Image classification method based on CNN fragmentation multi-scale feature fusion

Abstract

The invention discloses an image classification method based on CNN fragmentation multi-scale feature fusion, and the method is implemented according to the following steps: firstly obtaining a labeled image training set, carrying out the preprocessing of the labeled image training set, improving the sample diversity, and obtaining a complete image training set; secondly, constructing a feature extraction convolution module for fragmentation multi-scale feature fusion, and performing feature conversion and feature extraction on an image in the complete image training set by the convolution module to obtain an image feature vector representing an image sample; connecting the image feature vectors into a softmax classifier to serve as output of image recognition; and finally, training an obtained neural network model through a stochastic gradient descent method and a back propagation algorithm, and obtaining a finally completed model after loss function convergence training is finished.According to the image classification method based on CNN fragmentation multi-scale feature fusion, barriers among different network optimization normal forms are effectively broken through, the imagefeature extraction performance of a network model is further improved, and the model precision is improved.

Description

technical field [0001] The invention belongs to the technical field of computer vision based on deep learning, and in particular relates to an image classification method based on CNN-based sliced ​​multi-scale feature fusion. Background technique [0002] With the great success of CNN (Convolutional Neural Network) in the LSVRC image classification competition in 2012, it triggered a research boom of CNN in computer vision tasks. Traditional image classification methods need manual feature extraction based on a large amount of prior knowledge. This method is not only time-consuming but also the effect of the extracted features is not ideal. Compared with the traditional method, the biggest charm of CNN is that with the support of sufficient computing power, as long as there is sufficient training data, CNN can automatically learn the best features representing the original image according to the distribution of training samples, so "data-driven" is The most notable feature...

AI Technical Summary

Problems solved by technology

[0006] However, these optimization paradigms are all proposed in a single dimension such as network width or network depth, which has certain limitations for further improving the overall performance of the network.

Images