Endangered Species Object-Detection Model

• 

  Prediction

• 

  Database

• 

  Interface

Inspiration

We were inspired to create a model to run inferences on endangered species because computer vision and object detection is something that interested us. In addition, we really like animals... especially elephants!

What it does

This solution is developed to track the populations of endangered species overtime in a given area using overhead imagery. Ideally, photo samples would be taken of areas of land, for example an animal reservation, using equipment such as satellites or drones. This is a less intrusive way to track changes in endangered species numbers in areas that are not as accessible to humans. In addition, we have prepared the data such that it is easy to track the population and change in numbers overtime as is seen in our user interface.

How we built it

We used Python Tkinter as the main interface to the model and the database. We trained a few object-detection models using TensorFlow in a Google Colab notebook. We had difficulties integrating it into the interface so we used a Roboflow API to connect our model. As for our SQL database, the database that was most compatible with our interface, SQLite.

Challenges we ran into

Because object-detection training is so complex, we decided it would be easier to focus on one endangered species. For the sake of the hackathon, we decided to focus on just elephants because we had access to high-quality data to train. As was mentioned earlier, it was difficult for us to integrate our trained models into an interface. We had models that were actually better that we could not use because we could not figure out a good way to combine it with Tkinter.

Accomplishments that we're proud of

Deep learning is hard but we learned a lot through it! In addition, we were able to tie the model in to an interface and back the data in an SQL database... so many moving parts.

What we learned

We learned a lot about deep learning and how powerful convolutional neural networks are. Additionally, with training for computer vision the quality of the dataset is everything! Your model is only as good as what you put in afterall.

What's next for Endangered Species Object-Detection Model

Ideally, we would continue to make the model's mAP (mean Average Precision) score a bit better. We hope to include not only elephants, but other endangered species as well. We believe our interface could be improved as it now looks like it came from a couple decades ago. We hope to integrate the model into a website or application to be more aesthetic.

Built With

• python
• roboflow
• sql