Semantic segmentation method based on reverse attention model

Abstract

The invention relates to a semantic segmentation method based on a reverse attention model. The method mainly comprises the following steps: acquiring an image data set, and constructing a training set and a test set; constructing a deep semantic segmentation network model, wherein the deep semantic segmentation network model comprises a basic network model and a reverse attention model; and inputting the features output by the basic network into a reverse attention model to calculate attention views, gradually counteracting the attention views on the low-level output features of the basic semantic segmentation network, and fusing the attention views with the output features of the basic network and the up-sampling features thereof to obtain a final segmentation result. According to the model, only basic semantic segmentation network output features are used for calculating an attention view, and low-level features are guided to be fused into the basic semantic segmentation network output features, so that noise in the low-level features of the model is suppressed, and the robustness and segmentation precision of the semantic segmentation model are improved; meanwhile, a loss function based on Gumbel softmax is added to high-level output features of the basic semantic segmentation model, so that the model training speed is increased.

Description

technical field [0001] The invention belongs to the technical field of image processing, and in particular relates to a semantic segmentation method based on a reverse attention model. Background technique [0002] In recent years, deep learning has developed by leaps and bounds. The deep learning model represented by convolutional neural network (CNN) has once again ignited the silent neural network, setting off a wave of deep learning in academia and industry. [0003] The early DNN-based segmentation model was limited by the fixed size of the input image. To solve this problem, Long and Shelhamer of the University of Berkeley proposed a fully convolutional network (FCN) for image semantic segmentation, by using convolution instead of full Connect the layers, and use techniques such as deconvolution and upsampling to map the dense predictive image output by the network to the original image, thereby achieving end-to-end semantic segmentation, and the DNN model can handle i...

AI Technical Summary

Problems solved by technology

[0005] However, this method increases the number of model learning parameters and is more complicated.

Images