Research On Target Recognition Based On Deep Learning

Target recognition technology has been applied in all aspects of industrial production,especially in military,financial,high-end equipment,traffic safety and other fields,the key technology is how to improve the recognition accuracy and recognition speed,which directly determines the practicality and security of the technology.deep learning is a branch of the machine learning domain,which is an unsupervised learning algorithm,deep learning in natural language processing,natural image features of the learning effect far more than the traditional machine learning-related technologies,traditional image recognition algorithms are manually adding specific features to identify the target,not only the recognition rate is low,but also difficult to extract,because of the particularity of the neural network structure,the neural network can learn the inherent law in the training process,which is called the sample feature.this high abstract ability makes the training network have very high generalization ability.In this paper,we focus on the application of depth learning in target recognition.The main research works are as follows:(1)This paper introduces the basic structure of the neural network model,in which the convolution layer and the pool layer cross the stack to learn abstract features.The down sampling layer can preserve the original image information and reduce the data processing amount.finally,the back propagation algorithm is introduced.(2)A special unsupervised depth neural network is introduced--depth auto-encoder,input and output vectors have the same dimension,often according to some form of input vector,through the hidden layer to learn a data representation or the original data for effective coding.(3)A selective search algorithm combined with a binary normalized gradient detection algorithm is proposed.Compared with the traditional sliding model,the selective search can significantly reduce the search space.Because these two algorithms are combined to extract the pre-selected regions in the use of image information,,The algorithm proposed in this paper can improve the detection accuracy by 6% compared with other algorithms,and can reduce about 1500 pre-selected area windows when the same detection rate and recall rate are reached.