#

Building an AI Model to Identify Microorganisms Using Azure Custom Vision

#Martin Muchuki Sep 9th 2024
Read Aloud 405 Views

Imagine you're a microbiologist in a lab, tasked with quickly identifying various microorganisms under a microscope. Each slide presents a different challenge: Is it an amoeba? A paramecium? Maybe a type of bacteria, but which kind? Identifying these tiny organisms can be time-consuming and prone to human error, especially when the distinctions are subtle. What if you had an AI-powered assistant capable of identifying microorganisms instantly and with precision?

This isn’t science fiction—it’s possible today using Azure Custom Vision! In this blog, we’ll show you how to build an image classification model specifically designed to recognize different types of microorganisms, including:

- Amoeba

- Euglena

- Hydra

- Paramecium

- Spherical Bacteria

- Rod Bacteria

- Spiral Bacteria

- Yeast

Step 1: Set Up Your Azure Custom Vision Resource

Before we begin building the model, we need to set up the infrastructure that will power it.

1. Go to the Azure Portal and search for Custom Vision.

2. Create a new Custom Vision resource, selecting the Training and Prediction resource types.

3. Provide basic details like your subscription, resource group, region, and pricing tier. For this project, the free tier will work just fine.

A screenshot of a computer

Description automatically generated

Once you’ve set up the resource, head over to the Custom Vision portal (https://www.customvision.ai/ ) to start creating your microorganism classifier.

Step 2: Create a New Custom Vision Project

Now that your resources are in place, it's time to create your first project for microorganism identification.

1. In the Custom Vision portal, click on New Project.

2. Give your project a name, such as "Microorganism Classifier."

3. Select Classification as the project type, since we are identifying the types of microorganisms.

4. Choose whether you want Single-label (one tag per image) or Multi-label (multiple tags per image). For this project, Single-label works best.

5. In the Domains section, select General to ensure flexibility in recognizing diverse microorganism shapes and textures.

Step 3: Upload and Tag Images

Now, it’s time to teach your AI what each microorganism looks like by uploading images.

1. Gather a collection of labeled images for each microorganism. Ideally, aim for at least 50–100 images for each organism type to ensure accurate training. You can also take advantage of existing datasets eg from https://www.kaggle.com/

   - Amoeba: Typically blob-like with pseudopodia.

   - Euglena: An elongated, oval shape with a flagellum for movement.

   - Hydra: A small, tubular organism with tentacles.

   - Paramecium: A slipper-shaped organism with cilia.

   - Spherical Bacteria: Round-shaped bacteria, often in clusters.

   - Rod Bacteria: Cylindrical-shaped bacteria.

   - Spiral Bacteria: Corkscrew-shaped, highly distinctive.

   - Yeast: Spherical to oval-shaped fungi, often seen in budding forms.

2. In the Custom Vision Portal, click Add Images and upload your dataset.

3. Assign the appropriate tags to each image, like “Amoeba” or “Rod Bacteria.”

4. Once all your images are uploaded and tagged, you’ll have a strong foundation to train the model.

Step 4: Train the Microorganism Classifier

Now that your images are uploaded and labeled, it’s time to train the model.

1. In the project dashboard, click the Train button at the top right.

2. You can choose between Quick Training (good for small datasets) or Advanced Training (offers more customization options). For this project, let’s go with Quick Training.