2024 2nd International Conference on Artificial Intelligence and Machine Learning Applications Theme: Healthcare and Internet of Things

(AIMLA) | 979-8-3503-4922-1/24/$31.00 ©2024 IEEE | DOI: 10.1109/AIMLA59606.2024.10531390

CLOUD CLASSIFICATION AND RAINFALL PREDICTION

Kummara Charanya Kolla Ushasri Kurugunta Pragna

ECE, Kalasalingam Academy ECE, Kalasalingam Academy ECE, Kalasalingam Academy
of Research and Education of Research and Education Of Research and Education
Madurai, India Madurai, India Madurai, India
[email protected] [email protected] [email protected]

Manoranjan Kumar Dr. S.Diwakaran Paladugu Srikanth

Nanochipskills(Venky Labs) Associate Professor, ECE,Kalasalingam Academy
Bengaluru_560001 ECE, Kalasalingam Academy of Research and Education
Karnataka, India. of Research and Education Madurai, India
[email protected] Madurai, India. [email protected]
[email protected]

Abstract—Taking up over 60% of the planet's timeliness, and scalability. The advent of advanced
surface, clouds are essential in controlling the machine learning remote sensing technologies, and the
planet's radiation budget because they affect both availability of vast amounts of ground-based imagery
long- and short-wave warming and cooling. The past and meteorological information have opened new
several years have seen an increase in interest in methods for improving cloud classification and rainfall
ground-based autonomous cloud photo observation prediction. This project seeks to utilize the potential of
techniques. Ten commonly occurring cloud modern technology to tackle the challenges associated
formations and contrails are distinguished using with cloud classification and rainfall prediction. By
Convnet (CNNs), utilizing the latest developments in developing a robust and efficient system that merges the
deep learning. The efficiency and precision of machine learning algorithms in remote sensing data, and
Convnet cloud classification have not undergone real-time processing capabilities, we aim to provide
comprehensive examination, despite their more accurate and timely information about clouds and
impressive picture classification results. For the rainfall pattern. Machine learning remote sensing
purpose of determining a potentially dangerous technologies, and the availability of vast amounts of
weather activity, convective cloud detection ground-based imagery.
technologies in particular are extremely precise and
automated cloud classification techniques. II. LITERATURE REVIEW
Predictions regarding rainfall are also greatly
having an effect by the quantity and variety of Tuominen and Tuononen [1] research was performed on
clouds in the sky. That's why research on clouds ground-based cloud images, presenting a method that
draws in academics from diverse backgrounds and combines a neural network (NN) with the Lucas-Kanade
is regarded as part of the most significant and method to extract information about cloud coverage and
fascinating. movement. This method effectively detects clouds
across a range of conditions, including clear skies, dark
Keywords— CNN, GUI cloud cover, and varying degrees of cloudiness.
Dev et al. [2] SWIMSEG was launched, offering a
I. INTRODUCTION
exhaustive publicly available database of sky and cloud
Clouds are an integral part of Earth's atmosphere, images with partitioning capability, showcasing 1013
influencing our planet's climate, weather patterns, and images taken by a high-definition sky capture device
ecosystems. Understanding clouds and their behaviour imagery. Furthermore, a fully learning-based division
is crucial for a diverse array of applications, from frame work is being considered for the splitting of the
predicting rainfall to managing agricultural resources sky images including the evaluation and selection of
and mitigating the influence of extreme weather events.
suitable colour channels using two distinct datasets of
Cloud classification and rainfall prediction are
fundamental components meteorological science, with cloud/sky images.
noteworthy repercussions across various sectors, Gogoi and Devi [3] utilized image processing methods
including agriculture, water resource management and to analyse cloud images, extracting information such as
disaster preparedness traditional methods of cloud cloud category/state and sky through subdivision of
classification and rainfall prediction have relied on cloud evaluation procedure. By including parameters
manual observations and conventional meteorological such as height, colour, altitude, classification, and
techniques. While these methods have been valuable appearance, evaluated rainfall status. Additionally
they often have limitations in the terms of accuracy, various rainfall prediction methods, including

