[Objective] The extraction and change monitoring of cultivated land information is one of the hotspots of remote sensing application research. Northern Guangdong is the main grain base of Guangdong Province, and is the key area of cultivated land change monitoring by business management departments.[Method] In the present paper, using GF-2 and Sentinel-2 remote sensing images, combined with the phenological period of late rice in the study area, the vegetation index, humidity index, brightness index, color index and texture features were extracted for multi-scale segmentation, and the multi-scale image of paddy field extraction was constructed. The index difference image and time series multi feature difference image were classified to extract the paddy field area and evaluate the results.[Result]（i) The highest extraction accuracy of paddy field in time series multi feature difference image was 0.98,and the minimum difference between paddy field area and late rice sowing area was 240.05 hm2,and the best paddy field extraction effect was based on time series multi feature difference image;（ii）The difference of index feature needed to use the key phenological characteristics of crops and combine with multi-scale segmentation of image to improve the accuracy of paddy field extraction;（iii）The machine forest method has faster operation speed and higher classification accuracy in data classification with high dimensional features. [Conclusion] The research of the paper is to explore the method of high-resolution remote sensing in rapid and accurate updating of cultivated land information, which can provide support for the application of cultivated land management business. [ABSTRACT FROM AUTHOR]