20 December 2024 in IEEE Transactions on Geoscience and Remote Sensing
Change detection is an important task in earth observation. In the past few years, significant progress has been made in supervised change detection research. However, change labels are extremely expensive. Semi-supervised change detection has attracted increasing attention. In semi-supervised change detection, the problem of scarcity of positive samples is magnified. Moreover, the imbalance of change types (e.g., disappearance and appearance) exacerbates the missing detection phenomenon. To address the above problems, we propose a semi-supervised change detection method: CutMix-CD, which incorporates the change-aware CutMix augmentation into the consistency framework of change detection. The semi-supervised learning framework enriches change contexts and places special emphasis on the comparative process, facilitating more robust representations of changes with improved generalization capabilities. Firstly, mixed samples are synthesized using the change-aware CutMix operation. Then, we developed a student path and a teacher path to predict the changes of original samples and mixed samples respectively. Finally, the consistency loss is conducted between the two predictions to help the model learn change information of unlabeled samples. In addition, an unsupervised feature constraint loss is proposed to further optimize the change features. Experiments on four datasets validate the effectiveness of CutMix-CD. It can effectively alleviate the overfitting problem for unbalanced types of changes, and even outperforms the fully-supervised methods for some challenging samples. The code will be released in https://github.com/SQD1/CutMixCD.
19 November 2024 in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
Human-labeled training datasets are essential for convolutional neural networks (ConvNets) in satellite image scene classification. Annotation errors are unavoidable due to the complexity of satellite images. However, the distribution of real-world human-annotated label noises on satellite images and their impact on ConvNets have not been investigated. To fill this research gap, this study, for the first time, collected real-world labels from 32 participants and explored how their annotated label noise affects three representative ConvNets (VGG16, GoogleNet, and ResNet-50) for remote sensing image scene classification. We found that: (1) human-annotated label noise exhibits significant class and instance dependence; (2) an additional 1% of human-annotated label noise in training data leads to a 0.5% reduction in the overall accuracy of ConvNets classification; (3) the error pattern of ConvNet predictions was strongly correlated with that of participant's labels. To uncover the mechanism underlying the impact of human labeling errors on ConvNets, we compared it with three types of simulated label noise: uniform noise, class-dependent noise and instance-dependent noise. Our results show that the impact of human-annotated label noise on ConvNets significantly differs from all three types of simulated label noise, while both class dependence and instance dependence contribute to the impact of human-annotated label noise on ConvNets. Additionally, the label noise estimation algorithm (confident learning) cannot fully identify label noise. These observations necessitate a reevaluation of the handling of noisy labels, and we anticipate that our real-world label noise dataset would facilitate the future development and assessment of label-noise learning algorithms.
22 September 2026 in Scientific Data
Accurate mapping of human settlements is vital for understanding urbanization, human-environment interactions, and sustainable development. Existing medium-resolution products often underestimate the real spatial extent of settlements by focusing solely on built-up cores. Here, we present the China Settlement Footprint 1990–2020 (CSF30) dataset, a nationwide product depicting complete human settlement footprints from Landsat imagery at five-year intervals. CSF30 was generated using an Entity-Based Image Analysis framework integrated with a deep learning architecture. To ensure spatio-temporal generalizability and boundary accuracy, the model was trained via a two-stage strategy utilizing a built-up area dataset and a newly annotated settlement footprint dataset. Comprehensive validation demonstrates that CSF30 achieves a mean F1-score of 0.91 across the 30-year period, showing distinct advantages in representing rural and peri-urban areas compared to existing products. CSF30 provides a critical, high-quality baseline for diverse applications, including investigating settlement trajectories, assessing ecological and environmental impacts, and monitoring progress toward the Sustainable Development Goals (SDGs).