传统变化检测手段进行水稻遥感识别受“云污染”和影像间配准误差导致的变化检测误差累积及“椒盐”现象的影响,水稻遥感识别精度低。本文提出时空融合模型(Temporal-Spatial-Fusion Model,TSFM)进行水稻遥感识别,旨在综合像元在时间、空间维度上的信息定义像元的水稻时空归属度,根据时空归属度划分阈值提取水稻。实验结果表明:在不同窗口尺度下,TSFM在整体和“云污染”区域对水稻提取均达到了较高精度。当窗口尺度为3×3时,水稻提取的用户精度、制图精度和总体精度分别达到93.4%、83.5%和87.9%。在不同窗口尺度下水稻提取的用户精度、制图精度、总体精度均高于分类后比较PCC(Post-Classification Comparison)和多数投票法(Majority Voting,MV);在“云污染”区域,水稻识别总体精度均在92.0%以上,水稻制图精度比PCC、MV分别至少提高了14.0%、7.6%。有效地解决了传统变化检测作物遥感识别存在的误差累积问题,在一定程度上避免了“云污染”和“椒盐”现象对识别结果的影响。另外,初步探讨了TSFM水稻提取精度与景观特征关系,发现在景观规整区域适宜采用较小的窗口,在破碎区域适宜采用较大的窗口。该方法的成功实施,为大范围开展秋粮作物遥感识别,消除“云”影响进行了前期实验探讨。 Traditional remote sensing detection means that the accuracy of remote sensing of rice is low due to the influence of “detection error” and “salt and pepper” phenomenon caused by “cloud pollution” and registration error between images. In this paper, TSFM is used to identify rice remote sensing. The purpose of this paper is to define the spatial and temporal membership of paddy rice based on the information of pixels in time and space dimensions, and to extract the rice paddy according to the spatio-temporal membership degree threshold . The experimental results show that the TSFM achieves high precision for rice extraction in the whole and “cloud pollution” regions at different window scales. When the window size is 3 × 3, the accuracy of user extraction, mapping accuracy and overall accuracy of rice extraction are 93.4%, 83.5% and 87.9% respectively. The user accuracy, the precision of mapping and the overall accuracy of rice extraction under different window scales were higher than those of PCC (Post-Classification Comparison) and Majority Voting (MV) after classification. In the area of ​​“cloud pollution”, rice The overall accuracy of identification was over 92.0%, and the accuracy of rice mapping was increased by at least 14.0% and 7.6% respectively compared with PCC and MV. It effectively solves the problem of error accumulation in the traditional remote sensing detection of crop change and to some extent avoids the influence of “cloud pollution” and “salt and pepper” on the recognition result. In addition, the relationship between TSFM rice extraction accuracy and landscape characteristics was discussed preliminarily. It was found that a smaller window should be used in the landscape area and a larger window should be used in the area. The successful implementation of this method has carried on the pre-experiment discussion for the wide range crop remote sensing identification and eliminating the influence of “cloud”.