Abstract: Farmland shelterbelts are the direct ecological barrier of agroecosystems. The width of shelterbelts directly affects the fulfillment of the protective function and subsequent management. In the continued construction of farmland shelterbelts, it is critically needed to identify the width at the regional scale quickly and accurately. In this study, Fuqiang Town, Dazhong Town and Xingnong Town of Baiquan County, the core area of black land in Heilongjiang Province, were selected as the research area. The Gaofen6 Satellite Imagery (GF-6) data were selected to extract the patch information of farmland shelterbelt by object-oriented and buffer segmentation technology, and each shelterbelt was extracted and its width was calculated. The results showed that: (1) Farmland shelterbelt was sensitive in the blue band, near-infrared band, and normalized difference vegetation index, which can be used to separate farmland shelterbelt and non-farmland shelterbelt. (2) The automatic recognition results of the width of the farmland shelterbelts based on object-oriented and buffer segmentation technology showed a strong correlation with the measured width, with an R₂ value of 0.73. (3) The width of farmland shelterbelts in Fuqiang Town, Dazhong Town, and Xingnong Town was basically the same, mainly concentrated on 17-24 m, accounting for 32.64%, 33.24% and 30.29% of the total area of farmland shelterbelts in each township, respectively. The object-oriented and buffer segmentation technology created here will provide method support for the subsequent construction and management of farmland shelterbelts.

Key words: remote sensing, object-oriented, decision tree, buffer segmentation, equivalent rectangle