Research on Segmentation and Extraction of Mountain City Trail Landscape Elements Based on Improved PSPnet

As an important part of livable life, the efficiency and accuracy of landscape data management and scene design of urban footpath have become the new requirements of urban renewal. In this paper, a landscape element segmentation method which can be used in street view images is proposed. This method is based on the PSPnet semantic segmentation model. Firstly, through the cascades of different dimension features, it preserves more details of Street view image on the basis of enhanced scene parsing. And then, a lightweight semantic segmentation model is constructed using the depth separable convolution module to make it more efficient. Finally, through experimental comparison, the average pixel accuracy and average intersection ratio of the proposed method for the sample segmentation of mountain city walk street scene are 82.49% and 77.87%, respectively, which are 10.14% and 13.42% higher than before the improvement. Furthermore, the segmentation results are better than other models commonly used. This method can effectively divide the landscape elements of urban footpaths, which is of great significance for the improvement of urban streetscape data and the promotion of municipal management.