Skip to main content
Publications | Persons | Institutes | Projects
[ report an error in this record ]basket (0): add | show Print this page

Digital image analysis of flatfish bleeding injury
Uhlmann, S.S.; Verstockt, S.; Ampe, B. (2020). Digital image analysis of flatfish bleeding injury. Fish. Res. 224: 105470. https://dx.doi.org/10.1016/j.fishres.2019.105470
In: Fisheries Research. Elsevier: Amsterdam. ISSN 0165-7836; e-ISSN 1872-6763, more
Peer reviewed article  

Available in  Authors 

Keyword
    Marine/Coastal
Author keywords
    Algorithm; Automated image analysis; Haemorrhage; Observer bias

Authors  Top 

Abstract
    To improve accuracy of post-release mortality predictions and facilitate the routine collection of information about physical condition of catches after commercial fishing capture, traditional visual assessment by potentially subjective human observers or raters may be automated by digital image analysis. The purpose of this study was to develop a method and device that can eliminate subjectivity in scoring external injury of commercially beam-trawled flatfish by taking standardized, high resolution images to allow for automated calculation of the % surface area of visible bleeding injury relative to the whole fish based on digital image analysis. A reference library was compiled by photographing ventral sides of 67 fish of six flatfish species of different sizes and freshness (fresh vs defrosted). All fish were sourced from the R/V Simon Stevin while beam-trawling in the Belgian coastal zone of the Southern North Sea. All images that were neither over- nor under-exposed were compiled (n = 51) and scored for the extent (%) of multifocal cutaneous petechial ('point bleeding'), and suffusion or haemorrhaging ('bruising') of the ventral head and body region, respectively, by three experienced raters using a continuous scale (between 0 and 100 %). Then, several state-of-the-art computer vision algorithms were tested on the dataset to develop a protocol that can 1) align each image; 2) identify fin, body and head regions; and 3) quantify the surface area of bleeding injury of each region by using appropriate thresholding techniques. For validation of the computer-derived % surface coverage estimates of bleeding injury, these were compared to the average rater’s score. For bruising injury, a significant difference between human- vs computer-derived scores persisted. For point bleeding of the head region, computer-based estimates of % coverage were not different from those of the human raters. Overall, species, size and their freshness did not have a significant effect. By consistently recording the coverage of externally visible bleeding injury, this image analysis protocol may find its application in measuring the effect of different capture techniques on whole fish quality, and in improving vitality assessments as part of the transition towards a more sustainable fishery and the implementation of the European Landing Obligation.

All data in the Integrated Marine Information System (IMIS) is subject to the VLIZ privacy policy Top | Authors