Design And Implementation Of Multi-label Image Classification System Based On Deep Learning

With the rapid development of the Internet industry,image data continues to grow,and the era of image big data has followed.The original single-label image classification technology could not meet the image classification requirements with multiple complex semantics.Multi-label image classification technology came into being.However,we need to work hard to innovate a multi-label image classification method with fast classification and high accuracy.In this paper,we study the classification performance of some different network structures on different data sets for the process of multi-label image classification.The degree of fineness of the extracted features is important to the effect of the multi-label classifier on the final classification results.Different network structures and parameter settings will also affect the final classification accuracy.In this paper,we mainly explore a new type of network structure based on previous research,and strive to improve the classification accuracy of multi-label images.The main work of this paper is as follows:Explain the basic knowledge of deep learning,and focus on convolutional neural networks from various aspects.The model used in this paper is based on the VGG network proposed by the Department of Science and Engineering at Oxford University.It mainly improves the pooling layer which is composed of 13 convolutional layers and 3 fully connected layers,and replaces the pooled structure in the original network with spatial pyramid pooling.At the same time,the objective function in the original network is replaced by a new objective function combined by the Max-Margin objective function,the Max-Correlation objective function and the Correntropy loss function.According to the research,the proposed HCP network structure and our improved VGG network have their own strengths.Therefore,we have combined two different network structures to compensate each other for each other.Good effect.We experimented with the new network we designed on the PASCAL VOC 2007 and PASCAL VOC 2012 data sets.The network was trained and tested using two data sets.The test results show that the network we designed can handle multi-label image classification tasks very well.We used different methods for comparative experiments,such as manual design feature methods,shallow machine learning methods,and methods for different kinds of network structures to perform experiments on the same data set.The experimental results show that the method of deep learning network structure is more efficient than the artificial design feature and shallow machine learning method.we implemented a simple image classification system.The system is simple to use and the image classification effect is remarkable.