'Is the use of deep learning an appropriate means to locate debris in the ocean without harming aquatic wildlife?' Source Code
Description
These are the source code files for a simple convolutional neural network compared with the VGG-16 transfer learning model that we trained and tested for our paper: 'Is the use of deep learning an appropriate means to locate debris in the ocean without harming aquatic wildlife?'. Within our research we compared the following two CNN's by training them on our dataset of 1,644 images and testing on 100 images. As our database of imagery is small, we were only able to produce a 'prototype' on what this research could lead to. Our aim was to see if a CNN could safely distinguish between marine life and synthetic debris in a binary classification and we achieved promising results with an accuracy of 89% (custom CNN) and 95% (VGG-16). The full paper explains how with further development this project could be applied to automation and be a part of the process of cleaning up earth's oceans and waterways; the paper also outlines the importance of reducing marine litter on animal life, the environment and human health.
Files
Steps to reproduce
The source code files are kept at the same parameter settings we used for our database. However you should adapt these for your own data. Within the source code files, we have included comments on different aspects of the code as well as instructions on where to insert your own file directories. The convolutional neural network we built was influenced by Harrison Kinsley's (harrison@pythonprogramming.net) work.