Design And Implementation Of Image Recognition System Based On Deep Learning

Image is an important way for humans to get information.With the development of science and technology,people generate more images.The use of computers for image recognition plays a major role in many industries.It is a hot topic of current research.The traditional image recognition technology adopts artificial selection features and uses pattern matching,linear classification and other algorithms for image recognition.The accuracy of recognition depends on the quality of the selected features,and it is difficult to extract features that can express the key of the original data.The deep learning method is used for image recognition and has a great advantage over traditional methods.The commonly used method is to build a deep neural network and use some data for training.The deep neural network will automatically extract image features and achieve better recognition results.Convolutional neural network is a key network for deep learning in image recognition.By constructing a convolutional neural network,image features can be extracted layer by layer.The network construction and training method are very important for deep neural network.Excellent network design can achieve better training results with fewer parameters.Special network components can speed up the training process.The appropriate training methods can give full play to the capabilities of the network.This paper introduces the relevant knowledge of deep learning applied to image recognition,tests a variety of deep neural networks,uses different training methods,and analyzes their accuracy.Using the method of model splicing,multiple networks are integrated and parameters of network training are adjusted to obtain better results.Based on the image target recognition,a post-processing method is designed to calculate the position relationship and the size relationship of the two targets.The class histogram is generated and can be used to analyze the logical relationship of the target.The rationality of the target recognition can be calculated and used to modify the recognition result.A graphical user interface for deep learning image recognition is designed and implemented.The system supports the selection of parameters of the training network,feedbacks the training progress in real time,and provides a dedicated page view image test result.Finally,to provide users with image recognition network services.And use this service to identify street images.