Research On Image Classification Based On Modular Combination Convolutional Neural Networ

At present,the main research directions of machine vision are divided into: image classification,image segmentation and object detection.Among them,image classification is a classic research problem in computer vision.This problem has been explored by many computer scientists.This paper attempts to modularize the classical convolutional neural network.Developers can combine any existing convolutional neural network modules to explore higher recognition accuracy.Moreover,when new neural network modules are developed by researchers,they can also be modularly connected to existing neural networks for combination.The new network designed in this paper has 16 convolutional layers,which integrates the ideas of VGGnet,InceptionNet and ResNet,and has the network architecture characteristics of the classic convolutional neural network to construct a new convolutional neural network.After comparative analysis,the new network has better recognition effect than the original network in terms of accuracy and generalization performance.The structure of the article is arranged as follows: First,the main research methods in the field of machine vision and the advantages and disadvantages of existing research results are introduced.On this basis,the existing deep learning network is modularized and a new convolutional neural network model is built.Train on an existing dataset and calculate the accuracy.This paper mainly studies the modular packaging for convolutional neural networks,and then designs the combined neural network to compare with the original classical convolutional neural network.This paper modularizes the classic convolutional neural network and constructs a combined convolutional neural network.Compare an existing neural network with a combined neural network.Improved recognition accuracy.The research kernel test shows that the method achieves better results than the current classical convolutional neural network on the dataset..