Image Classification And Labeling In Complex Environments

The development of computer network technology,visual image processing technology and deep learning has provided us with more and more solutions for image feature extraction and image classification.In the face of massive data image information,how to effectively classify and annotate images is a great challenge at present.When dealing with data in a large number of complex environments,traditional neural network models are no longer able to meet the current requirements for image classification and annotation.Studying the convolutional networ k model has become an inevitable trend for the effective development of image classification and labeling applications.This paper first describes the research background and significance of image classification and annotation,image feature extraction,im age classification,image annotation,deep learning and convolutional neural network research at home and abroad.Then introduced the development and structure of artificial neural network,studied neural network back propagation algorithm,briefly describ ed deep learning,focused on the basic knowledge of convolutional networks,including local receptive fields,weight sharing techniques,multi-convolution kernels and pools.Operation.Then focus on designing convolutional networks in complex environments.This paper mainly studies the structure design of convolutional networks and the visualization of network characteristics.The number of layers in the network structure design and the corresponding parameters in the network layer largely determine the quality of subsequent image classification and labeling.Therefore,designing a reasonable network layer and optimizing parameters based on the actual situation is an important research content of image classification and annotation.Training the network model and adjusting the parameters through the training process is to better extract the features of the image.This paper uses the convolution network to extract the features of the image,the low-level network extracts the bottom features of the image,and gradually extracts the salient features of the image in favor of the classification task.The feature information learned at each level of the convolutional network is presented to visualize the image features.The convolutional neural network can be explored from a certain angle.It helps us better understand how the convolutional neural network extracts image features and helps us improve Network classification performance.Based on the deep learning framework,this thesis designs a convolutional network structure in a complex environment and optimizes the network parameters.First,the convolution network designed in this paper is initially identified as an 8-layer network model structure,including the first five convolutional layers and the three fully connected layers including the output layer.The third and fourth layers of the first five convolution layers are not connected to the pooling layer.This is designed so that the size of the input convolutional layer image is smaller than the size of the corresponding convolution kernel.The first convolutional layer is pooled randomly,and the second and fifth convolutional layers are pooled.Then,the specific parameters for each layer of network design,including each layer of convolution kernel,step size and whether to perform pixel compensation and other specific operations.Use the PReLU activation function instead of the ReLU activation function.The output layer chooses to use the Softmax classifier for image classification.Finally,on the basis of the data set under the complex environment,the convolution network designed was trained,and the parameters of the network network were adjusted.The photos taken at random were tested and the results were analyzed.Through experimental analysis,the convolutional network structure designed has achieved very good performance.