FishNet: Deep Neural Networks for Low-Cost Fish Stock Estimation
Authors:
Moseli Mots'oehli,
Anton Nikolaev,
Wawan B. IGede,
John Lynham,
Peter J. Mous,
Peter Sadowski
Abstract:
Fish stock assessment often involves manual fish counting by taxonomy specialists, which is both time-consuming and costly. We propose FishNet, an automated computer vision system for both taxonomic classification and fish size estimation from images captured with a low-cost digital camera. The system first performs object detection and segmentation using a Mask R-CNN to identify individual fish f…
▽ More
Fish stock assessment often involves manual fish counting by taxonomy specialists, which is both time-consuming and costly. We propose FishNet, an automated computer vision system for both taxonomic classification and fish size estimation from images captured with a low-cost digital camera. The system first performs object detection and segmentation using a Mask R-CNN to identify individual fish from images containing multiple fish, possibly consisting of different species. Then each fish species is classified and the length is predicted using separate machine learning models. To develop the model, we use a dataset of 300,000 hand-labeled images containing 1.2M fish of 163 different species and ranging in length from 10cm to 250cm, with additional annotations and quality control methods used to curate high-quality training data. On held-out test data sets, our system achieves a 92% intersection over union on the fish segmentation task, a 89% top-1 classification accuracy on single fish species classification, and a 2.3cm mean absolute error on the fish length estimation task.
△ Less
Submitted 27 June, 2024; v1 submitted 16 March, 2024;
originally announced March 2024.
