Automatic Counting Algorithm of Fry Based on Machine Vision System

Abstract

In this study, an automatic counting system for fry based on the machine vision is designed, and the dataset of Pseudosciaena crocea (2–4cm) and Grass Carp (3–6cm) are collected. Aiming at the problems of large moving distance and large deformation of fry between consecutive frames, a matching algorithm based on probability density function (PDF) and a dynamic counting strategy are proposed, which can ensure the accuracy of fry counting effectively. Finally, the experimental results show that the automatic counting system can effectively separate fry, and the average accuracy of counting is more than 97.9%.