I am a person with severe peanut allergies. I wanted to create something that could potentially help people like me. While some people might find identifying different nuts easy, people with visual disabilities or people who have been sheltered away from an allergen might have trouble with this. I wanted to use machine learning to create something that will help people identify different types of nuts.

What it does

Given an image of either an almond, a peanut, a walnut, a cashew, or a hazelnut, the application will identify which nut it is.

How I built it

I used the Wolfram language to tackle this problem. First I created sets of examples for each type of nut. To accomplish this, I used a web scraping technique to gather many images (about 60) for each type of nut. Next, I created a function called NutClassify that can classify the type of nut.

Challenges I ran into

Machine learning can use up a lot of memory especially when it comes to images, so my computer kept crashing. I was able to overcome it by confining the images to a fixed specification.

Accomplishments that I'm proud of

Despite numerous challenges, including the computer crashing multiple times, I was able to successfully create a microsite for the application.

What I learned

I made the application using machine learning, but in order for it to be very effective and accurate, I need large dataset. While the classifier was trained using numerous pictures for each nut (about 60), including more pictures will certainly help.

What's next for Nut Identification Using Machine Learning

I would like to improve the efficiency of this application by training it with a bigger dataset.

Subsequently, I would like to make it a self-learning application. Currently, I was able to get the dataset set by performing a google search and uploading the images on to Wolfram cloud. I would like to make it a self-learning solution.

It could also be improved by creating an app that does the same thing as the microsite. While the microsite can be used from a phone or a tablet, it may be easier to use an app.

Share this project: