from drone or camera
observation stations is an important method in weather forecasting, environmental
assessments, warning dangerous transportation... In terms of environmental
assessments, it is important to classify the respective weather phenomenon to alarm

CITA 2023 ISBN: 978-604-80-8083-9