Adventures in Machine Learning

Unveiling the Power of Convolutional Neural Networks for Image Classification

Machine learning has been a buzzword for quite some time now. One of the aspects of machine learning that has gained a lot of attention is image recognition.

The MNIST dataset is one of the most well-known datasets that are used as a benchmark for image recognition algorithms. In this article, we will learn about the MNIST dataset, its popularity, loading the dataset in Python, verifying the shape of the training and testing data, displaying images using Matplotlib, and plotting the dataset.

1) Loading the MNIST Dataset in Python:

The MNIST dataset consists of handwritten digits from 0 to 9. It is a voluminous dataset that is publicly available.

The dataset is popular for many reasons, including its sheer size and the fact that it requires little to no processing when compared to other datasets. Loading the MNIST dataset in Python is a simple task.

One of the easiest ways to load the dataset is by using Keras. Keras is a powerful and easy-to-use Python library for developing deep learning models.

The MNIST dataset is built into Keras, making it easy to access. Once you have Keras installed, you can access the dataset by using the following code:

“` python

from keras.datasets import mnist

(train_images, train_labels), (test_images, test_labels) = mnist.load_data()


This code will load the dataset into four variables: train_images, train_labels, test_images, and test_labels.

2) Verifying the Shape of Training and Testing Data:

Before we proceed, it is essential to verify the shape of the training and testing data. This step is crucial to ensure that we know how many images we have and how many of them belong to the training and testing sets.

To do this, we can use the following code:

“` python

print(“Train data shape: “, train_images.shape, train_labels.shape)

print(“Test data shape: “, test_images.shape, test_labels.shape)


When we run this code, we will get the shape of the training and testing data. The output will look something like this:


Train data shape: (60000, 28, 28) (60000,)

Test data shape: (10000, 28, 28) (10000,)


This output tells us that we have 60,000 training images and 10,000 testing images.

Each image is 28 x 28 pixels, and the training labels and testing labels are single-dimensional arrays. 3) Displaying Images in the MNIST Dataset using Matplotlib:

Matplotlib is an excellent Python library for data visualization.

It can be used to display images from the MNIST dataset. To display an image from the dataset, we can use the following code:

“` python

import matplotlib.pyplot as plt

# Select an image from the train dataset

image_idx = 1

plt.imshow(train_images[image_idx], cmap=’gray’)

plt.title(‘Label of the image: ‘ + str(train_labels[image_idx]))


This code will display the selected image from the MNIST training dataset.

We can also display multiple images side-by-side using Matplotlib’s subplot function. The following code displays multiple images:

“` python

import matplotlib.pyplot as plt

# Select 10 images from the train dataset

images = train_images[0:10]

labels = train_labels[0:10]

# Plot the images

fig, axes = plt.subplots(2, 5, figsize=(15, 6))

for i, ax in enumerate(axes.flat):

ax.imshow(images[i], cmap=’gray’)

ax.set_title(f”Label: {labels[i]}”)


This code will display ten images side-by-side.

We can see that each image has a corresponding label. 4) Plotting the MNIST Dataset:

Plotting the MNIST dataset is an excellent way to understand its distribution.

We can plot the MNIST dataset using the following code:

“` python

import seaborn as sns

import matplotlib.pyplot as plt

# Flatten the dataset

train_images_flattened = train_images.reshape(60000, 784)

# Plot the first 25 images

plt.figure(figsize=(10, 10))

for i in range(25):

plt.subplot(5, 5, i+1)



plt.imshow(train_images[i], cmap=’gray_r’)


# Plot the distribution of the dataset


plt.title(“Distribution of images in the MNIST dataset”)


The first plot shows the first 25 images of the training dataset. We can observe that some numbers are written differently, and some are written very similarly.

The second plot shows the distribution of the dataset. We can see that the dataset is well distributed, with each number having an almost equal number of samples.


In conclusion, the MNIST dataset is a popular dataset used for benchmarking image recognition algorithms. Loading the dataset, verifying the shape of the training and testing data, displaying images using Matplotlib, and plotting the dataset is a simple task.

