Fish Inspection System Using a Parallel Neural Network Chip and Image Knowledge Builder Application

Abstract

A generic image learning system, CogniSight, is being used for the inspection of fishes before filleting offshore. More than thirty systems have been deployed on seven fishing vessels in Norway and Iceland over the past three years. Each CogniSight uses four neural network chips (a total of 312 neurons) based on a natively parallel hardwired architecture performing real time learning and non-linear classification (RBF). These systems are trained by the ship crew using Image Knowledge Builder, a ”show and tell” interface for easy training and validation. Fishermen can reinforce the learning at anytime when needed. The use of CogniSight has reduced significantly the number of crewmembers on the boats (by up to six persons) and the time at sea has shortened by 15%. The prompt and strong return of the investment to the fishing fleet has increased significantly the market shares of Pisces Industries, the company integrating CogniSight systems to its filleting machines.