To accurately extract the growing area of paddy rice is a significant premise of paddy rice production and food security under the background of climate change.based on the current situation that paddy rice extraction is beset with difficulties in southern china,where clouds and rain appear in high frequency during growing seaon,how to take full advantage of the limited images to obtain accurate paddy rice planting area is a desiderated problem.In this study,we combined remotely sensed data from two different dates and brought out D-value bands,using object-oriented Random Forest to achieve the goal of rice extraction. The D-value bands,indicating the difference between a character derived from two different time phases, can be generated from traditional characteristic bands including vegetation indexes,water index,prominent component analysis and Tasseled Cap results,as well as the original bands.We applied this method to extract paddy rice planting area in Dingcheng District,Changde,Hunnan Province,China, and results show that,the accuracy of paddy rice extraction was improved to 93% by 3percent compared with single-phased method,and the kappa coefficient reaches 91in the study area.To further analyze the effect of D-value bands in other classifiers,we compared the accuracy of combination of D-value bands with decision tree and Random Forest,separately.Results show that the D-value bands provides infromation in both subject segementation and classification,which can effectively improve the accuracy of paddy rice planting area extraction.