Computer Vision - CNN for Image Classification by Wale Ayandiran Presented at AI Wales meetup, Tramshed Tech, Cardiff. 08-06-2022

We might start by identifying certain pictures of animals ourselves. Shall we? How does computer identify images with their labels?

Can you tell these wild cats aparts?

Leopard Cheetah Jaguar Tiger Image credit: Google

Machine Learning

Lion Cat Wolf Dog Fox

Cat Lion Dog Arctic Fox Wolf Image credit: Google

How are we as humans able to differentiate things easily?

Recognizing Patterns How are we as humans able to differentiate things easily?

Credit: MIT Feature Detection Low level feature detection to high level feature detection

What do computers see?

Images are matrix of numbers

Images are matrix of numbers What computers see

Images are matrix of numbers What computers see Credit: https://openframeworks.cc/ofBook/chapters/image_processing_computer_vision.html

Credit: https://www.keithbcarter.com/think-differently/how-a-machine-sees Computer detects features of an image as matrix of numbers.

How do we make computers identify images? Neural Networks. Neural networks are computing systems with interconnected nodes that work much like neurons in the human brain. Using algorithms, they can recognize hidden patterns and correlations in raw data, cluster and classify it, and – over time – continuously learn and improve. Credit: https://www.sas.com/en_us/insights/analytics/neural-networks.html

A simple neural network includes an input layer, an output (or target) layer and, in between, a hidden layer. The layers are connected via nodes, and these connections form a “network” – the neural network – of interconnected nodes. How Neural Network Works Credit: https://www.sas.com/en_us/insights/analytics/neural-networks.html

● Artificial Neural Network (ANN) ● Convolutional Neural Network (CNN) ● Recurrent Neural Network (RNN) 3 Popular Neural Networks

Convolutional Neural Networks

Learning Visual Features ➔ Filter extraction Applying filters and generating smaller pieces of image from the original image. ➔ Features Mapping Using the filters extracted to create a feature map of the image.

Image credit: MIT

Tiger Tiger’s eyes ? = Y Tiger’s nose ? = Y Tiger’s mouth/whiskers ? = Y Tiger’s paws ? = Y Tiger’s tail ? = Y Tiger’s head ? = Y Tiger’s body ? = Y Is it a Tiger ? = Y Image credit: Google

Main parts of a CNN Three main parts to a CNN ➔ Convolution Application of filters to generate feature maps ➔ Non-linearity (ReLU) Replacing negative values with zero (0) often ReLU (Rectified Linear Unit) ➔ Pooling Reduces dimensions and distortions, making the model tolerant towards image variations.

Feature Extraction Classification Image credit: codebasics

Some limitations of CNN in Image classification ❏ Requires large datasets to train ❏ Can take a lot of time to train if there are several layers ❏ Can be computing power intensive.

A CNN model to identify traffic signs https:/ /bit.ly/traffic_cnn

Thank You! Find slide and code link below Demo link: https:/ /bit.ly/traffic_cnn Github https:/ /github.com/waleCloud/solid-fortnight

Wale Ayandiran. Software and AI Engineer. Find me at https://walecloud.me Twitter @walecloud