基于卷积神经网络的图像分类研究进展

基于卷积神经网络的图像分类算法的优势是传统方法无法比拟的。卷积神经网络利用其设计好的网络结构和权值共享的特点，能够从数量庞大的训练数据中学习图像底层到高级语义的抽象特征，而且端到端的学习省去了在每一个独立学习任务执行之前所做的数据标注。多年来，卷积神经网络经过科研人员的探索和尝试，从最开始的多层神经网络模型，演变出多种优化结构，性能不断提高。本文介绍了基于卷积神经网络图像分类算法的研究进展，叙述了卷积神经网络在图像分类中的经典模型和近年来的改进方法，并对各个模型进行分析，展示各种方法在ImageNet公共数据集上的性能表现，最后对基于卷积神经网络的图像分类算法的研究进行总结和展望。

Research Progress of Image Classification based on Convolutional Neural Network

The advantages of image classification algorithms based on convolutional neural network are unmatched by traditional methods.Convolutional neural network uses its designed network structure and weight sharing characteristics to learn abstract features from the bottom of the image to the high-level semantics from a huge amount of training data.End-to-end learning eliminates the need for data labeling before the execution of each independent learning task.Over the years,after research and experimentation by researchers,the convolutional neural network has evolved a variety of optimized structures from the first multilayer neural network model,and its performance has been continuously improved.This article introduces the research progress of image classification algorithm based on convolutional neural network,describes the classic model of convolutional neural network in image classification and the improved methods in recent years.Each model is been analyzed,and the performance of various methods on ImageNet public dataset are shown.Finally,the research of image classification algorithm based on convolutional neural network is summarized and prospected.