Research On The Key Technology Of Image Classification Based On Deep Learning

Image classification problem belongs to the machine learning and computer vision fields,it aims to recognize and classify the objects in image contents.Applying deep learning algorithms into the image classification problem plays a critical role in solving the problem.The classification accuracy of state-of-art algorithm on the large scale dataset as ImageNet is almost the same with human now.This achievement makes the focus of research switch to the image caption problem,which is more challenging.The deep learning frameworks are foundations and standards of deep learning applications in specific tasks.Many technology companies and research institutions have launched and maintained their own deep learning frameworks,in order to become the industry standard.Based on the image classification task,this paper researches and compares the linear classification algorithms and deep learning classification algorithms,focus on the implementation of different algorithms and deep learning models.Based on the implementation of algorithms,it successfully built a lightweight deep learning framework,and verified its validity through experiments.The image caption problem is a new challenge after the image classification problem.It aims to descrip the image contents with natural language sentences.This paper researches the application of deep learning models in the image caption problem,introduces the current research progress of image caption problem from three aspects,which are the datasets,evaluation metric,and the models.Utilizing the lightweight deep learning framework to train a convolutional neural network as a classification model,this paper constructed an on-line image classification system,which includes a WeChat public account as the front end and a aliyun cloud server as the back end,and verified the validity of the image classification system.