Shortest path problem Wu Xing Shortest path problem

最短路径 Shortest path problem Wu. Xing

Shortest path problem What is Shortest Path Part Why to Solve it How to Solve it

D C A Edge 1 Intersections -> node B road -> edge Edge 2 Edge 3

Shortest Path Problem New Road? Road Closure?

Shortest Path Problem How to Extract the Road From Our Real Life

Shortest Path Problem New method D-Link Net: Link Net with Pretrained Encoder and Dilated Convolution for High-resolution Satellite Imagery Road Extraction Based on Link Net, we can better update the map on the software by way of road extraction of satellite remote sensing images.

Shortest Path Problem Link Net D—Link Net

Shortest Path Problem

Shortest Path Problem Convolution Receptive field（感受野） Output: �(n+2 p−f)/s+1�×�(n+2 p−f)/s+1�

Shortest Path Problem

Shortest Path Problem

Shortest Path Problem Pooling • The function of the pooling layer is to reduce the size of the model, improve the computing speed, enlarge the receptive field and reduce the noise to improve the robustness of the extracted features • Max Pooling:

Shortest Path Problem

Shortest Path Problem Res-Blocks Plain-Network Residual-Network

Shortest Path Problem Significance： Simply increasing the depth can lead to gradient dispersion(弥散) or gradient explosion and degradation problems. The layer is represented as a learning residual function based on the input. Experiments show that the residual network is easier to optimize and can improve the accuracy by Instead of learning the entire output, adding considerable depth. the residual focus on the residual of the previous network output!

Shortest Path Problem Part A is the Encoder part，and now Part C is the Decoder one.

Shortest Path Problem Transposed convolution

Shortest Path Problem

Shortest Path Problem

Shortest Path Problem D—Link Net

Shortest Path Problem

Shortest Path Problem ACKNOWLEDGEMENT （1）http: //openaccess. thecvf. com/content_cvpr_2018_workshops/w 4/html/Zhou_D-Link. Net_Link Net_With_CVPR_2018_paper. html （论文） （2）https: //github. com/zlkanata/Deep. Globe-Road-Extraction-Challenge （开源项目） （3）https: //cloud. tencent. com/developer/news/360857 （CNN提取特征） （4）https: //blog. csdn. net/cxmscb/article/details/71023576 （CNN及实现） （5）https: //www. cnblogs. com/skyfsm/p/6790245. html （CNN总结） （6）http: //www. cnblogs. com/alanma/p/6877166. html （Reset-net笔记） （7）https: //blog. csdn. net/qq_38906523/article/details/79838000 （Encoder-Decoder模型） （8）https: //www. cnblogs. com/fourmi/p/10040619. html （Link Net笔记） （9）https: //www. cnblogs. com/yangperasd/p/7071657. html （Convolution network及其变种） （10）https: //www. jianshu. com/p/2 b 968 e 7 a 1715 （Receptive field） （11）https: //zhuanlan. zhihu. com/p/39866557 （空洞卷积/膨胀卷积） REFERENCE D-Link Net: Link Net with Pretrained Encoder and Dilated Convolution for High Resolution Satellite Imagery Road Extraction Lichen Zhou, Chuang Zhang, Ming Wu Beijing University of Posts and Telecommunications

Thank You!