基于SFNet-F地物识别技术的农业大棚信息提取

Agricultural greenhouse information extraction based on SFNet-F land feature recognition technology

农业大棚是现代设施农业的重要组成部分,准确识别和动态监测其分布为政府开展农业补贴核算和农业生产决策提供可靠科学依据。针对传统方法图像识别准确度不高的问题,本文提出了SFNet-F图像处理技术。通过收集不同类型、时相和区域的大棚数据集,把FixMatch半监督学习模块与SFNet结合,以提高样本库建立的效率和质量,并缩减成本,从而实现高精度半监督自适应分割。为了评估该方法的可行性,以河北省平泉市为研究区域,选取多个精度评价指标进行精度验证,并与U-Net、HBRNet和DeepLabV3+进行对比。结果表明,基于SFNet-F嵌入SuperMap平台深入学习模型可以大范围、快速、精准识别农业大棚,识别效果相较于传统方法,各项精度指标均为最优。

Agricultural greenhouses are an essential component of modern agricultural facilities,and accurate identification and dynamic monitoring of their distribution provide a reliable scientific basis for the government to carry out agricultural subsidy accounting and agricultural production decision-making.In this paper SFNet-F image processing technology is proposed to address the issue of low accuracy in traditional image recognition methods.By collecting agricultural greenhouse datasets of different types,periods,and regions,the FixMatch semi-supervised learning module is combined with SFNet to improve the efficiency and quality of sample library establishment,reduce costs,and achieve high-precision semi-supervised adaptive segmentation.In order to evaluate the feasibility of this method,multiple accuracy evaluation indicators were selected for accuracy validation in Pingquan city,Hebei province,and compared with U-Net,HBRNet,and DeepLabV3+.The results show that the deep learning model based on the SFNet-F embedded SuperMap platform can identify agricultural greenhouses on a large scale quickly and accurately.The recognition effect is the best in all accuracy indicators compared to several popular methods.