Tail pattern recognition

Migration studies have been a key component of IPHC research since its inception in the 1920s. A variety of tags have been since utilized for mark-recapture investigations including external tags (wire, dart), internal tags (passive integrated transponder tags), satellite tags, and internally embedded archival tags. Each of these methods is able to aid in our understanding of a component of migration or behavior of interest, but have challenges such as variable reporting rates and high deployment costs. In recent years, there have been investigations with other species regarding the use of natural tags, i.e. marks or patterns indigenous to specific species and unique to individuals, in migration studies. Pacific halibut have natural patterning on both sides of the body and in 2017, the IPHC conducted a pilot study to explore the extent of uniqueness that can be determined from images of an individual fish with the use of pattern recognition software.

For the pilot study, data were collected in IPHC Regulatory Area 2C off the coast of Southeast Alaska (near Sitka and Cape Ommaney) in June 2017. The tail of the fish was chosen as the area most likely to yield favorable results, and photographs of both the white (blind) side and pigmented side of the tail for each fish in the study were taken. Analysis of the tail images was conducted using the Intelligent Individual Identification System (I3S) Spot (large user input required) and I3S Pattern (less user input required) software. Images were queried for matches in the database and matching scores, based on the numerical geometry of key points in the patterns, and ranks were given. Images were compared to copies of themselves and to images from the same fish taken at port offload events. Other variables (i.e. sex, pattern type, fork length, and all combinations of aforementioned variables) were analyzed in regard to their use of increasing the probability of the correct image match being ranked in first place.

The I3S Pattern software yielded larger proportions of correctly matching images than the I3S Spot software when image quality was high for both images being compared. However, if image quality was poor, I3S Spot was the more capable pattern recognition software to correctly identify matched individuals. A total of 88% and 82% correctly matched identical images for both blind side and pigmented side analyses was found, respectively, along with high proportions (67%) of correctly matched initial and offload images when the variable “length” (and any such combination with length) was used in conjunction. The blind side was determined to be the better side of the halibut for unique tail pattern recognition.

The pilot study was considered a success in that Pacific halibut tails present unique enough patterns for computer-assisted recognition software to match two images of the same tail, and this matching ability will potentially allow natural tagging in future IPHC ecological studies. However, there were image quality issues at the point of the offload due to variations in lighting and sampling station set-ups which limited the use of the software. Additionally, the software used here required a fair amount of additional user input for each fish which is inconsequential in a small study, but much more of a challenge when dealing with a database containing thousands of images. 

In the coming year, the IPHC is planning to continue investigating methods for obtaining high-quality images and testing software that will require as little user input as possible. If implemented as a coastwide project in the future, this “natural” tagging method could provide a low-cost solution with high precision for coastwide migration studies.