Deformable convolution hybrid task cascade semantic segmentation method based on embedded balance

Abstract

The invention designs a deformable convolution hybrid task cascade semantic segmentation method based on embedded balance, which is used for realizing image target recognition and semantic segmentation, and comprises the following steps: inputting a cut image into a pre-trained neural network; mapping the two samples to the same scale space through a feature pyramid network; performing informationfusion on semantic features extracted from different hierarchies; predicting a pixel-level segmentation result by adopting a convolution layer; performing feature extraction on the input image by adopting a deformable convolutional neural network at the convolution and pooling part of the feature pyramid network to obtain a feature map; dividing the feature map into parts with the same size; inputting a feature map obtained after passing through the feature pyramid network into a regional candidate network for training the network; wherein the region candidate network comprises a target detection classifier and a candidate frame positioning classifier, the target detection classifier outputs a target recognition result and prediction accuracy, and the candidate frame positioning classifier can provide accurate positioning for candidate regions and output candidate frames of a plurality of candidate regions. According to the method, the semantic segmentation positioning accuracy and the segmentation accuracy are improved.

Description

technical field [0001] The present invention designs a deformable convolution mixed task cascade semantic segmentation method based on embedding balance, which is used to realize image target recognition and semantic segmentation. Background technique [0002] In traditional semantic segmentation, the semantic segmentation task is defined as dividing an image into several disjoint parts, and these parts have their own semantics, that is, the segmented parts only contain one type of target or object. In traditional semantic segmentation, there are a lot of research work on semantic segmentation methods based on user interaction. This type of method often selects an area by the user, and then uses the color similarity, texture similarity, and edge features of other areas and the selected area as the connection weights of these areas, and finally uses Conditional Random Fields (Conditional Random Fields, CRF) or The Graph-Cut method is used to segment the image. Many ideas in ...

AI Technical Summary

Problems solved by technology

However, this type of method has certain disadvantages: when there are multiple targets in the image, human participation is required, and the adaptability is low

Images