Scribd
Upload
9 views30 pages

Group 14

This document presents a case study on object classification using deep learning, specifically Convolutional Neural Networks (CNNs), applied to the CIFAR-10 dataset. The project highlights the advantages of deep learning over traditional methods, outlines the dataset's characteristics, and describes the CNN architecture and training process. The model achieved an accuracy of 85.196%, with suggestions for future enhancements including the use of ResNet/VGG and transfer learning.

Uploaded by

sstories569
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
9 views30 pages

Group 14

This document presents a case study on object classification using deep learning, specifically Convolutional Neural Networks (CNNs), applied to the CIFAR-10 dataset. The project highlights the advantages of deep learning over traditional methods, outlines the dataset's characteristics, and describes the CNN architecture and training process. The model achieved an accuracy of 85.196%, with suggestions for future enhancements including the use of ResNet/VGG and transfer learning.

Uploaded by

sstories569
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
You are on page 1/ 30

OBJECT CLASSIFICATION

USING DEEP LEARNING (CNN)


- A Case Study on CIFAR-10 Dataset

Supervisor: Fahima Tabassum, Ph.D


Professor, Institute of Information Technology
TEAM MEMBERS

Md. Sohel Rana Anik Kaisar Md. Ashraful


ID: 2065 ID: 2058 Islam
IIT, JU IIT, JU ID: 2076
IIT, JU

2
HELLO
!
SEEKING ATTENTION!!!!

3
INTRODUCTIO
N
4
CONTINUE…
•Object classification is a fundamental task in computer
vision.
•Deep Learning, especially CNNs, has revolutionized image
recognition.
•This project focuses on classifying images in the CIFAR-10

dataset.

5
What is Object
Classification?

6
CONTINUE…

•It is the process of identifying and labeling objects in an image.


•Applications:
•Autonomous vehicles (detect pedestrians, traffic signs).
•Medical Imaging (tumor detection).
•Security & Surveillance (face recognition).
•Retail & E-commerce (product recognition).

7
Why Use Deep Learning
for Object Classification?

8
CONTINUE…

• Traditional Approaches (SIFT, HOG, etc.) struggle with complex


images.
Deep Learning Advantages:

• Learns features automatically from images.

• Can handle large datasets and variations in lighting, orientation, etc.

• Highly scalable using GPUs.

9
UNDER
STANDING

CIFAR-10
DATASET 10
Overview of CIFAR-
10 Dataset
•Contains 60,000 images (32×32 pixels).
•10 classes: Airplane, Automobile, Bird, Cat, Deer, Dog, Frog, Horse,
Ship, Truck.
•50,000 images for training, 10,000 images for testing.

11
CHALLENGES WITH
CIFAR-10
∎ Low-resolution ∎ Similar-looking ∎ Prone to
images (32x32 classes (e.g., cat overfitting due
pixels). vs. dog, airplane to small dataset
vs. bird). size.
SAMPLES IMAGES
FROM CIFAR-10
• Each class contains 6,000 images, ensuring
a balanced dataset

13
DEEP
LEARNIN
G AND
CNN
BASIS
14
CONTINUE…

What is Deep Leaning?? Introduction to CNN

•Subset of Machine Learning using •CNNs are specialized neural networks


artificial neural networks. for image processing.
•Uses multiple layers (hidden layers) to •Consist of convolutional layers,
extract meaningful patterns from data. pooling layers, and fully connected
layers.

15
CNN
ARCHITECTURE
Convolutional Layers: Extract features (edges, textures).
Pooling Layers: Reduce dimensions and computation.
Fully Connected Layers: Make the final classification.

Activation Functions (ReLU, Softmax): Introduce non-linearity.

16
Data
Preprocessing &
Augmentation

17
DATA PREPROCESSING STEPS

•Convert images to tensors


(numerical arrays).
•Normalize pixel values to
[0,1] for better learning.
•Split dataset into training and
validation sets.

18
DATA AUGMENTATION

• Techniques Used:Rotation
• Flipping
• Cropping
• Brightness adjustments

19
BUILDING THE
CNN MODEL
DIMENSIONALITY
FEATURE EXTRACTION CLASSIFICATION
REDUCTION

20
CONTINUE…
•Input: 32×32×3 (RGB image).
•Layers:
•2 Convolutional layers (feature
extraction).
•2 Pooling layers (dimensionality
reduction).
•3 Fully Connected layers (classification).

21
TRAINING AND
OPTIMIZING THE
MODEL
22
CONTINUE…
Training the Model DEFINING LOSS FUNCTION AND
•Use GPU acceleration if available.
OPTIMIZER
•Backpropagation and optimization at each epoch.

23
MODEL
EVALUATION
Evaluating the model

24
TESTING
IMAGES
frog truck truck deer
automobile automobile bird horse
ship cat deer horse
horse bird truck truck
truck cat bird frog
deer cat frog frog
bird frog cat dog
deer airplane airplane truck

VS
PREDICTION
IMAGES
frog truck truck deer
automobile automobile frog
horse
ship horse deer horse
horse truck truck ship
truck cat frog frog
deer cat frog frog
horse cat cat automobile
horse airplane airplane truck
RESULT & DISCUSSION

85.196%
ACCURACY
CONCLUSION
•CNNs are powerful for object classification.
•Successfully classified CIFAR-10 dataset with high
accuracy.

27
FUTURE
ENHANCEMENTS

∎ Implement ResNet/VGG for higher accuracy.


∎ Explore Transfer Learning on larger datasets.

28
REFERENCES

•CIFAR-10 Dataset:
https://www.cs.toronto.edu/~kriz/cifar.html
•Deep Learning Frameworks: TensorFlow, PyTorch.

29
THANK
THANK YOU
YOU!!!
Q??

30

You might also like