Extraction Spatial Distribution of Winter Wheat Based on Deep Learning
Accurate spatial distribution data of winter wheat is of significant importance for government departments to guide agricultural production and optimize industrial structure.This study addresses the need to obtain spatial distribution data of winter wheat from high-resolution remote sensing images by establishing an efficient semantic segmentation model based on convolutional neural networks(Dense-U-Coord Net)to acquire high-precision spatial distribution information of winter wheat.The Dense-U-Coord Net model uses DenseNet as its backbone to create a"U"-shaped network structure,which facilitates the fusion of features at different levels through dense connections.It also establishes a multi-dimensional feature attention mechanism based on coordinate attention and OCR modules(object-contextual representations),which optimizes the fused features based on contextual and spatial location information to enhance the model's ability to generate consistent features.Dense-U-Coord Net employs Softmax as the classifier for image segmentation to extract spatial distribution data of winter wheat.The study area selected is Guantao County,Handan City,Hebei Province,using GF-6 PMS(Gaofen-6 Panchromatic and Multispectral Scanner)remote sensing images as the data source.SVM,U-Net,ERFNet,and RefineNet models are chosen as comparative models for comparative experiments.The experimental results demonstrate that the Dense-U-Coord Net model outperforms the comparative models with Precision(92.5%),Recall(93.4%),Mean Pixel Accuracy(MPA)(94.2%),and Mean Intersection over Union(MIoU)(91.7%).These findings confirm the advantages of Dense-U-Coord Net in extracting spatial distribution information of winter wheat.The proposed method provides essential data for modern agriculture.
国家科技期刊平台
NSTL
万方数据
