Researches Advanced in Image Recognition based on Deep Learning

Abstract

In the discipline of computer vision, which tries to create a model to predict the category of objects present in a given image, image recognition has always been a research hotspot. Early image recognition models were mainly based on manual features, and their recognition accuracy and generalization ability often fluctuated greatly with changes in the scene, which could not meet the actual application requirements. Convolutional neural networks have progressed quickly, which has accelerated the development of deep learning-based image recognition. in this paper, we detail the development of image recognition technology. Specifically, we introduce the three approaches of convolutional neural network, recurrent neural network, and graph neural network as the traditional methods of image recognition. including their design ideas, key steps, advantages and disadvantages. We also compare the recognition accuracy of different methods on the ImageNe dataset to try to explore the application boundaries of different methods. Finally, we go over the difficulties in the area of image identification and project its future growth.