Int J Performability Eng ›› 2020, Vol. 16 ›› Issue (5): 711-719.doi: 10.23940/ijpe.20.05.p4.711719

• Orginal Article • Previous Articles     Next Articles

Semantic Segmentation Method based on Super-Resolution

Dulei Zheng, You Fu, Hao Zhang, Minghao Gao,  and Jianzhi Yu*()   

  1. College of Computer Science and Engineering, Shandong University of Science and Technology, Qingdao, 266590, China
  • Submitted on ; Revised on ; Accepted on
  • Contact: Jianzhi Yu
  • Supported by:
    This research was funded by the National Key Research and Development Program of China, grant number 2017YFB0202002; Natural Science Foundation of Shandong Province, grant number ZR2018BF001; Shandong Science and Technology Development Project, grant number 2013GGX10118; National Virtual Simulation Experimental Teaching Center Development Fund for Coal Mine Safety Mining, grant number 20190102; and Shandong Key Research and Development Project, grant number 2019GGX101066.


Convolutional neural network is an important method to solve most computer vision problems nowadays. Although increasing the computing cost and the scale of model will make most tasks achieve satisfactory results, the difficulty of increasing the computing cost and high-quality data also limit the increase of model scale. In this paper, when using neural network to segment the remote sensing image, aiming at the problem that the classification effect of the internal pixels of the target is not ideal, a multi-scale fusion structure about the dimension of the feature map is proposed as the classifier module of the model. In order to further improve the performance of semantic segmentation model, inspired by the Generative adversarial nets, combined with super-resolution, generative semantic segmentation architecture is proposed. In order to verify the effect of the two methods, the kappa coefficient was selected as the evaluation to conduct the semantic segmentation experiment of the remote sensing image of seaculture. With little to no increase in the scale of the model, the classification ability of the model is improved, and the effect is compared intuitively from the segmentation image.

Key words: semantic segmentation, generative, single image super-resolution, remote sensing image