首页 >  , Vol. , Issue () : 1993-2002

摘要

引用本文:

DOI:

10.11834/jrs.20208285

收稿日期:

2018-07-10

修改日期:

2019-03-04

PDF Free   EndNote   BibTeX
从高分影像中提取冬小麦空间分布信息的方法
宋德娟1, 张承明1, 杨晓霞1, 李峰2, 韩颖娟3, 高帅4, 董海燕1
1.山东农业大学信息科学与工程学院;2.山东省气候中心;3.中国气象局旱区特色农业气象灾害监测预警与风险管理重点实验室;4.中国科学院遥感与数字地球研究所
摘要:

高精度的农作物空间分布信息对于资源、环境、生态、气候变化和粮食安全问题具有重要的意义,使用全卷积神经网络从遥感影像中提取的农作物空间分布信息中,种植区域边缘往往比较粗糙。本文以冬小麦为研究对象,以国产Gaofen 2影像为数据源,通过将RefineNet模型与最大后验概率模型耦合,建立了冬小麦高精度遥感影像分割模型(Winter Wheat Remote Sensing Segmentation Model,WWRSSM),获取了精细的冬小麦空间分布信息。WWRSSM模型使用RefineNet网络的卷积结构提取像素的特征;使用改进的SoftMax模型逐像素计算类别概率向量,根据类别概率向量的分量差值将像素分为差值较大的像素集合和差值较小的像素集合,对于差值较大的像素集合,直接使用概率最大的类别作为相应像素的类别,对于差值较小的像素集合,进一步结合最大后验概率模型确定每个像素的类别。利用随机梯度法对模型进行训练,使用训练成功的模型从遥感影像上提取冬小麦的空间分布信息。选择SegNet、DeepLab、RefineNet作为对比模型进行实验,实验结果显示,WWRSSM结果的精度为92.9%,比SegNet提高了13.8%,比DeepLab提高了10.9%,比RefineNet提高了8.6%,表明WWRSSM模型在提取冬小麦空间分布信息方面具有一定的优势。研究结果对于提高大范围冬小麦空间分布制图精度和自动化水平具有一定的意义,也可以为农作物空间分布信息提取和面积统计提供技术参考。

Extracting winter wheat spatial distribution information from GF-2 images
Abstract:

Abstract: Objective:Winter wheat is one of the main food crops in China. The accurate spatial distribution information of winter wheat is of great significance for yield estimation and food security. However, the existing methods for extracting spatial distribution information of winter wheat using full convolutional neural network ignore the characteristics. The influence of the difference between the probabilities in the coded class probability vector on the judgment of the pixel class attribution leads to the occurrence of misclassification or missing points at the edge, which affects the accuracy of the result. In this paper, the RefineNet model is coupled with the Maximum aposteriori probability (MAP), and the WWRSSM (Winter Wheat Remote Sensing Extraction) model is established to form a method to extract the spatial distribution information of Winter Wheat from gf-2 (gaofen-2) Remote Sensing image. Method: WWRSSM uses the convolution structure of RefineNet network to extract the features of pixels. Using improved SOFTMAX model to get the pixel category probability vector, and with winter wheat category probability value with other category as a probability vector poor, according to the category probability vector difference between can be difference divided into smaller pixel and difference collection between the larger pixel collection, collection for larger difference between pixels,directly using the probability biggest category as the pixel category, for small difference between pixels, combined with the maximum a posteriori probability model to determine the type of each pixel; Then the model was trained by using the stochastic gradient method, and the spatial distribution information of winter wheat was extracted from the remote sensing image using the successfully trained model. Result:SegNet, DeepLab and RefineNet were selected as the comparison models. Experimental results showed that WWRSSM accuracy was improved by 4.2%, 7.6% and 8.6%, respectively, compared with the comparison models, and the overall extraction accuracy reached 93%, indicating that the method proposed in this paper has certain advantages in proposing the spatial information distribution of winter wheat. Conclusion: This method deeply explores the information contained in the class probability vector of the output of the full convolutional network, and finds that the classification error is closely related to the small component difference in the class probability vector. Based on this, the classification result of the full convolution network is revised. The effectiveness and feasibility of the proposed method are proved by experiments.

本文暂时没有被引用!

欢迎关注学报微信

遥感学报交流群 分享按钮