Image feature extraction and training method based on three-dimensional convolutional neural network

Abstract

The invention discloses an image feature extraction and training method based on a three-dimensional convolutional neural network. The method comprises the steps that 1, dimension normalization processing is performed on input images for feature extraction; 2, the three-dimensional convolutional neural network comprising a convolutional layer, an activation layer, a pooling layer, a full-connection layer and an output layer is constructed; and 3, the constructed three-dimensional convolutional neural network is trained to obtain an optimized three-dimensional convolutional neural network, feature extraction is performed on the input images, and classified recognition of the input images is completed. According to the method, feature extraction and recognition of the three-dimensional images are performed through the three-dimensional convolutional neural network, the three-dimensional convolutional neural network directly performs convolution on the three-dimensional images, three-dimensional spatial features of the images are extracted, feature modes of the three-dimensional images can be expressed more effectively, and therefore the purpose of classified recognition of the imagesis achieved.

Description

Technical field [0001] The invention belongs to the field of image recognition and deep learning, relates to three-dimensional image feature extraction and recognition, and is specifically an image feature extraction and training method based on a three-dimensional convolutional neural network. Background technique [0002] Image recognition is a technology in which computers process, analyze, and understand images to identify targets and objects in various patterns. It has been applied to industrial security, life, education and other aspects. Image recognition is an important field of artificial intelligence. In order to teach computers to recognize images like humans, people have proposed many image recognition methods. The traditional recognition process includes image preprocessing, image segmentation, feature extraction and judgment matching. Therefore, there are a large number of different algorithms in each intermediate step, and each intermediate step affects the final...

AI Technical Summary

Problems solved by technology

For these three-dimensional images, the current solution is to average all images in a certain dimension to obtain a two-dimensional image, and then use a two-dimensional deep learning algorithm to identify it. This method combines all images in a certain dimension Averaging is performed, so a lot of information is lost, and not all features can be effectively extracted

Another method is to regard a certain dimension as a channel of a two-dimensional convolutional neural network, that is, how many slices the image has in this dimension, then how many channels there are, and then use the same two-dimensional convolutional neural network Algorithm for recognition, although this method seems to have no loss of information, it turns the three-dimensional image into an isolated two-dimensional image, extracts features from the two-dimensional image, and extracts two-dimensional features, without considering the characteristics of the two-dimensional image The relevance in the third dimension, and the amount of calculation is large, so it does not conform to the essence of the three-dimensional image, the information loss in the recognition process is large, and the recognition degree is low

Images