基于改进U-Net模型的农业大棚遥感提取方法

Remote sensing extraction method of agricultural greenhouse based on an improved U-Net model

农业大棚是一种农业设施,依据表面透光度分为阳棚和阴棚。大范围统计农业大棚对农业设施普查、农业政策制定、县域经济发展规划具有重要意义。针对人工统计费时费力问题,基于卷积神经网络和高分辨率遥感影像实现农业大棚遥感信息提取。针对U-Net模型对遥感影像中语义信息提取不足和边缘信息利用不充分问题,提出以下改进方案:1)优化语义分割任务,使用ConvNeXt和注意力机制提取遥感影像中农业大棚的深层语义信息;2)引入边缘检测任务,采用门控卷积层和拼接操作融合编码器的语义特征和解码器输出的图像梯度,进而结合边缘信息优化分割结果。经测试,改进后的模型能同时提取阴阳2种大棚信息且识别效果良好,相对传统方法有较大提高。

The agricultural greenhouse is a kind of agricultural facility,which is divided into transparent and non-transparent according to the surface transmittance.The large-scale statistics of agricultural greenhouses are of great significance to the survey of agricultural facilities,the formulation of agricultural policies,and the planning of county economic development.Aiming at the problem that manual statistics are time-consuming and laborious,this paper utilizes the convolutional neural network to extract agricultural greenhouses information from high-resolution remote sensing images.To solve the problems of insufficient semantic information extraction in remote sensing images and insufficient utilization of edge information of the U-Net model,this paper proposes the following improvements:1)The semantic segmentation task is optimized,and ConvNeXt and attention mechanism is utilized to extract deep semantic information of agricultural greenhouses in remote sensing images.2)The edge detection task is introduced,and the gated convolution layer and concate operation are used to fuse the semantic features of the encoder and the image gradient output by the decoder,and then the edge information is combined to optimize the segmentation results.After testing,the improved model can extract both transparent and non-transparent agricultural greenhouses information at the same time and the recognition effect is good,which is greatly improved compared with the traditional method.