A multi-scale weakly supervised learning method with adaptive online noise correction for high-resolution change detection of built-up areas

Author: Yinxia Cao, Xin Huang, Qihao Weng | Paper link | Date: 2023

Dataset

download link in google drive or [Baiduyunpan] (link：https://pan.baidu.com/s/17iHRoWs8XOY_93s9s5IqZQ , code：mhiz).
BA dataset excluding test cities (Beijing, Shanghai, Xian, and Kunming)

Setup datasets:

• training dataset: E:/yinxcao/ZY3LC/datanew8bit/datalist_posneg_train_0.6.csv
• test dataset: E:/yinxcao/ZY3LC/datanew8bit/datalist_posneg_test_0.6.csv

Part1. Training for BA detection

• training a multi-scale classification network
• generating the multi-scale cam
• generating pseudo-labels with CRF and thresholding
• adaptive online noise correction for BA detection: 1) obtain correction time; 2) correct labels

python train_mitb1_0.6_cam_stride_tlcmulti4.py
python demo_cues_torch_lvwang_mitb1_cam_stride_tlc_multi4.py
python cam_to_ir_label_tlcmult4.py
python train_mitb1_0.6_cam_stride_tlcmulti4_RRM_adele.py
python ttest_mitb1_0.6_cam_stride_rrm_tlcmulti4_adele.py

Part2. Training for BA change detection

• predict BA results for each date

python predict_rrm_tlcmulti4_adele_wholeimg.py

1. generate pseudo labels at pixel, object, and pixel+object levels

demo_1116_gen_pix_beijing.m
demo_1117_gen_obj_diff_beijing.m
demo_1117_gen_change_cert.m

• clip sample

demo_1117_clipsample_imglab_bj.m
demo_1118_testsample_diffarea.m

2. training change detection models see directory BANetCD

• stats model parameters see package torchstat