Image recognition

EIKON

Team: Portia Cooper, Emily Wang, Unity Dill, Anu Nair, Kemka Ihemelandu, & Shubham Bhatnagar

Faculty: Ryen Krusinga, Kamal Gupta, & Dr. David Jacobs

AI4All Facilitators: Andrew Lambeth & Janielle Jackson

Project Overview

Humans classify and identify objects based on visual cues. Computers are able to replicate human perception by utilizing neural networks; this is referred to as image recognition. For our project, we developed a convolutional neural network that classifies images of leaves and identifies tree species.

What is Image recognition?

Image recognition refers to computer technologies that utilize algorithms and machine learning to identify objects.

How is this possible?

Artificial intelligence is basically giving a human brain to a computer. Just as human brains have neurons, machines utilizing neural networks make similar connections. The neural network will work with the program as neurons work with human brains.

Project Question

Can students develop a convolutional neural network that approximates the performance of the pre-existing resnet18 neural network?

Action Steps

Team divided into two work groups:

Group 1 created the neural network.
Group 2 wrote code for the training algorithm.

The neural network was trained alongside a pre-existing neural network (resnet18) with the leafsnap dataset using 50 epochs.

Team implemented neural network on web app.

Deliverable

Convolutional neural network functioning at 55% accuracy developed by student team.

Training Loss and Validation Accuracy by Epoch Results for Team Eikon's Neural Network and Resnet18

Web app

Link to web app: http://a9ce0c9fe78a.ngrok.io

Here is Team Eikon's presentation. Check it out for more information!

Eikon: Image Recognition Presentation

Page updated

Report abuse