Matplotlib is a great library for data visualization and can be used to display images from the MNIST dataset. Plotting the MNIST dataset gives us an insight into its distribution.

Understanding the MNIST dataset is a great way to start working with machine learning algorithms.Convolutional Neural Networks (CNNs) are a type of deep neural network that is commonly used for image classification. They are inspired by the structure and function of the visual cortex in animals and have been shown to be very effective in recognizing images.

CNNs use a combination of convolution and pooling layers to filter, reduce the size and represent pixels of an image. This article will provide an overview of CNN networks for image classification and explain in detail the convolution, pooling, and flattening layers.

1) Overview of CNN Networks for Image Classification:

The input to a CNN is a raw image, and the output is a classification label. CNNs are made up of multiple convolutional and pooling layers, followed by one or more fully connected layers.

Convolutional layers perform the filtering and feature extraction, while pooling layers reduce the spatial size of the output feature maps. Fully connected layers are used to map the output of the final pooling layer to the target label.

CNNs use backpropagation to learn the weights of the filters in the convolutional layers and the weights of the fully connected layers. 2) Convolution Layer for Filtering the Image:

Convolution layers in a CNN perform the filtering and feature extraction of the input image.

The convolution operation involves sliding a window over an image and taking the dot product of the window and the underlying input pixels. This window is commonly referred to as a filter or kernel.

By sliding the window across an image, we can extract specific features of interest in the image. For example, if we are looking for edges in an image, we can use a Sobel filter.

A CNN has multiple convolution layers, with each subsequent layer learning more complex features of the image. The size of the filters used in the convolution layers is typically small, with 3×3 or 5×5 filters being common.

Convolution layers are usually followed by activation functions, such as the rectified linear unit (ReLU), to introduce non-linearity into the model. The output of a convolution layer is referred to as the feature map.

3) Pooling Layer for Reducing the Spatial Size of the Image:

Pooling layers in CNNs reduce the spatial size of the output feature maps. They are used to downsample the feature maps, reducing the spatial size while retaining the essential information.

There are two common types of pooling: max pooling and average pooling. Max pooling returns the maximum value in a pooling window, while average pooling returns the average value.

Max pooling is more commonly used in CNNs for image classification. By reducing the spatial size of the output feature maps, pooling layers help with regularization and prevent overfitting.

They also make it possible to use more filters in the convolution layers, making the network more complex and capable of learning more detailed features. Pooling layers are typically added after each convolution layer in a CNN.

4) Flattening Layer for Representing Multi-Dimensional Pixel Vector as a One-Dimensional Pixel Vector:

The output of a CNN is typically a multi-dimensional array of feature maps. To use this output in a fully connected layer, we need a one-dimensional pixel vector.

The flattening layer in a CNN performs this conversion. It takes the output of the final convolution or pooling layer and flattens it into a one-dimensional pixel vector.

A one-dimensional pixel vector is easier to process by a fully connected layer compared to a multi-dimensional array. The output of the flattening layer is fed into a fully connected layer, which maps the output to the target label.


Convolutional Neural Networks (CNNs) are a powerful and effective way to perform image classification. They use a combination of convolution and pooling layers to filter, reduce the size, and represent pixels of an image.

Convolution layers perform the feature extraction by filtering the image, while pooling layers downsample the feature maps. Flattening layers convert the output of the final pooling or convolution layer into a one-dimensional pixel vector, which is fed into a fully connected layer for classification.

Understanding these layers is essential in building a effective Convolutional Neural Network for image classification. In this article, we explored the essential components of Convolutional Neural Networks (CNNs) for image classification.

CNNs use convolution and pooling layers to filter, downsample, and represent the pixels of an image. The flattening layer is used to convert multi-dimensional feature maps into a one-dimensional pixel vector.

Understanding the functioning of these layers is crucial in building an effective CNN for image classification. By utilizing CNNs, we can solve problems of image recognition, drive autonomous vehicles, and develop innovative solutions in healthcare.

Popular Posts