Identification of Shark Species Based on Their Dry Dorsal Fins through Image Processing

Abstract

Shark populations worldwide have suffered a decline that has been primarily driven by overexploitation to meet the demand for meat, fins, and other products for human consumption. International agreements, such as CITES, are fundamental to regulating the international trade of shark specimens and/or products to ensure their survival. The present study suggests algorithms to identify the dry fins of 37 shark species participating in the shark fin trade from 14 countries, demonstrating high sensitivity and specificity of image processing. The first methodology used a non-linear composite filter using Fourier transform for each species, and we obtained 100% sensitivity and specificity. The second methodology was a neural network that achieved an efficiency of 90%. The neural network proved to be the most robust methodology because it supported lower-quality images (e.g., noise in the background); it can recognize shark fin images independent of rotation and scale, taking processing times in the order of a few seconds to identify an image from the dry shark fins. Thus, the implementation of this approach can support governments in complying with CITES regulations and in preventing illegal international trade.