Deep Learning-Based Vegetation Extraction and Vegetation Change Monitoring by using Very High-Resolution Satellite Imagery

Vegetations play an important role in the management of physical activities and the public health of urban residents. However, with the rapid urbanization in the world, vegetation regions are changing constantly. In order to prevent the decrease in vegetation areas, constant vegetation monitoring is required. In this study, we performed vegetation extraction and vegetation change monitoring in very high-resolution (VHR) satellite imagery through deep learning-based techniques. To this end, two deep learning networks (i.e., DeepLabV3-plus, and deeply supervised image fusion network (DSIFN)) were used for vegetation extraction and change detection, respectively. Firstly, the two networks were trained on the two datasets each for their respective purpose. Then, a DSIFN was tested to detect all the changes occurring in VHR bitemporal satellite images. Moreover, the binary vegetation maps from bitemporal images were independently generated by using DeepLabv3-plus. Later, the vegetation maps and the change detection result were combined to figure out the change tendency related to vegetation. To show the effectiveness of the proposed method, an accuracy assessment was carried out. The proposed method can be used to determine the amount of change occurring within a period in the vegetation of urban areas.