Coral Reef Detection and Analysis (Industrial Project)

• Tech Stack: Python, Tensorflow, Keras, Visual Transformers, YOLOV5
• Github URL: Project Link

This project focuses on automating manual coral counting. The initial step involved deploying a U-Net model to create accurate coral reef masks, yielding a mean IOU score of 0.72.

Engineered a Custom-tailored Visual Transformers model utilizing 8 transformer encoders, trained over 200 epochs, culminating in an exceptional 96% accuracy for coral reef identification and analysis.

Devised a coral detection system employing YOLOv5, yielding an 80% accuracy rate. Employed the watershed algorithm to compute coral areas, which proved accurate within approximately 20% deviation from the original sizes.