Where are poachers? Real-time drone images processing.

Inspiration

Animal trafficking is one of the worst sides of mankind. It ruins animal's lives, destroys populations and ecosystems. We are probably the last generation with abilities to eradicate this problem.

What it does

Our solution is a neural network that detects poachers and animals on infrared drone photo. It can be deployed on a drone using Intel Neural Compute Stick 2. This drone can make real-time decisions and send alerts to authorities.

How I built it

We conducted a study to answer the question: “what machine learning algorithm do researchers use in similar cases? ”. We found out that deciase segmentation tasks as example lung cancer segmentation and brain tumor segmentation tasks looks very similar to our task. Datasets with x-rays of lungs with tumors looks alike infrared desert photos. Objects to detect looks very similar too. This means that we can find solution with the best performance in deciase segmentation cases and adapt solution’s neural network. We found out that Mask-RCNN usually has the best overall performance in similar cases, and we used this recursive neural network in our solution. Also Mask-RCNN can be launched on drone. We trained Mask-RCNN on given dataset, our model has way better performance in some metrics than existing solutions. There is more to come.

Challenges I ran into

The most challenging for us were to train model on this large dataset in a such a limited time. For achieving better performance we need more than a week.

Accomplishments that I'm proud of

1)High scalability of this solution 2) Importance for future wildlife trafficking prevention 3) Box software solution which could be used in national parks worldwide shortly after hackathon.