The Ultimate Guide to Building Powerful Keras Image Classification Models

Image classification is one of the most important applications of deep learning and Artificial Intelligence. Image classification refers to assigning labels to images based on certain characteristics or features present in them. The algorithm identifies these features and uses them to differentiate between different images and assign labels to them. In this tutorial titled ‚ÄėThe ultimate guide to building powerful Keras Image Classification Models‚Äô, you will learn how to perform image classification with Keras, a deep learning library.¬†

Post Graduate Program in AI and Machine Learning

In Partnership with Purdue UniversityExplore Course
Post Graduate Program in AI and Machine Learning

What Is Image Classification?

Image classification is the process of assigning classes to images. This is done by finding similar features in images belonging to different classes and using them to identify and label images. 

Image classification is done with the help of neural networks. Neural networks are deep learning algorithms. The below picture shows a neural network.

ImageClassification_Models_1

Figure 1: Neural Network

Neural Networks contain multiple layers of neurons that perform prediction, classification, etc. The output of each neuron is fed to the neurons in the next layer which helps fine-tune the output until we reach the final output layer. The different layers which are present in a neural network are :

  • Input Layer: This is the layer through which we give the input to your neural network
  • Hidden Layer: This layer contains various neurons which process the input received from the input layer
  • Output Layer: The final layer in the network which processes your data one last time and gives you the output

Neural networks can be easily implemented using a deep learning library like Keras, Tensorflow, or Pytorch. You will be using Keras to implement your Neural Networks. Keras is a Python library that supports other deep learning libraries as backends while providing a user-friendly frontend. It offers strong computational power while maintaining ease of implementation. 

FREE Machine Learning Certification Course

To become a Machine Learning EngineerExplore Course
FREE Machine Learning Certification Course

Intel Image Classification Dataset

The dataset that you will be using is the Intel Image Classification dataset which contains images of different landforms such as forests, glaciers, mountains, sea, buildings, and streets. It is a well-collected dataset with images meticulously collected and stored in different folders. 

ImageClassification_Models_3.

Figure 2: Intel Image Classification Dataset

Loading the Data

You will begin by loading your data and importing the necessary modules.

ImageClassification_Models_4

Figure 3: Importing modules

You will also declare variables that contain the class names and the corresponding label and define the desired IMAGE_SIZE for our pictures. You must define the labels corresponding to the various classes in the dataset.

ImageClassification_Models_5.

Figure 4: Defining class labels

Now, load our data. You will define a function called load_data which you can use to load your train and test data. 

  • You have two folders within the data folder, the seg_train folder, and seg_test folder.
  • Within each of these folders, you also have a folder containing images from each class.¬†
  • You will read individual images from each folder and push them into your image array after converting them to RPG form and resizing.
  • You will append a label to this image and append it to the output array.

ImageClassification_Models_6

ImageClassification_Models_6_1

Figure 5: Loading our data 

You will then call your load_data() function and save your training and testing data. To better train the model, you will shuffle the data in your train dataset. 

ImageClassification_Models_7.

Figure 6: Creating your training and testing dataset

Free Course: Introduction to Neural Network

Learn the Fundamentals of Neural NetworkEnroll Now
Free Course: Introduction to Neural Network

How to Create a Convolution Neural Network?

Creating a Convolution Neural Network with Keras is relatively easy. You can define which model you want. In this case, you will be using a sequential model. You then define the different layers.

  • You are using one hidden layer for our model.
  • You must define your input layer as a convolution layer followed by a MaxPooling layer.
  • You have a hidden layer of another convolution layer and a hidden layer.
  • You flatten your outputs to reduce the number of features and you have an output layer consisting of a dense relu layer and a dense softmax layer.

ImageClassification_Models_8

Figure 7: Creating a CNN

Following this, you have to compile our model. You need to use an Adam optimizer to optimize your model and a loss function to calculate the loss. The metrics define which metric you want to calculate. In this case, it is accuracy.