orized licensed use limited to: AMRITA VISHWA VIDYAPEETHAM AMRITA SCHOOL OF ENGINEERING. Downloaded on November 07,2024 at 13:18:59 UTC from IEEE Xplore. Restrictions a
 traditional statistical and numerical approaches were III. METHODOLOGY
explored in the discussion.
Zhang et al. [4] CloudNet, a novel Convnet based Convolutional Neural Networks (CNNs) can have an
algorithm for classifying ground-based cloud images impact tool for cloud classification and rain prediction
was introduced. The CCSN dataset, composed of digital when utilized in combination with meteorological data
cloud images taken from the ground, contains a large and ground-based images. Convolutional Neural
number of entries, compared to previous datasets, such Networks (CNNs) are robust tools for cloud
as SWIMCAT. This dataset presents a anew cloud type classification and rain prediction when they work with
meteorological data and ground-based images.
known as contrails, which are line shaped clouds
Assemble a sizable collection of aerial photos taken
produced by aircraft engine emissions, with the aim of from the ground that incorporates various cloud types
studying their impact on global warming. and meteorological data such as wind, temperature, and
Lai et al. [5] performed research on the classification of humidity. Ensure that the data is suitably classified for
cloud images focusing on eight distinct cloud types. By cloud types and contains information on historical
utilizing both a ten-layer and a seven-layer CNN, rainfall records. Make changes such as rotation, scaling,
evaluated the outcomes and noted a ten percent and flipping to enhance the diversity of the dataset. This
improvement in test accuracy upon reducing the layers. improves the model's ability to generalize. Create
The experiments covered 1400 epochs with batch training, validation, and testing sets from the dataset to
normalization integrated into the CNN after the Relu evaluate the CNN's efficacy. Our system will
function to address over fitting concerns. incorporate cloud data, meteorological conditions, and
García Fernández [6] This study offers a comprehensive historical rainfall records into predictive models.
examination of CNN performance in autonomously Machine learning methods will be employed and
classifying ground-based sky images. Two derivative numerical weather prediction techniques for
datasets are generated from the original. One containing determining the amount and patterns of rainfall. Several
the cloud images of two categories and the other six models and data sources will be combined in our
types. This experiments involve activating and ensemble forecasting techniques to increase precision
deactivating sub sampling layers and testing various and dependability. By helping to take prediction
uncertainties into account, this will improve the overall
optimizers (SGD and Adagrad) to attain satisfactory
standard of the forecast.
precision.
Muruvvet Kalkan [7] Commonly we experienced false IV. PROPOSED SYSTEM
prediction of rainfall so we need to make the weather
predictions as accurate as possible by F1-Score, namely We suggest a thorough method for classifying clouds
VGG-16 prediction. This method proposes to compute and forecasting rainfall. To make better precision and
these estimations with pre-trained based models. promptness of the cloud and rainfall forecasts, our
Anuska Sarkar, Riya [8] As we know prediction of the suggested system combines cutting-edge deep utilizing
rainfall is greatly influenced by the clouds. In this CNN deep learning in conjunction with real-time data
is used to classify the clouds and predict the rainfall. processing. Convnet (CNNs) and HSVs. will employed
Based on the precipitation the clouds are divided into for create and train models in machine learning it will
three types such as no rain to low rain, low rain to be able to recognize various cloud types and their
medium rain, and lastly medium to high rainfall. attributes. The key features used to differentiate
Teena Varma [9] introduced an algorithm called remote between different cloud types in the classification
sensing is nothing but the science and application of process include color, texture, shape, size, altitude, and
obtaining the given information regarding an object appearance. These features are extracted through image
without actually coming into contact withit. In more processing techniques such as segmentation and color
suitable terms for our understanding, remote sensing is analysis.Predictive models created by our system will
incorporate historical rainfall records, atmospheric
a mechanism for samplifying reflected and emitted
conditions, and data on cloud classification. We intend
electro-magnetic(EM) radiation from the earths utilizing numerical weather prediction techniques in
terrestrial and acquatic ecosystems and atmosphere. combining machine learning techniques to assess the
Sushant Jadhav [10] In this machine learning process is intensity and patterns of rainfall. Several models and
used to carry out the complex tasks in the nature. K- data sources will be combined in our ensemble
means clustering is used to classify the clouds in to high forecasting techniques to increase precision and
level clouds medium level clouds and low level clouds. dependability. By helping to take prediction uncertainty
Manzo and Pellino [11] A framework for cloud based into account, this will improve the forecast's overall
image identification is developed, incorporating various quality.
CNN methods through transfer learning. During
classification, a voting system is utilised, taking into
account each methods prediction likelihood.

orized licensed use limited to: AMRITA VISHWA VIDYAPEETHAM AMRITA SCHOOL OF ENGINEERING. Downloaded on November 07,2024 at 13:18:59 UTC from IEEE Xplore. Restrictions a
 V. BLOCK DIAGRAM including name and email address. Provide the
password under the requirements needed to ensure its
START
high level of security.

Load image to the

dataset.

Data preprocessing and data

analysis

Training
Fig3. Admin Login Page
After entering the password and username provided,
Images are classified by CNN
click "Login."

- Output

Classify

sunny Rainy Partially etc

cloudy Fig4. Select Image
Select an image from the dataset. The images were
Fig1. Block Diagram captured using a fish-eye lens and hemispheric dome
Here firstly we need to undertake the images of the camera from the ground.
cloud from the ground. Then we need to load images to
the data-set. Now we have to analyse and preprocess the
data. Again we have to instruct the data and then
proceed for the categorization of images by using
convnet algorithm. Now we will get the output. From
the output we can classify the clouds as Sunny, Rainy,
Partly Cloudy.
VI. OUTPUT

Fig5. Submit Image

The screen is showing the selected image. To discover
the sky condition and rainfall, click the submit button
now. To discover the weather, upload the most recent
photos of the sky.

