Small target semantic segmentation method and system based on low-rank mixed attention mechanism

Abstract

The invention provides a small target semantic segmentation method and system based on a low-rank mixed attention mechanism, and the method comprises the steps: constructing a segmentation model, specifically, constructing a segmentation data set, and carrying out the data preprocessing; performing feature extraction on the preprocessed segmented data set: expanding a receptive field based on expansion convolution; as the substitution of a pooling layer, adding a low-rank mixed attention mechanism in the network to aggregate feature information and extract high-order semantic features; for the extracted features, calculating weighted cross entropy loss and Dice loss, and training the network by using a mixed loss function to obtain a network model; and for the to-be-segmented object, outputting a small target segmentation result by using the segmentation model. The method effectively solves the problem that a relatively small target size in an image data set is easy to cause serious class imbalance, and the problem that accurate reconstruction of an image is difficult to realize in a decoder due to information loss in a feature learning process.

Description

technical field [0001] The invention belongs to the technical field of image segmentation, and in particular relates to a small target semantic segmentation method and system based on a low-rank mixed attention mechanism. Background technique [0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art. [0003] With the continuous development of deep learning technology, the semantic segmentation methods of images have made remarkable progress. As a fundamental topic in computer vision, the goal of semantic segmentation is to assign a semantic class label to each pixel in an image. In recent years, deep learning-based semantic segmentation has been actively studied by many scholars in various challenging tasks, such as autonomous driving, virtual reality, and computer-aided diagnosis, etc. Among them, semantic segmentation can usually achieve satisfactory results for the segmenta...

AI Technical Summary

Problems solved by technology

However, since the operation of non-local matrix operations is easy to introduce large computational complexity, it is also not conducive to the landing of the model

Images