.png){kind=icon}
.png){kind=icon}
Return
Advancing Automated Fish Size Estimation From Images: Applications, Challenges and a Case Study for Images Without a Specified Reference Object
.png){kind=icon}
PRE
AI
C
R
F
S
A
DOI：10.1111/faf.70097.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Automatic and accurate estimation of fish sizes from images is essential for many monitoring, fisheries management, stock assessment and conservation efforts. However, current methods often rely on physical reference objects or stereo-camera systems that are not always available. This paper explores the advancements, applications and challenges of automated fish body-size estimation from images, using artificial intelligence (AI) and machine learning (ML) methods. We first introduce key concepts in AI and ML for a non-specialised audience and review existing literature on models used for fish size estimation. We identify key barriers such as a lack of high quality and publicly available datasets, image variability, scattered efforts and the challenges of model generalisation across diverse species. Then we present a novel framework for size estimation from monocular (non-stereo) images without a specified reference object, using a dataset from an angling app. Our approach utilises an efficient, pretrained deep learning-based feature extraction tool integrated with an automated regression pipeline. Our findings demonstrate a promising pathway for size estimation in images without a reference object, with most estimated fish lengths within 10% of their true length. Future research and collaborative efforts should focus on diversifying and sharing training data and integrating metadata. To this end, we created a user-friendly online application, where the community can test the model performance and contribute photos. Finally, it is essential to rigorously test and refine the robustness of current models in real-world fisheries applications and to adopt standardised, comparable metrics for evaluating fish size estimation models across studies.
Keywords:
artificial intelligence
computer vision
fish size
fisheries management
machine learning
monitoring
recreational fisheries
Journal
IF:
6.1
Papers:
1.3K
Citations:
7.3K
Organization
Cited Papers