Multi-Class Segmentation Based on the B-UNet Model

To solve the problem that UNet performs poorly in multi-class segmentation on large datasets, this paper proposes a new neural network model, B-UNet. UNet has some problems, such as poor feature processing method and large splicing span of features, leading to its poor feature extraction ability for complex images. Although UNet++ solves the problem of large feature span, the number of parameters in the model is greatly increased. Based on UNet, the B-UNet model obtained by introducing bridge features not only solves the problem that UNet has poor performance in multi-class segmentation on large datasets, but also reduces the number of parameters compared with UNet++. In order to make the introduced bridge features play a greater role in the process of feature splicing, B-UNet model introduces an attention mechanism, so that the model can focus more on the features that contribute more to the segmentation, and the reference of residual module also enhances the stability of the model. On VOC2007 dataset, MIoU, MPA and Accuracy were used as evaluation indexes, Experimental results show that B-UNet model is more suitable for multi-class segmentation on larger datasets than UNet model.