A Convolutional Neural Network (CNN) is a type of neural network that can process grid-like data like images. It works by applying filters to the input image to extract features at different levels of abstraction. The CNN takes the pixel values of an input image as the input layer. Hidden layers like the convolution layer, ReLU layer and pooling layer are applied to extract features from the image. The fully connected layer at the end identifies the object in the image based on the extracted features. CNNs use the convolution operation with small filter matrices that are convolved across the width and height of the input volume to compute feature maps.

1. Convolution Neural
Network Tutorial

2. How image recognition works?
Do you know how Deep Learning recognizes the objects in an image?
It does it using a Convolution Neural Network
Pixels of image fed as input
Dog
Bird
Cat
Hidden Layers
Input Layer Output Layer

3. How image recognition works?
Let’s see how CNN identifies the image of a bird
Pixels of image fed as input
Dog
Bird
Cat
Hidden Layers
Input Layer Output Layer
Input layer accepts the pixels of the image as
input in the form of arrays
1
2
1
9
2
1
7 40 2
30 11 35 70 11
1
4
3307552613
60 45 50 10 89 23

4. How image recognition works?
Let’s see how CNN identifies the image of a bird
Pixels of image fed as input
Dog
Bird
Cat
Hidden Layers
Input Layer Output Layer
Hidden layers carry out feature extraction by
performing certain calculation and manipulation
1
2
1
9
2
1
7 40 2
30 11 35 70 11
1
4
3307552613
60 45 50 10 89 23

5. How image recognition works?
Let’s see how CNN identifies the image of a bird
Pixels of image fed as input
Dog
Bird
Cat
Hidden Layers
Input Layer Output Layer
There are multiple hidden layers like
Convolution layer, ReLU layer, Pooling layer,
etc that perform feature extraction from the
image
Convolution Layer
This layer uses a matrix filter and
performs convolution operation to
detect patterns in the image
1 0 1
10 0
1 0 1
Matrix Filter

6. How image recognition works?
Let’s see how CNN identifies the image of a bird
Pixels of image fed as input
Dog
Bird
Cat
Hidden Layers
Input Layer Output Layer
There are multiple hidden layers like
Convolution layer, ReLU layer, Pooling layer,
etc that perform feature extraction from the
image
ReLU
ReLU activation function is applied
to the convolution layer to get a
rectified feature map of the image

7. How image recognition works?
Let’s see how CNN identifies the image of a bird
Pixels of image fed as input
Dog
Bird
Cat
Hidden Layers
Input Layer Output Layer
There are multiple hidden layers like
Convolution layer, ReLU layer, Pooling layer,
etc that perform feature extraction from the
image
Pooling
Pooling layer also uses multiple
filters to detect edges, corners,
eyes, feathers, beak, etc

8. How image recognition works?
Let’s see how CNN identifies the image of a bird
Pixels of image fed as input
Dog
Bird
Cat
Hidden Layers
Input Layer
1
2
1
9
2
1
7 40 2
30 11 35 70 11
1
4
3307552613
60 45 50 10 89 23 Finally there is a fully connected layer that
identifies the object in the image
Output Layer

9. What’s in it for you?
How CNN recognizes images?
What is Convolution neural network?
Use case implementation using CNN
Introduction to CNN
Layers in convolution neural network

10. Introduction to CNN
Yann LeCun
Pioneer of Convolution Neural Network
Director of Facbook’s AI Research Group
Built the first Convolution Neural Network called LeNet in 1988
It was used for character recognition tasks like reading zip codes, digits

11. Introduction to CNN
Yann LeCun
Pioneer of Convolution Neural Network
Director of Facbook’s AI Research Group
Built the first Convolution Neural Network called LeNet in 1988
It was used for character recognition tasks like reading zip codes, digits

12. Introduction to CNN
Yann LeCun
Pioneer of Convolution Neural Network
Director of Facbook’s AI Research Group
Built the first Convolution Neural Network called LeNet in 1988
It was used for character recognition tasks like reading zip codes, digits

13. Introduction to CNN
Yann LeCun
Pioneer of Convolution Neural Network
Director of Facbook’s AI Research Group
Built the first Convolution Neural Network called LeNet in 1988
It was used for character recognition tasks like reading zip codes, digits

14. What is a Convolution Neural Network?
CNN is a feed forward neural network that is generally used to analyze visual images by processing data
with grid like topology. A CNN is also known as a “ConvNet”
Orchid
Rose
Flowers of 2 varieties
(Orchid/Rose)
Identifies the flowers
Hidden Layers
Input Layer
Output Layer

