FPGA IMPLEMENTATION OF NEUROCOMPUTERS TO RECOGNIZE THE STATE OF DEVELOPMENT OF CHICKEN EMBRYOS

Abstract

The hatchery industry is one of the main industries providing food for the population and plays an important role in poultry production. Egg hatchability is affected by many factors such as egg handling, egg fertility, parent flock problem, etc. However, the most important factor is the assurance that the eggs placed in the incubator are indeed fertilized. In most hatcheries, the process of separating fertilized and infertile eggs is carried out by specialists in the traditional way with the help of human vision using ovoscopes. During the hatching of poultry, eggs are periodically ovoscoped in order to determine the condition of the embryos of the chicks. Early detection of infertile eggs and eggs with dead embryos allows hatcheries to save energy, handling costs and prevent contamination of good eggs from broken eggs. The ovoscoping process is laborious and inefficient due to eye fatigue and operator errors, which have to check up to a thousand eggs per day. The article solves the problem of automating the process of eggs ovoscoping by adding to the machine vision system the neurocomputers capable of recognizing the embryos possible states at different stages of incubation. The two neurocomputers projects are implemented in Xilinx FPGA, which are designed to automate the monitoring of chicken embryos development by recognizing their condition during hatching. The first neurocomputer implemented in the xc3s500e FPGA contains 23 neurons not covered by feedback and counts the dark sections of the egg under study. Then the value of the threshold set for this period is subtracted from the received sum, and the obtained result is used to generate the output signals “good” or “bad”. The threshold value for different periods of ovoscoping can be changed by using interchangeable connecting blocks, which set the threshold values for the neurocomputer operation. The second neurocomputer, implemented in the xcv600e FPGA, contains 15 neurons covered by feedback, performs the functions of a Hopfield network and allows to recognize good eggs and eggs with dead embryos at late hatching periods with high reliability. The developed low-cost neurocomputers can complement the machine vision system for detecting fertilized eggs in the hatchery industry.