Medical Image Segmentation Method Based on Lightweight Fully Convolutional Neural Network

Abstract

The invention claims to protect a medical image segmentation method based on a lightweight fully convolutional neural network. First, the data set is preprocessed by grayscale, normalization, contrast-limited adaptive histogram equalization (CLAHE), gamma correction, etc.; then, the training set is randomly extracted from the patch and the test set is sequentially extracted from the patch image to Complete data enhancement; then, build a fully convolutional neural network architecture consisting of a contraction path (left) and an expansion path (right), and design leave‑one‑out training for data sets with a small number of images Method; finally, through channel sparse regularization training, cutting channels whose scale factor is smaller than the set threshold, and fine-tuning the cut network to complete the BN channel model cutting, obtain a lightweight fully convolutional neural network, and input the test data to the network In the fast test to complete the image segmentation. The lightweight fully convolutional neural network not only ensures the high segmentation accuracy advantages of the deep network, but also improves the test speed of the image segmentation network.

Description

technical field [0001] The invention belongs to the technical field of image feature extraction and segmentation methods, in particular to a medical image semantic segmentation method for a lightweight fully convolutional neural network. Background technique [0002] Fully Convolutional Neural Network (FCN) has dominated the field of image segmentation, especially semantic segmentation, because it has the characteristics of accepting input images of any size, high computing speed, etc., and can achieve pixel-level image segmentation effects. [0003] At present, a lot of experiments have been done on the fully convolutional neural network. Since FCN in 2015, a batch of classic models have been produced, such as UNet and UNet-like structures. These models are based on FCN and continuously deepen the number of network layers to improve the network. of the segmentation accuracy. But it also brings a series of problems, such as high network training environment requirements, sl...

AI Technical Summary

Problems solved by technology

It is widely used in convolutional neural networks, but the research and application of model clipping for full convolutional neural networks is still relatively small

Images