Morphological injuries are well-established Operational Welfare Indicators (OWIs) for farmed animals including fish. They are often scored manually by human observers and this process can be laborious and prone to subjectivity and error. In this study we evaluated the use of a hyperspectral imaging system to quantify the presence and severity of external haemorrhaging in Atlantic salmon focusing on dorsal fins as a proof of concept OWI.
Two inexperienced observers manually audited dorsal fin injuries on 234 post-smolt Atlantic salmon following a standardized protocol that scored fin erosion on a 0-3 scale and also classified the injury as active/healed. The same fish were then imaged with a hyperspectral camera system and the manually scored visual assessments were compared with hyperspectral images of the same fin. Hyperspectral images were processed to segment out the dorsal fin of each fish and the presence of blood in the tissue was quantified by analysing the spectral information, yielding a fin haemorrhaging index.
The hyperspectral imaging platform was robust at detecting blood in fins and could help classify active injuries more accurately than human observers. The agreement between human scorers and the image analysis tool for classifying active bleeding vs healed/undamaged fins was good with a Cohen’s kappa of 0.81 and 0.90. Accuracy between the fin haemorrhaging index and the human observers was moderate (0.61 and 0.57) and on par with the agreement between the two human observers (0.68), demonstrating the difficulty in classifying injuries that result in a reduction in fin size but may or may not result in fin haemorrhaging.
These results demonstrate the potential power of hyperspectral imaging to improve welfare audits in aquaculture, especially where manual injury classification schemes have potentially mixed traits that involve haemorrhaging. The data also suggests that the hyperspectral camera can detect bleeding that is not readily visible to the human eye. There is a need for further testing and validation to integrate these tools into existing welfare auditing programs, but the potential advantages of the automated approach include increased sensitivity, accuracy and throughput, while producing quantitative data for researchers or management.