15. What is a Convolution Neural Network?
CNN is a feed forward neural network that is generally used to analyze visual images by processing data
with grid like topology. A CNN is also known as a “ConvNet”
Convolution operation forms the basis of any
Convolution Neural Network
In CNN, every image is represented in
the form of arrays of pixel values
Real Image of the digit 8 Represented in the form
of an array
0 0 1 1 0 0
0
0
0
01 1
1
1
1 10
0
0
0 0
1
0
1
0 0
0 0
0 0
Digit 8 represented in the form of
pixels of 0’s and 1’s

16. What is a Convolution Neural Network?
Let’s understand the convolution operation using 2 matrices a and b of 1 dimension
a b* Sum the product
b = [1, 2, 3]
a = [5, 3, 2, 5, 9, 7]
b = [1, 2, 3]
a = [5, 3, 7, 5, 9, 7]
Matrix a and b
Convolution

17. What is a Convolution Neural Network?
Let’s understand the convolution operation using 2 matrices a and b of 1 dimension
a b* Sum the product
b = [1, 2, 3]
a = [5, 3, 2, 5, 9, 7]
[5, 6, 6]
a b* = [17, ]
Multiply the arrays
element wise
17
b = [1, 2, 3]
a = [5, 3, 7, 5, 9, 7]
Matrix a and b
Convolution

18. What is a Convolution Neural Network?
Let’s understand the convolution operation using 2 matrices a and b of 1 dimension
a b* Sum the product
b = [1, 2, 3]
a = [5, 3, 2, 5, 9, 7]
a b* = [17, 22 ]
Multiply the arrays
element wise
17
b = [1, 2, 3]
a = [5, 3, 7, 5, 9, 7]
Matrix a and b
Convolution
[5, 6, 6]
[3, 4, 15] 22

19. What is a Convolution Neural Network?
Let’s understand the convolution operation using 2 matrices a and b of 1 dimension
a b* Sum the product
b = [1, 2, 3]
a = [5, 3, 2, 5, 9, 7]
a b* = [17, 22, 39,…….. ]
Multiply the arrays
element wise
17
b = [1, 2, 3]
a = [5, 3, 7, 5, 9, 7]
Matrix a and b
Convolution
[5, 6, 6]
[3, 4, 15] 22
[2, 10, 27] 39
………
………

20. How CNN recognizes images?
image for the symbol image for the symbol /
Consider the following 2 images:
When you press , the above image is processed

21. How CNN recognizes images?
image for the symbol image for the symbol /
Consider the following 2 images:
When you press /, the above image is processed

22. How CNN recognizes images?
Image represented in the
form of a matrix of numbers
000000
11
0000000
1
1
1
0
1
1
0
0 0 0 0 0
0 0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0
0
000000
0 0 1 0
Real Image Represented in the form of black
and white pixels

23. Layers in Convolution Neural Network
Pooling Layer
ReLU LayerConvolution Layer
CNN
Fully Connected Layer
1 2
34

24. Convolution Layer
A Convolution Layer has a number of filters that perform convolution operation
Every image is considered as a matrix of pixel values.
Consider the following 5 5 image whose pixel values are only 0 and 1*
1 0 1
10 0
1 0 1
Filter
1 1 1 0 0
0
0
0
0
1
0
0
1
1 1 0
1 1 1
1 1 0
1 0 0
Image
pixels
4 3 4
2 4 3
2 3 4
Convolved
Feature
Sliding the filter matrix over the
image and computing the dot
product to detect patterns

25. Convolution Layer
A Convolution Layer has a number of filters that perform convolution operation
Every image is considered as a matrix of pixel values.
Consider the following 5 5 image whose pixel values are only 0 and 1*
1 0 1
10 0
1 0 1
Filter
1 1 1 0 0
0
0
0
0
1
0
0
1
1 1 0
1 1 1
1 1 0
1 0 0
Image
pixels
4
x
1
x0 x
1
x0 x
1
x0
x
1
x0 x
1
1 1 1
0 1 1
0 0 1
Convolved
Feature
Sliding the filter matrix over the
image and computing the dot
product to detect patterns

26. Convolution Layer
A Convolution Layer has a number of filters that perform convolution operation
Every image is considered as a matrix of pixel values.
Consider the following 5 5 image whose pixel values are only 0 and 1*
1 0 1
10 0
1 0 1
Filter
1 1 1 0 0
0
0
0
0
1
0
0
1
1 1 0
1 1 1
1 1 0
1 0 0
Image
pixels
4 3
x
1
x0 x
1
x0 x
1
x0
x
1
x0 x
1
1 1 0
1 1 1
0 1 1
Convolved
Feature
Sliding the filter matrix over the
image and computing the dot
product to detect patterns

