AImEe

• 

  Fire Map built with react

• 

• 

  First experiments with Inception Model with azure machine learning

• 

  Slow start. Our model incorrectly classifies most sunsets as fire

• 

  After training the model using more data, we are getting better results

• 

  Retraining the data with deformed images gives even better results

• 

  Testing on different data (clothing pattens) reveals our model identifies bright fabric patterns as fire

AImmEe is a loose acronym for AI Machine learning Model available Everywhere to Everyone

Inspiration

While learning theoretical concepts about image recognition, computer vision, linear algebra, and vector calculus, I was inspired by advancement in technology, and was eager to see how all this theory could be used and implemented in real life. How can the beauty and elegance of of mathematics are able to create useful results in real life and have a positive impact on human life? In this instance I was inspired to explore how AI could be used to detect fires.

What it does

AImmEe is an AI that can "live" inside and direct the actions of any autonomous system. In this version, AImmEe is directing drone's movements, and is able to process the data (images) the drone is collecting in real time. AImmEe classifies collected images as "fire" or "nofire" and sends the alerts if the fire was detected.

How I built it

1. After researching several machine learning models, I have chosen InceptionV3 model because of its high results in classifying images. I used an Azure Notebook VM in Azure machine learning services to run experiments, I learnt how to build Docker Images and deploy the model as a web service with ACI.

2. Although InceptionV3 was great for detecting many objects, it was not working well with classifying fire and smoke. I used transfer learning technique and retrained the model on different data using only two categories (fire and no fire). During the process I experimented with another model in Azure machine learning service, but it didn't provide the desired results.

3. Front-end is built with react. It shows the process of receiving and mapping the data in real time from a drone working in a remote location to detect fires. Orange indicates the fire was detected. This video corresponds to the scene 21: "Notre Dame in Flames" Although it doesn't display the images, you can start seeing the fire patterns, and send help to the areas where the most help is needed

Challenges I ran into

Steep learning curve of learning multiple things in a very short time. Deploying a web service on as ACI. The front end of the service was not working.

Accomplishments that I'm proud of

Learning basics of AZURE machine learning service, running my first experiment in this environment. Building my first application from start to end.

What I learned

Azure machine learning service ACI containers How to create DOCKER Image How to deploy models as a web service How to work with Jupyter notebooks using state in react

What's next for AImEe

To optimize the movement of the drone based on the info received from running inferences.

Built With

• azure
• azuremachinelearning
• cv2
• flask
• python