A student of Novosibirsk State Technical University (NETI), creates a feeder that will be able to recognize an animal in the absence of the owner and provide the necessary amount of feed and water.
"The main problem that my invention solves is providing access to food and water for pets during the day through the use of neural networks that are being implemented everywhere in human life. There are many smart feeders, but in these projects, feeders and drinkers are considered as separate devices. The goal of my project is to integrate the feed and water supply process into a single smart device," said Polina Gumennikova, a 2nd year student at the Faculty of Mechatronics and Automation of NSTU-NETI.
Polina started working on the project back in 2022, when she thought about the topic of her college thesis, the development was supported by her supervisor and went beyond the scope of the certification work. Now the work is being carried out on the basis of NSTU-NETI with the support of Nikita Petrov, a physics teacher.
"The main difference from the presented analogues is the use of a neural network. Smart feeders, which are currently on the market, work using various sensors: motion, sound, infrared, but such a tool can respond to any activity nearby. And the neural network will insure against a reaction to random movement or noise and will be able to distinguish a specific animal that came to the feeder purposefully," Polina noted.
The feeder is based on a microcomputer for interacting with a neural network that receives a signal from a camera built into the feeder body, which allows you to recognize the animal and start the feed and food supply mechanism. Portions are formed using a stepper motor, its rotation is carried out at a strictly set angle, which forms an individual portion size for each animal. The water supply to the bowl is controlled by a solenoid valve controlled by a neural network.
According to the student, a special time control function will help to avoid overeating in a pet. The neural network will be able to track how much time has passed since the last meal and water. The feeding interval of the animal can be set individually at will or according to the indications of the veterinarian.
The project is under development — a 3D model of the feeder already exists, new components for printing the prototype are expected. After this stage, the neural network will be programmed and trained: many photos of the pet from different angles are uploaded to the database to create a reference image, with which the neural network will later compare the information received from the camera. The more different images are uploaded to the database, the more likely it is that the feeder will work accurately.
The developer plans to conduct the first tests of the feeder, in which it is supposed to calibrate the operation of valves that supply food and water, as well as minimize possible neural network errors in animal recognition.