sample

Citation Author(s):
WEI
TONGBIAO
Submitted by:
CAO Ligang
Last updated:
Tue, 10/11/2022 - 07:54
DOI:
10.21227/c009-a371
License:
67 Views
Categories:
0
0 ratings - Please login to submit your rating.

Abstract 

This data is a conversion of remote sensing data into a VOC2012 dataset.

Instructions: 

This paper is aimed at the improvement of the Deeplabv3+ model, aiming at the excessive number of convolutional layers of the residual network 101 in the Deeplabv3+ model and the shortcomings of feature fusion, the reduction of the number of layers and the fusion of the residual blocks are realized, the accuracy of water extraction is improved, and the river discontinuity is improved.

Comments

Although the improved algorithm in this paper has achieved good water segmentation results on this dataset, there are still many shortcomings, which will guide my future research direction and improve the research work. Firstly, the data used in this paper are landsat8 remote sensing data with a spatial resolution of 30 meters. There are many water systems in Dongting Lake, with a large area and many tributaries. It may be inaccurate to label some small rivers. The sample data is manually labeled, so the labeled samples may have certain errors, which will affect the final water body segmentation accuracy. Although the improved algorithm in this paper has achieved good segmentation, there are still some shortcomings. The next step will be to test the algorithm for other water data sets, and further optimize and improve the model.

Submitted by CAO Ligang on Mon, 10/10/2022 - 04:26