Research on Image Segmentation based on Multibranch Convolution Model

Image semantic segmentation is a very important research field in computer vision task. In view of the discontinuity of segmentation in existing semantic segmentation models, poor segmentation effect on small target objects and large amount of computation, a module similar to Iception is designed, which is called Multi-Branch Convolution (MBC) in this paper to decompose the convolution layer of the atrous space pyramid pooling (ASPP). The experimental results show that the improved network model achieves segmentation results of 71.3%(MIoU) and 90.9%(FWIoU).