Application of the Hopfield Neural Network for the State of Chicken Embryos Development Recognition

Abstract

Monitoring chicken embryos development is an important part of the hatching process to determine when the egg is not developing and when the egg is close to hatching. The ability to closely monitor the embryo development allows determining the optimal time to change incubation parameters, such as humidity, to create the best conditions for hatching, leading to more efficient chick production. To monitor the chicken eggs development during incubation, it was proposed to automate the process of ovoscoping through determining the chick embryos state by using the NI EVS-1464R technical vision system from National Instruments, where a number of hardware is available for image acquisition (video capture cards for digital cameras with different interfaces, real-time systems and smart cameras), and the Hopfield’s artificial neural network. This allows associative memory to be implemented. The main task of associative memory is reduced to storing input (training) samples in such a way that when a new sample is presented, the system can generate an answer – which of the previously stored samples is closest to the received image. This neural network changes its internal state for each iteration and stops when the current one coincides with the previous one. In this case, the neural network is said to converge to one of the states stored in its memory. If a similar image is not found in memory, the network may issue a nonexistent attractor. The modeling of recognition of the chicken embryos various states is carried out. When perform the ovoscoping, the light sector of the egg area corresponds to a signal equal to 0, and the dark sector to signal 1. In total, during the modeling, the states of 23 sectors of the bird’s egg area are analyzed. A software tool based on the Hopfield neural network is developed in C++ and its ability to identify live and non-living embryos of chicken chicks has been tested. Correct identification of distorted vectors allows the use of the Hopfield net in the poultry industry, which will reduce excessive operator fatigue and eliminate the false rejection of good eggs.