Fig2. Admin Register Page

The URL 27.0.0.1:5000/admin login results in the login
page where the admin can view the rainfall. The admin
can also view the names and localities of new users. Go
to the login page and enter the login user's details, Fig6. Output

orized licensed use limited to: AMRITA VISHWA VIDYAPEETHAM AMRITA SCHOOL OF ENGINEERING. Downloaded on November 07,2024 at 13:18:59 UTC from IEEE Xplore. Restrictions a
 VII. CONCLUSION
The output is displayed in the graphical user interface
(GUI) with cloud characteristics such as cloud type and Based on images taken from the ground, CNNs can be
estimated rainfall. There are four categories for the utilized to automatically and precisely categorized
different kinds of clouds. This allows for a deeper
cloud status: Sunny, White, Blue and Grey.
understanding of cloud behaviour and how it affects
weather patterns. Predicting rainfall more accurately
and promptly is made possible by the integration of data
with meteorological data and numerical weather
prediction models. This is very helpful for disaster
preparedness, agriculture, +and water resource
management. The system's ability to process data in
almost real-time guarantees that rainfall forecasts and
cloud classification are current, giving decision-makers
across various sectors vital information. Since the
methodology combines scalability with effective data
Fig7. Output
integration, it is adaptable enough to handle large-scale
Select the various cloud types to learn about the various
data processing and a range of data sources, including
kinds of rainfall from the provided image. The HSV
radar data, satellite imaging, and ground-based
colour space utilization fixed for the colour range, so for
observations. Cloud classification and rainfall
example, blue sky low = [90,50,50].
prediction are now available to a wider range of users,
The pixels are employed to calculate the cloud area.
including meterological farmers and disaster
Rainfall in mm = cloud area * 0.1 is the formula used to
management organizations, thanks to generate an
estimate rain.
intuitive dashboard.

TABLE VIII. REFERENCES

CLOUD TYPE CLOUD AREA AMOUNT OF 1) Auska, Sankar, Chethan, show, Sourishmitra,” Cloud
(in pixels) RAINFALL
classification for prediction of rainfall using CNN”,
BLUE SKY 4689.5 468.95 2013.
2) TingtinZhu, Liang Wei,Yiren Guo,” Cloud
WHITE SKY 7723.5 772.35 classification of ground-based cloud images derived
GREY SKY 9432.9 953.29 on CNN”2021.
3) Shraddha R.Palaskar,Aishwarya R Jagtap,”
SUNNY DAY 2388.5 238.85 Classification of clouds for prediction of
rainfall”,2017 2(10),242-244.
4) Jiang Zhang,Liu pu,“A ground based cloud
GRAPH classification method with a convnet” 2018
5) Geeta Mahadeo Ambildhuke,Barnali Gupta Banik
,“An efficient method of predict rainfall based on
ground based cloud images”,2021.
6) Kumar Abhishek,Abhay Kumar,“A rainfall
predictionmodel using ANN”,2012
7) Niyati Salot, Dr. Priya Swaminarayan"Classification
of cloud types for rainfall forecasting”,2015 ,07
(01),2626-2629.
8) Zijun I and hoiio kong" Neural Network-Based
Identification regarding types of clouds from
Ground-Based Images of Cloud Layers”,
13(7),2023.
9) Pu liu, Feng zhang,“CloudNet: cloud classification
employing deep convolution neural networks for
ground based observations ”, 45(16),2018.
Fig8. Graph Output

orized licensed use limited to: AMRITA VISHWA VIDYAPEETHAM AMRITA SCHOOL OF ENGINEERING. Downloaded on November 07,2024 at 13:18:59 UTC from IEEE Xplore. Restrictions a
 10) Sourav deyroy,Dipak Hrishi Das,“Cloud Pattern
Classification making use of convolution neural
networks for rainfall prediction”,2021,83-90.
11) XinZhao, Haikum Wei,“3D-CNN-based
prediction”,volume:181,510-518,2018.

12) Wen Zhu,tianliang extracting features from ground

cloud images derived from ground observations for
direct normal irriadiance estimation chen,
“Classification of Ground-Based Cloud Images by
Improved Combined Convolutional
Network”,12(3),2022.
13) Yuzhu Tang and Jianyu Chen "Improving Cloud
Type categorization of Ground-Based Images Using
Region Covariance Descriptors”,2018.
14) Lysha M.Matsunobu, HugoT.C.Pedro, “Cloud
detecting with convolutional neural networks on
remote sensing images”,volume:230, 1020-
1032,2021.
15) Min Wang, Zhihao ZhuangKangWang,“Intelligent
classification of ground-based visibile cloud images
using a transfer convnet and fine tuning”,29(25)
4117641190,2021.
16) Niyati Salot, D. P,“An exploration of rainfall
prediction through Image Processing Technique” I03
(02), 126-132,2015.
17) Ezzat, D., Hassanien, A.E., Taha, M.H.N.,
Bhattacharyya,“Transfer learning with a fine-tuned
CNN model for classifying augmented natural
images”,843-856,2020.
18) K.Kaviarasu, P.Sujith and Mr. G. Ayaappan
“Prediction Of Rainfall Using Image Processing”
2010.

orized licensed use limited to: AMRITA VISHWA VIDYAPEETHAM AMRITA SCHOOL OF ENGINEERING. Downloaded on November 07,2024 at 13:18:59 UTC from IEEE Xplore. Restrictions a