27. Convolution Layer
A Convolution Layer has a number of filters that perform convolution operation
Every image is considered as a matrix of pixel values.
Consider the following 5 5 image whose pixel values are only 0 and 1*
1 0 1
10 0
1 0 1
Filter
1 1 1 0 0
0
0
0
0
1
0
0
1
1 1 0
1 1 1
1 1 0
1 0 0
Image
pixels
4 3 4
x
1
x0 x
1
x0 x
1
x0
x
1
x0 x
1
1 0 0
1 1 0
1 1 1
Convolved
Feature
Sliding the filter matrix over the
image and computing the dot
product to detect patterns

28. Convolution Layer
A Convolution Layer has a number of filters that perform convolution operation
Every image is considered as a matrix of pixel values.
Consider the following 5 5 image whose pixel values are only 0 and 1*
1 0 1
10 0
1 0 1
Filter
1 1 1 0 0
0
0
0
0
1
0
0
1
1 1 0
1 1 1
1 1 0
1 0 0
Image
pixels
4 3 4
2
x
1
x0 x
1
x0 x
1
x0
x
1
x0 x
1
0 1 1
0 0 1
0 0 1
Convolved
Feature
Sliding the filter matrix over the
image and computing the dot
product to detect patterns

29. Convolution Layer
A Convolution Layer has a number of filters that perform convolution operation
Every image is considered as a matrix of pixel values.
Consider the following 5 5 image whose pixel values are only 0 and 1*
1 0 1
10 0
1 0 1
Filter
1 1 1 0 0
0
0
0
0
1
0
0
1
1 1 0
1 1 1
1 1 0
1 0 0
Image
pixels
4 3 4
2 4
x
1
x0 x
1
x0 x
1
x0
x
1
x0 x
1
1 1 1
0 1 1
0 1 1
Convolved
Feature
Sliding the filter matrix over the
image and computing the dot
product to detect patterns

30. Convolution Layer
A Convolution Layer has a number of filters that perform convolution operation
Every image is considered as a matrix of pixel values.
Consider the following 5 5 image whose pixel values are only 0 and 1*
1 0 1
10 0
1 0 1
Filter
1 1 1 0 0
0
0
0
0
1
0
0
1
1 1 0
1 1 1
1 1 0
1 0 0
Image
pixels
4 3 4
2 4 3
x
1
x0 x
1
x0 x
1
x0
x
1
x0 x
1
1 1 0
1 1 1
1 1 0
Convolved
Feature
Sliding the filter matrix over the
image and computing the dot
product to detect patterns

31. Convolution Layer
A Convolution Layer has a number of filters that perform convolution operation
Every image is considered as a matrix of pixel values.
Consider the following 5 5 image whose pixel values are only 0 and 1*
1 0 1
10 0
1 0 1
Filter
1 1 1 0 0
0
0
0
0
1
0
0
1
1 1 0
1 1 1
1 1 0
1 0 0
Image
pixels
4 3 4
2 4 3
2
x
1
x0 x
1
x0 x
1
x0
x
1
x0 x
1
0 0 1
0 0 1
0 1 1
Convolved
Feature
Sliding the filter matrix over the
image and computing the dot
product to detect patterns

32. Convolution Layer
A Convolution Layer has a number of filters that perform convolution operation
Every image is considered as a matrix of pixel values.
Consider the following 5 5 image whose pixel values are only 0 and 1*
1 0 1
10 0
1 0 1
Filter
1 1 1 0 0
0
0
0
0
1
0
0
1
1 1 0
1 1 1
1 1 0
1 0 0
Image
pixels
4 3 4
2 4 3
2 3
x
1
x0 x
1
x0 x
1
x0
x
1
x0 x
1
0 1 1
0 1 1
1 1 0
Convolved
Feature
Sliding the filter matrix over the
image and computing the dot
product to detect patterns

33. Convolution Layer
A Convolution Layer has a number of filters that perform convolution operation
Every image is considered as a matrix of pixel values.
Consider the following 5 5 image whose pixel values are only 0 and 1*
1 0 1
10 0
1 0 1
Filter
1 1 1 0 0
0
0
0
0
1
0
0
1
1 1 0
1 1 1
1 1 0
1 0 0
Image
pixels
x
1
x0 x
1
x0 x
1
x0
x
1
x0 x
1
1 1 1
1 1 0
1 0 0
4 3 4
2 4 3
2 3 4
Convolved
Feature
Sliding the filter matrix over the
image and computing the dot
product to detect patterns

