Wildify - Animal Recognition

Inspiration

As we know there are thousands of national parks and wildlife sanctuaries around the world spread in large areas. So it’s not a feasible task to keep a reliable record of the number of different animals habituated there. As of current scenarios, authorities keep track of the wildlife by monitoring them through the CCTV cameras installed in their natural habitats. However, it’s a strenuous, expensive, time-consuming and monotonous task for the surveillance personnels to manually monitor them. Here comes the use of our AI based application, which makes it easy to monitor the wildlife by recognizing their image.

What it does

Wildify is a tool that takes an image as input and returns the name of the animal present in the image as output. After the tool is created, it can be deployed to the web for everyone to use.

How we built it

By leveraging the power of CNNs, we aim to create a system that can automatically identify animals in a given image. We used the Animals 10 dataset obtained from Kaggle that contains about 26 thousand images of 10 distinct animals. By using the TensorFlow package, we created and trained a CNN and saved the model. For the purpose of reducing training time, we trained the system to recognize only 10 distinct animals i.e. dog, cat, sheep, horse, elephant, butterfly, spider, chicken, squirrel and cow. The AI was trained using a Convolutional Neural Network (CNN) and achieved an accuracy of 71.65%. A website was made to host the AI model and was deployed on the localhost server.

Challenges we ran into

1) We were limited to small datasets and large datasets required high memory to execute. 2) We were troubled with some bugs and errors while developing Wildify and it took us a lot of time and efforts to debug them.

Accomplishments that we're proud of

1) We found a suitable dataset using which we can train the AI. 2) Created a neural network suitable for our dataset. 3) We successfully trained the neural network until a high enough accuracy is reached and save the resultant model. 4) We implemented the model by creating a website with a usable User Interface.

What we learned

While developing Wildify we learned about all the technicalities involved. We learned about the AI, created neural networks and learned to train them. We also tried our hands to build a user-friendly frontend interface.

What's next for Wildify - Animal Recognition

We plan to develop Wildify even further as we want to realize it for the real-world usage. At present this project can recognize just a few number of animals. But we plan to make this project inclusive of almost all the species. This is possible by training our AI models on some broader datasets. In order to accomplish this we intend to seek the help of the official wildlife databases provided by the government and the concerned authorities.

Built With

• flask
• kaggle
• pillow
• python
• spyder
• tensorflow
• visual-studio