关键词: 城市绿地, Ostu算法, MeanShit算法, FNEA算法, DeepLab-v3+

Abstract: Urban green space is an important part of ecological civilization construction, and the extraction of green space information is the basis and premise of urban green space planning and construction. Remote sensing image segmentation is the key step of green space information classification. Selecting appropriate image segmentation method can effectively improve the accuracy of urban green space extraction. The traditional remote sensing segmentation method results in a serious edge jagged phenomenon, which is quite different from the green field boundary, and does not meet the requirements of green space information extraction. In this paper, the high-resolution Worldview image is used as the data source, and the urban green space information is extracted based on the deep learning network DeepLab-v3+. At the same time, the classification accuracy of the proposed algorithm is compared with that of Ostu, MeanShift and FNEA. The results show that: DeepLab-v3 + has the best segmentation performance, its segmentation edge is smooth, and it has a high degree of coincidence with the green field boundary, which can effectively solve the edge jagged problem of traditional segmentation algorithm. Because of the good segmentation effect, it also effectively improves the classification accuracy. Among various segmentation and classification algorithms, the accuracy of DeepLab-v3 + is the highest, reaching 98.01%.

Key words: urban green space, Ostu algorithm, MeanShift algorithm, FNEA algorithm, DeepLab-v3+