用于特征提取与融合的卷积神经网络结构和应用综述OA

Convolutional neural network is one of the core network structure types in deep learning,widely used in com-puter vision.Convolutional neural networks use local connections and different receptive fields to achieve multi-scale fea-ture extraction and multi-dimensional fusion,ultimately obtaining feature maps of different scales and abstraction levels that contain comprehensive and complete information about the original object.Based on the learned feature maps,vari-ous image segmentation,object localization,recognition,and object detection tasks can be achieved.In order to clarify various convolutional neural network structures for feature extraction based on convolutional neural networks,this paper provides an in-depth summary and induction of various convolutional neural network structural units for feature extrac-tion and fusion,including convolution,pooling,downsampling,etc.In addition,in order to gain a deeper understanding of the process of data extraction and fusion,this article introduces several classic convolutional neural network structures from the perspective of feature extraction and fusion,including AlexNet,VGG Nets,GoogLeNet,ResNet,SENet,and DenseNet.In addition,a summary of the mainstream convolutional neural networks that have emerged in the past two decades are also provided,allowing readers to take a higher perspective on the development of convolutional neural net-works for local feature extraction and fusion processing in different scales of length,width,and depth directions.Based on practical applications,this article introduces commonly used convolutional neural networks and convolutional hybrid net-works from the perspectives of image classification and recognition,object detection,and image segmentation.Finally,this article explores the future research and development directions of convolutional neural networks from the aspects of model compression,model security,model mixing,and the introduction of advanced data models,especially in the combi-nation of CNN and attention mechanisms,and evaluates different ways of combination.