Abstract: [Objective] This paper was to explore a high precision automatic extraction method of farmland shelterbelt in northeast China based on analyzing its spectral and spatial geometric characteristics. And the results will provide basic data support for a wide range of farmland shelterbelt extraction and remote sensing monitoring.[Method] In this paper, part zones of Dehui City and Nong'an County of Jilin Province were took as the study area. We analyzed the vegetation index and spatial geometric features of the farmland shelterbelt based on ZY-3 multi-spectral image. The residential boundary was extracted from Landsat 8 OLI data. Then we put forward using the object-oriented method to deal with binary image data. The vector results of farmland shelterbelt were extracted in combination with the mathematical morphology and the GIS technology.[Result] The total length of farmland shelterbelt is 304.46 km within the 50 km×50 km study area. The correct extraction of farmland shelterbelt is 286.42 km, the excess extraction of 18.05 km and missing extraction is 14.19 km. In this study, we used the region verification, filed verification and high resolution images verification based on existing outcome data, filed observation data, GeoEye image and ZY-3 image. As for reqion verification, the extraction accuracy is 89.89%, the redundancy error is 5.66% and the missing error is 4.45%. All 22 belts collected in field were extracted correctly and the extraction accuracy of length is 93.93%.[Conclusion] The ratio vegetation index(RVI) is better than the normalized difference vegetation index(NDVI) when extracting the farmland shelterbelts in high vegetation coverage. Mathematical morphology method and object-oriented method have their unique advantages in processing linear characteristic features which has a certain gap, especially for the farmland shelterbelt. It should be given full consideration to the phenology information, spectral information and spatial geometry information of farmland shelterbelt when extracting farmland shelterbelt automatically with remote sensing technology. Accuracy verification results show that the combination of morphology, object-oriented methods and GIS technology to extract farmland shelterbelts can obtain higher accuracy based on ZY-3 image. This method can give a reference for extracting the farmland shelterbelt automatically and widely in northeast China. It also can provide technical support for the spatial analysis of landscape and the dynamic monitoring and management in the future.

Key words: farmland shelterbelt, remote sensing, feature extraction, morphology, object-oriented