An image classification method based on fusion clustering in a capsule network

Abstract

The invention discloses an image classification method based on fusion clustering in a capsule network, and solves the problems of serious calculation energy consumption, poor network expansibility and unstable classification accuracy when a routing iteration algorithm used in the prior art is used for solving a weight coefficient of a high-grade feature of a combined image. The method comprises the following implementation steps: (1) inputting a natural image to be classified; (2) obtaining a training sample set and a test sample set; (3) constructing a capsule network; (4) extracting prediction feature vectors of samples in the training sample set; (5) obtaining a clustering center vector of samples in the training sample set; (6) training a capsule network; and (7) classifying the testsample set. The method has the advantages of being simple in model, high in training speed and good in network expansibility, and can be used for natural image classification.

Description

technical field [0001] The invention belongs to the technical field of image processing, and further relates to an image classification method based on clustering integrated into a capsule network in the technical field of image classification. The invention can obtain the clustering center vector of the natural image according to the capsule network, and use the clustering center vector to classify the natural images containing different types of objects. Background technique [0002] Natural images refer to the images obtained by shooting natural scenes with cameras and other shooting equipment, or the images captured by video cameras shooting and recording natural scenes. Image processing, which classifies images according to the image features obtained through processing. [0003] In their paper "An Optimization View on Dynamic Routing Between Capsules" (International Conference on Learning Representations, 2018), Dilin Wang et al. proposed a capsule network to classify...

AI Technical Summary

Problems solved by technology

Although this method uses the capsule network to extract the advanced features of the image to obtain better classification results, the method still has the disadvantage that, since the routing algorithm requires multiple iterations, the optimal number of iterations is difficult to find. The calculation cost is quite high, which limits the scalability of the capsule network structure, making the calculation energy consumption problem more serious when the capsule network deepens the number of layers, and cannot guarantee the network to obtain good classification performance

The specific steps of this method are: constructing a capsule network, inputting image training sets into the capsule network, completing image classification, recognition and calibration after training and learning, inputting images to be classified into the capsule network, and obtaining the model value with the largest value among all output vectors of the capsule network as Classify the results, and set the reconstructed network structure of the capsule network as a deconvolution layer, restore the vector to an image through the deconvolution operation, and adjust the network parameters by comparing the error between the restored image and the original image. This method is used in the capsule network. Using the deconvolution layer as the reconstruction network structure reduces the amount of calculation parameters, but the method still has the disadvantage that in image classification tasks, it is not necessary to use the deconvolution operation to restore the vector to an image. The deconvolution layer and the network still have the disadvantage of a large number of parameters, which leads to the method having a large number of model network layers in image classification, complex structure, high energy consumption in the training process, and low model classification efficiency.

Images