Graph Neural Network-Based Method of Spatiotemporal Land Cover Mapping Using Satellite Imagery

Abstract

Multispectral satellite imagery offers a new perspective for spatial modelling, change detection and land cover classification. The increased demand for accurate classification of geographically diverse regions led to advances in object-based methods. A novel spatiotemporal method is presented for object-based land cover classification of satellite imagery using a Graph Neural Network. This paper introduces innovative representation of sequential satellite images as a directed graph by connecting segmented land region through time. The method’s novel modular node classification pipeline utilises the Convolutional Neural Network as a multispectral image feature extraction network, and the Graph Neural Network as a node classification model. To evaluate the performance of the proposed method, we utilised EfficientNetV2-S for feature extraction and the GraphSAGE algorithm with Long Short-Term Memory aggregation for node classification. This innovative application on Sentinel-2 L2A imagery produced complete 4-year intermonthly land cover classification maps for two regions: Graz in Austria, and the region of Portorož, Izola and Koper in Slovenia. The regions were classified with Corine Land Cover classes. In the level 2 classification of the Graz region, the method outperformed the state-of-the-art UNet model, achieving an average F1-score of 0.841 and an accuracy of 0.831, as opposed to UNet’s 0.824 and 0.818, respectively. Similarly, the method demonstrated superior performance over UNet in both regions under the level 1 classification, which contains fewer classes. Individual classes have been classified with accuracies up to 99.17%.

A deep inverse convolutional neural network-based semantic classification method for land cover remote sensing images

“…11 , it can be seen that the proposed method obtained a PRI index close to 1 during segmentation performance testing, indicating good segmentation performance. The PRI index of the methods in reference 15 and reference 16 are both below 0.9. Through comparison, it can be seen that the proposed method has significantly better test results than these two methods.…”

“…According to experimental analysis, the proposed method exhibits lower average time consumption, and as the number of images increases, the time consumption of all methods also shows an increasing trend. Compared to the methods in references 15 , 16 , the proposed method has higher computational efficiency and performs well in semantic classification tasks of large-scale image data. Therefore, the proposed method can quickly and effectively process land cover remote sensing images, and has broad application prospects.…”

“…The experiment tested the image segmentation performance of different methods and calculated their respective PRI values. The segmentation performance of the proposed method was compared with the methods in references 15 , 16 . The test results of the image segmentation performance of different methods are shown in Fig.…”

“…A set of remote sensing images of land cover with different resolutions and sizes was selected as the input dataset, including different types of landforms such as urban areas, farmland, and forests. In the experiment, the proposed method, methods from references 15 , 16 were used for testing, and the experimental results are shown in Table 3 .…”

“…The model trained by this method in specific regions or scenarios may not have good generalization ability, and the classification effect on data from other regions or under different conditions may not be ideal. Reference 16 Kavran et al proposed an object based spatiotemporal method for satellite image classification using graph neural networks. Construct a directed graph by connecting segmented land regions, then use convolutional neural networks to extract features, and use graph neural networks for node classification.…”