The State Key Laboratory for Marine Population in City University of Hong Kong is working on a special project to preserve Chinese white dolphins.They are applying image diagnostic techniques to find out the health condition of Dolphins. Data is collected with frequent boat trips along the coastal line of Hong Kong every month. Each Dolphin sighting trip contains thousands of images out of which only few are worthy of diagnostic purposes. The aim of my project is to automate the process of manual image processing
This project consists of implementation of various deep learning methods which are used for image diagnostic purposes.
- Classify images of dolphin based on grades
- important preprocessing step to remove redundant images
- Trained variety of models , including Resnet , Faster-RCNN , Scaled YOLOv4 model
- Faster-RCNN to crop images of dolphin to reduce dataset size. This mehtod also helps us deal with the issue with multiple dolphins in a picure.
- The goal of this program is to cluster images based on visual simmilarity.
- We have used t-SNE dimention reduction mehtods to reduce dataset size.
- To perform unsupervised clustering, you can use K-means or HDBSCAN to form clusters