Automatic discard registration in cluttered environments using deep learning and object tracking: class imbalance, occlusion, and a comparison to human review

Automatic discard registration in cluttered environments using deep learning and object tracking: class imbalance, occlusion, and a comparison to human review\ Rick van Essen, Angelo Mencarelli, Aloysius van Helmond, Linh Nguyen, Jurgen Batsleer, Jan-Jaap Poos and Gert Kootstra Paper: https://doi.org/10.1093/icesjms/fsab233

About

This repository contains the code beloning to the paper "Automatic discard registration in cluttered environments using deep learning and object tracking: class imbalance, occlusion, and a comparison to human review".

Installation

Python 3.8 is needed with all dependencies listed in requirements.txt. Optionally, apex can be installed for faster training:

commandline pip install -r requirements.txt pip install detection/apex

Content

The software contains 5 notebooks:

| Notebook | | Description | |----------------------------------------|-------------------------|--------------------------------------------------------------------------------------| | createsyntheticdata | | Notebook to create synthetic data. | | train | | Notebook to train the YOLOv3 neural network. | | detect | | Notebook to detect fish in the images. | | track | | Notebook to track the fish over consequtive images. | | evaluate | | Notebook to evaluate the detection and count the number of tracked fish. |

Citation

If you find this code usefull, please consider citing our paper:

text @article{vanEssen2021, author = {vanEssen, Rick and Mencarelli, Angelo and vanHelmond, Aloysius and Nguyen, Linh and Batsleer, Jurgen and Poos, Jan-Jaap and Kootstra, Gert}, title = {Automatic discard registration in cluttered environments using deep learning and object tracking: class imbalance, occlusion, and a comparison to human review}, journal = {ICES Journal of Marine Science}, volume = {78}, number = {10}, pages = {3834-3846}, year = {2021}, month = {11}, issn = {1054-3139}, doi = {10.1093/icesjms/fsab233} }

The dataset belonging to this repository can be found at https://doi.org/10.4121/16622566.v1. A small sample dataset is available for quickly testing this repository.

Funding

The study was carried out under the Fully Documented Fisheries project initiated by the Dutch Ministry of Agriculture, Nature and Food Quality and funded by the European Maritime and Fisheries Fund.