Identification of individual animals using computer vision methods has been demonstrated for many species with unique external markings. In this study, tropical rock lobster (TRL) species Panulirus ornatus are shown to be uniquely identifiable from the analysis of fingerprint-like patterns between their two supra-orbital horns.
The use of an identification algorithm based on the feature detection and description method ORB delivered a superior performance with relation to speed and accuracy compared to other state of the art methods: SIFT, SURF and BRISK. Contrast limited adaptive histogram equalisation (CLAHE) and random sample co-nsensus (RANSAC) methods were found to reduce the false match rate of the algorithm. On a sample population of 15 adult tropical rock lobster, a false match rate of 0.0% was achieved for images where the unique patterns of each individual were clearly shown. A population of 1500 lobsters was simulated using computer generated patterns.
Non-ideal image effects such as cutout, blur, and scale were found to produce a false match rate of 2.9%, compared to 25.0% for ORB without the proposed improvements.
The versatility of the algorithm was tested by confirming its ability to identify animals both after moulting and cooking (boiling). While market sized tropical rock lobster were the focus of this work. Testing of the algorithm on juveniles found that the development of their unique patterns over time increased the false match rate. The proposed identification algorithm has been deployed in a prototype mobile phone application, developed for iOS in the native Swift language. The application demonstrates the feasibility of a consumer-side autonomous, real-time identification tool, for verifying the authenticity of tropical rock lobster products. This developed technology also has potential uses for commercial aquaculture stock management, citizen science, and adaptation for other species.