Dog breed recognition using convolutional neural networks
Loading...
Date
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Данная работа имеет практическое применение в современном мире для безопасности людей. С помощью такой системы можно различать собак на бойцовые породы и нет. По закону к выгулу бойцовых пород предъявляются особые требования(намордник, поводок, бирка). Во многих городах на улицах расположены камеры и их можно использовать для нахождения нарушителей.
В данной работе было реализовано решение, позволяющее определить породу собаки на изображении. Решение основывается на сверточных нейронных сетях, а в основе модели лежит архитектура ResNet152. С применением техники Transfer Learning нейросеть обучается под задачу распознавания породы собак и хорошо ее решает. В рамках работы была выбрана база данных Stanford Dogs Dataset. Также все изображения были предобработаны и аугментированы для расширения базы данных. Нейронная сеть реализована с помощью фреймворка PyTorch на языке Python. В итоге была получена хорошая точность в данной задаче.
This work has practical application in the modern world for the safety of people. With this system, you can distinguish between fighting and non-fighting breeds. According to the law, special requirements(muzzle, leash, tag) are imposed on the walking of fighting breeds. In many cities, there are cameras on the streets and they can be used to find offenders. In this work, we implemented a solution that allows us to determine the breed of dog in the image. The solution is based on convolutional neural networks, and the model is based on the ResNet152 architecture. Using the Transfer Learning technique, the neural network is trained for the task of recognition dog breeds and solves it well. The Stanford Dogs Dataset was selected as part of the work. Also, all images were pre-processed and augmented to expand the database. The neural network is implemented using the Python framework PyTorch. As a result, we obtained good accuracy in this problem.
This work has practical application in the modern world for the safety of people. With this system, you can distinguish between fighting and non-fighting breeds. According to the law, special requirements(muzzle, leash, tag) are imposed on the walking of fighting breeds. In many cities, there are cameras on the streets and they can be used to find offenders. In this work, we implemented a solution that allows us to determine the breed of dog in the image. The solution is based on convolutional neural networks, and the model is based on the ResNet152 architecture. Using the Transfer Learning technique, the neural network is trained for the task of recognition dog breeds and solves it well. The Stanford Dogs Dataset was selected as part of the work. Also, all images were pre-processed and augmented to expand the database. The neural network is implemented using the Python framework PyTorch. As a result, we obtained good accuracy in this problem.