Land-Cover Classification with High-Resolution
Remote Sensing Images Using Transferable Deep Models

Xin-Yi Tong, Gui-Song Xia, Qikai Lu, Huanfeng Shen,
Shengyang Li, Shucheng You, Liangpei Zhang.

1. Abstract

In recent years, large amount of high spatial-resolution remote sensing (HRRS) images are available for land-cover mapping. However, due to the complex information brought by the increased spatial resolution and the data disturbances caused by different conditions of image acquisition, it is often difficult to find an efficient method for achieving accurate land-cover classification with high-resolution and heterogeneous remote sensing images. In this paper, we propose a scheme to apply deep model obtained from labeled land-cover dataset to classify unlabeled HRRS images. The main idea is to rely on deep neural networks for presenting the contextual information contained in different types of land-covers and propose a pseudo-labeling and sample selection scheme for improving the transferability of deep models. More precisely, a deep Convolutional Neural Networks (CNNs) is first pre-trained with a well-annotated land-cover dataset, referred to as the source data. Then, given a target image with no labels, the pre-trained CNN model is utilized to classify the image in a patch-wise manner. The patches with high confidence are assigned with pseudo-labels and employed as the queries to retrieve related samples from the source data. The pseudo-labels confirmed with the retrieved results are regarded as supervised information for fine-tuning the pre-trained deep model. To obtain a pixel-wise land-cover classification with the target image, we rely on the fine-tuned CNN and develop a hybrid classification by combining patch-wise classification and hierarchical segmentation. In addition, we create a large-scale land-cover dataset containing $150$ Gaofen-2 satellite images for CNN pre-training. Experiments on multi-source HRRS images, including Gaofen-2, Gaofen-1, Jilin-1, Ziyuan-3, Sentinel-2A, and Google Earth platform data, show encouraging results and demonstrate the applicability of the proposed scheme to land-cover classification with multi-source HRRS images.


2. GID: A Well-Annotated Gaofen Image Dataset (GID) for Land-Cover Classification

For evaluating and learning transferable models for land-cover classification, we annotated the Gaofen Image Dataset (GID). It contains 150 high-resolution Gaofen-2 (GF-2) images acquired from more than 60 different cities in China. And these images cover the geographic areas that exceed 50,000 km2. Images in GID have high intra-class diversity coupled with low inter-class separability. Therefore, GID can provide the research community with a high-quality data resource to advance the state-of-the-art in land-cover classification.

- Data Source

Gaofen-2 (GF-2) is the second satellite of “High-definition Earth Observation System (HDEOS)” promoted by China National Space Administration (CNSA). Two panchromatic and multispectral (PMS) sensors with spatial resolution of 1 m panchromatic (pan)/4 m multispectral (MS) are onboard the GF-2 satellite, with a combined swath of 45 km. GF-2 images achieve a combination of high spatial resolution and wide field of view, allowing the observation of detailed information over large areas.

- Categories

GID consists of two parts: a large-scale classification set and a fine land-cover classification set. The large-scale classification set contains 150 pixel-level annotated GF-2 images, and the fine classification set is composed of 30,000 multi-scale image patches coupled with 10 pixel-level annotated GF-2 images. In the large-scale classification set, 5 major categories are annotated: built-up, farmland, forest, meadow, and water. The fine land-cover classification set is made up of 15 sub-categories: paddy field, irrigated land, dry cropland, garden land, arbor forest, shrub land, natural meadow, artificial meadow, industrial land, urban residential, rural residential, traffic land, river, lake, and pond. Its training set contains 2,000 patches per class, and validation images are labeled in pixel level.

- Annotated Samples

The large-scale classification set (training images and validation images are respectively marked with orange and cyan in the distribution map):

The fine land-cover classification set:

- Data Download

Once you register on the data download page, you can get the download link. (Data will be updated in September)


3. Experiments on Gaofen-2 images

- Comparison Methods

Features: spectral feature, gray-level co-occurrence matrix (GLCM), differential morphological profiles (DMP), and local binary patterns (LBP), and their fusion features.
Classifiers: Maximum likelihood classification (MLC), random forest (RF) , support vector machine (SVM), and multi-layer perceptron (MLP).

- Evaluation Metrics

Kappa coefficient (Kappa), overall accuracy (OA), and user’s accuracy.

- Classification Accuracy

- Examples of Classification Results


4. Testing of the Transferability of the Proposed Method

- Multi-source Images

- Overall Accuracy (%)

- Examples of Classification Results


Citation

If you want to make use of GID, please cite our following paper:

@article{Tong_2019_GID,
  title={Land-Cover Classification with High-Resolution Remote Sensing Images Using Transferable Deep Models},
  author={Tong, Xin-Yi and Xia, Gui-Song and Lu, Qikai and Shen, Huanfeng and Li, Shengyang and You, Shucheng and Zhang, Liangpei},
  journal={Remote Sensing of Environment},
  year={2019}
  }
	

Contact

If you have any problem, please contact:

  • Xin-Yi Tong at xinyi.tong@whu.edu.cn.
  • Gui-Song Xia at guisong.xia@whu.edu.cn.