Image Feature Extraction Based On Deep Learning

In this paper,the image feature extraction method based on depth learning is compared with the method based on machine learning.By using the Common object image database provided by Microsoft Company,we measure the image feature extra draw action a way of depth learning and the image feature extraction of machine learning,respectively,under the normal state,the image normalization and the working effect under the condition of indefinite angle.The features of the image pick up method based on convolution neural network is used respectively,the traditional method are applied to different databases.Commonly used the features of the image extraction measures,such as directional gradient histogram and scale invariant feature transformation,require a priori knowledge of the relevant domain in the design of features.The structure of the paper is divided into three parts.First,the paper mainly introduces the theoretical knowledge and the related principles,including the correlation principle of the directional gradient histogram and the scale invariant feature transformation,as well as the support vector machine and the non-maximal suppression related theory in the algorithm,and the theory of the convolution neural network.The second part of the article is an empirical analysis link.The thesis will be divided into two steps.The first step is to clean up the data,select one thousand human and aircraft pictures,eight hundred of them as training sets and two hundred test sets,and then use the direction gradient histogram and the scale invariant respectively.Feature extraction and feature extraction model of convolution neural network for feature extraction.The third part of the article is the summary and analysis.Image feature extraction technology can be applied in many fields.By comparing traditional feature extraction methods and depth networks,it can understand their respective characteristics using environment and research ideas,for specific environment,such as fixed view video surveillance video,sloshing background foreground extraction,multi view pedestrian recognition and UAV search and rescue targets.The results of this study show that the method of image feature extraction based on depth learning is higher than the method of image feature extraction based on machine learning in the normal state,and the accuracy rate of image gray normalization and angle uncertainty are all higher than that of machine learning.