针对不同季节影像水稻田提取适用性不高的问题，借鉴卫星影像智能解译方法，提出了基于高分辨率无人机影像和顾及季节变化的水稻田提取方法。以无人机影像为数据源，分析了水稻不同生长时期的影像特征，并建立水稻田季节性特征样本库；结合卷积注意力机制、空洞空间金字塔池化、残差结构改进U-Net网络模型，以提高在复杂背景中获取目标地物重要特征和上下文信息的能力；利用季节性特征样本库进行模型训练验证，以提高模型在不同季节的适用性。对比分析结果表明，该方法能从不同季节的高分辨无人机影像中准确提取水稻田，各季节提取结果的平均召回率、平均交并比分别达到95.54%、91.20%以上，优于原始的U-Net、DeepLabV3等模型。

In response to the challenges posed by the varying phenological characteristics of rice paddy during different growth stages and the limited sample size, we put forward an extraction method of rice paddy based on high-resolution UAV images and accounting for seasonal variations. Taking UAV images as the data source, we analyzed the image characteristics of rice paddies at different growth stages, and established their seasonal characteristic sample library. Combined with the convolutional block attention module, the atrous spatial pyramid pooling and the residual block, we improved the U-Net model to enhance its ability to obtain important features and contextual information of target objects in complex backgrounds. We used the seasonal characteristic sample library for model training and validation to improve the feasibility of model in different seasons. Experimental comparative analysis demonstrates that this method can precisely extract rice paddies from high-resolution UAV images across various seasons. In each season, the improved U-Net model achieved the average recall and average IoU exceeding 95.54% and91.20%, respectively, surpassing the original U-Net, DeepLabV3, and other models.