Semantic Segmentation of Remote Sensing Image Based on Convolutional Neural Network and Mask Generation

Figure 5

The selected images of the Potsdam dataset. The dataset is composed of high-resolution remote sensing images. The surface targets in the image are divided into 6 categories. (a) Low vegetation, including green plants such as grassland, woodland, and agricultural land. (b) Buildings, all types of buildings belong to this category. (c) Impervious surfaces, including road. (d) Cars, all types of cars belong to for this category. (e) Trees, all types of trees belong to this category. (d) Others, it means the clutter background. The semantic information annotation table is shown in Table 1, where R, G, and B, respectively, represent red, green, and blue.

(a)

(b)

(c)

(d)

(e)