The project is aimed to be implemented at an Autonomous Underwater Vehicle at project AUV manipal. We have used transfer learning from mobile net V2 to pretrain the model, therefore got a higher prediction accuracy even without much image preprocessing
Images scraped from website using a chrome plugin
Dataset link on drive: https://drive.google.com/drive/folders/1Ab6JDBHJd9dpD3WISHS-LTRL_xKncdR5?usp=sharing
Underwater images are in 8 different classes
- 'auv': 0
- 'fish': 1
- 'jellyfish': 2
- 'scubadiver': 3
- 'shark': 4
- 'starfish': 5
- 'turtle': 6
- 'wreckage': 7
Result:- Prediction accuracy: 94.48%