Using Convolutional Neural Networks for Image Detection and Recognition Based on a Self-Written Generator

Abstract

Object recognition is a branch of artificial vision and one of the pillars of machine vision. It consists of the identification of forms pre-described in a digital image and, in general, in a digital video stream. While it is generally possible to perform recognition on video clips, the learning process is usually performed on images. In this paper, we consider an algorithm for classifying and recognizing objects using convolutional neural networks. The purpose of the work is to implement an algorithm for detecting and classifying various graphic objects fed from a webcam. The task is to first classify and recognize an object with high accuracy according to a given dataset and then demonstrate how to generate images to increase the size of the training data set using a self-written generator. The classification and recognition algorithm used is invariant to translation, translation, and rotation. A significant novelty in this work is the creation of a self-written generator that allows you to apply various types of augmentation (artificial increase in the size of the training sample by modifying the training data) to form new groups (batches) of modified images each time.

Keywords

• image detection
• image recognition
• convolutional neural networks
• r-cnn (regional convolutional neural networks) model
• augmentations
• subsampling