Blue Vision

## Inspiration

Marine ecosystems are critical for maintaining the planet's ecological balance, regulating the climate, providing food, and preserving biodiversity. But there are growing threats : invasive species, endangered species, and plastic pollution. Our team is composed of two students in artificial intelligence, thus we thought of creating a tool capable of identifying and monitoring these threats. Our third member worked to propose the best data visualisation to guide our preservation actions.

## What it does

Our project uses AI-powered image recognition (Yolov5 for bounding boxes and classification, helped by a Kernelized Correlation Filter to track the individuals) to detect invasive species, endangered species, and plastics in marine ecosystems. The system identifies specific areas where these threats are present, enabling targeted computer vision and preservation efforts, thus optimizing cleanup actions, and better resource management. 

## How we built it

We started by researching underwater videos and databases to train a detection model. Using YOLOv5, we fine-tuned the model on datasets from distinct regions:
- Pacific: Featuring white sharks, manta rays, and plastic waste.
- Caribbean: Featuring turtles, lionfish, and plastic waste.

We improved the system with object tracking using TrackerKCF, enhancing accuracy in dynamic environments. Kaggle and Google Colab provided the computational power for training and testing.

## Challenges we ran into

1. Data availability: Finding rich datasets and preprocessing them was time-consuming.
2. Model compatibility: Switching between YOLOv5 and YOLOv8 led to compatibility issues with dependencies.
3. Tracking limitations: Integrating DeepSORT faced compatibility challenges, leading us to adapt TrackerKCF.
4. Combining our own prefered languages

## Accomplishments that we're proud of

We successfully trained YOLOv5 to detect key marine threats and implemented object tracking to improve monitoring capabilities. Despite dataset constraints and technical challenges, our model achieved reliable results for identifying and tracking invasive species, endangered species, and plastic waste.

## What we learned

We learned a lot about image recognition, AI model fine-tuning, and the complexities of data preprocessing. We learned the challenges of compatibility issues and dataset limitations, with innovative solutions. 

## What's next for Blue Vision

We aim to:
- Expand our dataset to include more regions and species.
- Improve tracking algorithms.
- Develop a real-time monitoring system.
- Perhaps enventually collaborate with local communities, NGOs, and policymakers to implement conservation strategies and raise awareness about marine ecosystem threats.

Built With

• fastapi
• kaggle
• php
• python