34. ReLU Layer
Once the feature maps are extracted, the next step is to move them to a ReLU layer
1050-5-10
0
2
4
6
8
10
R(z) = max(0, z)
ReLU
Performs element wise
operation
Sets all negative pixels to 0 Introduces non-linearity to
the network
The output is a rectified
feature map

35. ReLU Layer
Real image is scanned in multiple convolution and ReLU layers for locating features

36. ReLU Layer
Real image is scanned in multiple convolution and ReLU layers for locating features

37. Note for the instructor
While explaining, please mention there are multiple Convolution, ReLU and Pooling layers
connected one after another that carry out feature extraction in every layer. The input
image is scanned multiple times to generate the input feature map.

38. Pooling Layer
The rectified feature map now goes through a pooling layer. Pooling is a down-sampling operation that
reduces the dimensionality of the feature map.
1
2
4
6
2 7
58
3 04
1 2 3 1
7
6 8
4 7
max pooling with 2x2 filters and
stride 2
Max(3, 4, 1, 2) = 4
Pooled feature map
Rectified feature map

39. Pooling Layer
Identifies the edges, corners and other features of the bird
Pooling layer uses different filters to identify different parts of the image like edges, corners, body,
feathers, eyes, beak, etc.

40. Pooling Layer
Structure of the Convolution Neural Network so far
1 1 1 0 0
0
0
0
0
1
0
0
1
1 1 0
1 1 1
1 1 0
1 0 0
Convolution Pooling
Input Image
Convolution Layer
Pooling Layer
ReLU

41. Flattening
6 8
4 7
Pooled feature map
6
8
4
7
Flattening
Flattening is the process of converting all the resultant 2 dimensional arrays from pooled feature map into
a single long continuous linear vector.

42. Flattening
Pooling Layer Input Layer
Flattening is the process of converting all the resultant 2 dimensional arrays from pooled feature map into
a single long continuous linear vector.
Flattening

43. Flattening
1 1 1 0 0
0
0
0
0
1
0
0
1
1 1 0
1 1 1
1 1 0
1 0 0
Convolution Pooling
Input Image
Convolution Layer
Pooling Layer
Flattening
Input to the to final layer
Structure of the network so far
ReLU

44. Fully Connected Layer
………… Flattened Matrix
The Flattened matrix from the pooling layer is fed as input to the Fully Connected Layer to classify the
image

45. Fully Connected Layer
………… Flattened Matrix
Dog
Bird
Cat
The Flattened matrix from the pooling layer is fed as input to the Fully Connected Layer to classify the
image
Pixels from the flattened
matrix fed as input

46. Fully Connected Layer
Dog
Bird
Cat
Identifies the image
The Flattened matrix from the pooling layer is fed as input to the Fully Connected Layer to classify the
image
Pixels from the flattened
matrix fed as input

47. Fully Connected Layer
1 1 1 0 0
0
0
0
0
1
0
0
1
1 1 0
1 1 1
1 1 0
1 0 0
Convolution Pooling
Input Image
Convolution Layer
Pooling Layer
Flattening
ReLU
Fully Connected
Layer

48. Fully Connected Layer
Lets see the entire process how CNN recognizes a bird
Dog
Bird
Cat
Feature Extraction in multiple hidden layers Classification in the output layer
Convolution + ReLU + Max Pooling Fully Connected Layer

49. Use case implementation using CNN
We will be using CIFAR-10 data set (from Canadian Institute For Advanced
Research) for classifying images across 10 categories
01
03
05
07
09
02
04
06
08
10
airplane automobile
bird cat
deer dog
frog horse
ship truck

50. Use case implementation using CNN
1. Download data set
2. Import the CIFAR data set

51. Use case implementation using CNN
3. Reading the label names

52. Use case implementation using CNN
4. Display images using matplotlib

53. Use case implementation using CNN
4. Display images using matplotlib

54. Use case implementation using CNN
4. Display images using matplotlib

55. Use case implementation using CNN
5. Helper function to handle data

56. Use case implementation using CNN
5. Helper function to handle data

57. Use case implementation using CNN
6. To use the previous code, run the following
7. Creating the model

58. Use case implementation using CNN
8. Applying the helper functions

59. Use case implementation using CNN
8. Create the layers
9. Create the flattened layer by reshaping the pooling layer
10. Create the fully connected layer

60. Use case implementation using CNN
12. Apply the Loss function
11. Set output to y_pred
13. Create the optimizer
14. Create a variable to initialize all the global tf variables

61. Use case implementation using CNN
15. Run the model by creating a Graph Session

62. Key Takeaways