Overview / Usage

The importance of coral reefs is already well known, it is one of the most diverse ecosystems on the planet, responsible for providing habitat and shelter for a large number of marine organisms, it also helps to recycle nutrients.
The method used to evaluate coral growth in the reef currently requires a specialist, a number of points are chosen, these points are randomly positioned in the photo of the coral wall, the expert classifies what is below each point, so it is estimated a proportion of each coral on the wall.
This project intends to transfer the expert's knowledge to deep convolutional neural networks, thus classifying the points automatically.

Methodology / Approach

The images are from the expert himself, segments of each coral were labeled. Using well known CNN architectures such as AlexNet, GoogleNet, VGG, to investigate and find the best architecture for the problem.

Technologies Used

Intel Distribution for Python, TensorFlow* Optimizations for the Intel Architecture