ImageClassification_Models_9.

Figure 8: Compiling the model

After compiling, fit the model to your training data, ie: train the model. You will train it in batch sizes of 128 with 6 epochs and use 20% of the data as the validation set.

ImageClassification_Models_10

Figure 9 : Training our model

Artificial Intelligence Engineer

Your Gateway to Becoming a Successful AI ExpertView Course
Artificial Intelligence Engineer

Plotting Accuracy

To plot the accuracy of the model, define a function plot_accuracy_loss(). The data returned after your neural network also includes the final accuracy and loss of the model. You plot the accuracy of the training set and validation set for each epoch to understand the variation in your accuracy. Along with this, you also plot the loss and validation loss. 

ImageClassification_Models_11

Figure 10: Plotting accuracy of the model

Now call your plot function with the results of your training.

ImageClassification_Models_12

Figure 11: Accuracy of your model

Now, evaluate the loss and accuracy of your model on the test data. From the graphs, you can see that the accuracy of the model increases with every epoch for both training and testing sets. The loss of the model decreases with every epoch as your model learns and gets better.

 Along with this, you must also save the model predictions and use them to make a classification report of different metrics such as precision, recall, etc to get a clear view of how well the model is performing.

ImageClassification_Models_13

Figure 12: Creating a classification report 

How to Create a VGG16 Model?

VGG16 is a pre-trained CNN model which is used for image classification. It is trained on a large and varied dataset and fine-tuned to fit image classification datasets with ease.

Now, import a VGG16 model. You must initialize the model and add input and output layers.

ImageClassification_Models_14.

Figure 13: Importing VGG16

You must now extract the features from the dataset and save them as train and test features.

ImageClassification_Models_15

Figure 14: Feature Extraction

Now, create the final model by specifying the inputs and outputs. You use convolution and MaxPooling layers as input layers and then flatten and use Dense layers for the output.

ImageClassification_Models_16.

Figure 15: Creating the model

Free Course: Introduction to AI

Learn the Core AI Concepts and Key Skills for FREEStart Learning
Free Course: Introduction to AI

You then need to compile your model using the adam optimizer and use the accuracy metric. 

ImageClassification_Models_17

Figure 16: Compiling our model

You then train the model on your test data.

ImageClassification_Models_18.

Figure 17: Training the VGG16 Model

You can plot the model error by calling the plot_accuracy_loss() function.

ImageClassification_Models_19.

Figure 18: Plotting accuracy of VGG16

Are you an AI and Machine Learning enthusiast? If yes, the AI and Machine Learning course is a perfect fit for your career growth.

Conclusion 

In this tutorial titled ‚ÄėThe ultimate guide to building powerful Keras Image Classification Models‚Äô, you explored image classification and understood the basic principle behind neural networks.¬† You then looked into your dataset and the Intel Image Classification Dataset. Moving on, you learned how to load data for the program and implement image classification using Keras with CNN. You calculated and plotted the accuracy of your model and finally took a look at the VGG16 architecture.

If you wish to learn more about Image Classification and Deep Learning check out the Post Graduate Program in AI and Machine Learning by Simplilearn in Collaboration with Perdue University and IBM. You will learn many crucial topics like NLP, Keras, Tensorflow, and much more.

Hope this tutorial taught you the basics of image classification and how to perform it using Keras. Do you have any doubts or questions for us? Mention them in this tutorial’s comments section, and we'll have our experts answer them for you at the earliest!

About the Author

Kartik MenonKartik Menon

Kartik is an experienced content strategist and an accomplished technology marketing specialist passionate about designing engaging user experiences with integrated marketing and communication solutions.

View More
  • Disclaimer
  • PMP, PMI, PMBOK, CAPM, PgMP, PfMP, ACP, PBA, RMP, SP, and OPM3 are registered marks of the Project Management Institute, Inc.