C2F-SemiCD: A Coarse-to-Fine Semi-Supervised Change Detection Method Based on Consistency Regularization in High-Resolution Remote Sensing Images

22 Apr 2024  ยท  Chengxi Han, Chen Wu, Meiqi Hu, Jiepan Li, Hongruixuan Chen ยท

A high-precision feature extraction model is crucial for change detection (CD). In the past, many deep learning-based supervised CD methods learned to recognize change feature patterns from a large number of labelled bi-temporal images, whereas labelling bi-temporal remote sensing images is very expensive and often time-consuming; therefore, we propose a coarse-to-fine semi-supervised CD method based on consistency regularization (C2F-SemiCD), which includes a coarse-to-fine CD network with a multiscale attention mechanism (C2FNet) and a semi-supervised update method. Among them, the C2FNet network gradually completes the extraction of change features from coarse-grained to fine-grained through multiscale feature fusion, channel attention mechanism, spatial attention mechanism, global context module, feature refine module, initial aggregation module, and final aggregation module. The semi-supervised update method uses the mean teacher method. The parameters of the student model are updated to the parameters of the teacher Model by using the exponential moving average (EMA) method. Through extensive experiments on three datasets and meticulous ablation studies, including crossover experiments across datasets, we verify the significant effectiveness and efficiency of the proposed C2F-SemiCD method. The code will be open at: https://github.com/ChengxiHAN/C2F-SemiCDand-C2FNet.

PDF Abstract
Task Dataset Model Metric Name Metric Value Global Rank Benchmark
Change Detection CDD Dataset (season-varying) C2FNet F1-Score 95.93 # 9
Precision 95.46 # 2
F1 95.93 # 2
Overall Accuracy 99.04 # 2
IoU 92.18 # 2
Recall 96.41 # 2
KC 95.39 # 1
Change Detection DSIFN-CD C2FNet F1 64.03 # 4
IoU 47.09 # 3
Overall Accuracy 86.19 # 3
Precision 57.45 # 1
Recall 72.31 # 3
KC 55.62 # 1
Change Detection GoogleGZ-CD C2FNet F1 86.86 # 4
Precision 85.46 # 2
Recall 88.31 # 1
Overal Accuracy 93.43 # 1
KC 82.48 # 1
IoU 76.77 # 1
Change Detection LEVIR+ C2FNet F1 79.15 # 3
Prcision 77.19 # 4
Recall 81.22 # 3
OA 98.26 # 3
KC 78.25 # 3
IoU 65.50 # 3
Change Detection LEVIR-CD C2FNet F1 91.83 # 10
IoU 93.69 # 1
Overall Accuracy 90.04 # 8
F1-score 99.18 # 1
Precision 84.89 # 6
Recall 91.40 # 2
Semi-supervised Change Detection LEVIR-CD - 10% labeled data C2F-SemiCD IoU 83.15 # 1
F1 90.80 # 1
Precision 92.44 # 1
Recall 89.22 # 1
OA 99.08 # 1
KC 90.31 # 1
Semi-supervised Change Detection LEVIR-CD - 20% labeled data C2F-SemiCD IoU 83.75 # 1
F1 91.16 # 1
Precision 93.26 # 1
Recall 89.15 # 1
OA 99.12 # 1
KC 90.69 # 1
Semi-supervised Change Detection LEVIR-CD - 40% labeled data C2F-SemiCD IoU 84.62 # 1
F1 91.67 # 1
Precision 93.41 # 1
Recall 89.99 # 1
OA 99.17 # 1
KC 91.23 # 1
Semi-supervised Change Detection LEVIR-CD - 5% labeled data C2F-SemiCD IoU 81.76 # 1
F1 89.97 # 1
Precision 91.45 # 1
Recall 88.53 # 1
OA 98.99 # 1
KC 89.44 # 1
Change Detection S2Looking C2FNet F1-Score 62.83 # 6
Precision 74.84 # 1
Recall 54.14 # 4
OA 99.22 # 1
KC 62.44 # 3
IoU 45.80 # 3
F1 62.83 # 2
Change Detection SYSU-CD C2FNet F1 77.97 # 2
Precision 75.44 # 4
Recall 80.67 # 1
OA 89.25 # 4
KC 70.87 # 3
IoU 63.89 # 3
Semi-supervised Change Detection WHU - 10% labeled data C2F-SemiCD IoU 76.33 # 2
F1 86.58 # 1
Precision 87.35 # 1
Recall 85.81 # 1
OA 98.94 # 1
KC 86.03 # 1
Semi-supervised Change Detection WHU - 20% labeled data C2F-SemiCD IoU 81.93 # 1
F1 90.07 # 1
Precision 91.83 # 1
Recall 88.36 # 1
OA 99.23 # 1
KC 89.66 # 1
Semi-supervised Change Detection WHU - 40% labeled data C2F-SemiCD IoU 86.97 # 1
F1 93.03 # 1
Precision 93.20 # 1
Recall 92.86 # 1
OA 99.45 # 1
KC 92.74 # 1
Semi-supervised Change Detection WHU - 5% labeled data C2F-SemiCD IoU 74.87 # 2
F1 85.63 # 1
Precision 86.51 # 1
Recall 84.77 # 1
OA 98.87 # 1
KC 85.04 # 1
Change Detection WHU-CD C2FNet F1 94.36 # 1
Overall Accuracy 99.56 # 1
Precision 96.57 # 1
Recall 92.26 # 1
KC 94.14 # 1
IoU 89.33 # 1

Methods