改进卷积神经网络的冬小麦提取方法
Winter Wheat Extraction Method Based on Improved Convolution Neural Network
崔兆韵 1崔焕淼 2宋德娟 3李瑶瑶4
作者信息
- 1. 山东省气象防灾减灾重点实验室,济南 250031;泰安市生态与农业气象中心,山东泰安 271000
- 2. 山东工程职业技术大学人工智能学院,济南 250200
- 3. 临沂市河东区农业农村发展服务中心,山东临沂 276000
- 4. 泰安市生态与农业气象中心,山东泰安 271000
- 折叠
摘要
为提高遥感影像冬小麦识别精度,通过对传统的卷积神经网络进行优化,设计实现一种改进的卷积神经网络Im-SegNet(Improved-SegNet).模型增加了冬小麦和非冬小麦概率向量差值信息,对于概率向量差值较小的像素进行了二次判断.以肥城市冬小麦生长期的77张GF-2(Gaofen2)影像作为实验数据,利用精度、准确率以及查全率3项指标对Im-SegNet模型提取效果进行验证评价.结果表明,提取精度为92.1％,准确率为91.8％,查全率为84.9％,3项指标均高于经典的SegNet模型提取结果,Im-SegNet模型可以有效改善遥感影像作物分类效果.
Abstract
In order to extract winter wheat from high-resolution images with high accuracy,the traditional convolutional neural network were improved,and the implemented neural network Im SegNet(Improved SegNet)is got.The model adds the difference information of the probability vectors of winter wheat and non-winter wheat,and makes a secondary judgment on the pixels with smaller difference of the probability vectors,which improves the extraction accuracy of the convolution neural network model.77 GF-2(Gaofen 2)images of winter wheat in Feicheng City,were collected and used as experimental data,and three indicators of accuracy,accuracy and recall aws used to verify and evaluate the extraction effect of Im-SegNet model.The accuracy of the extraction results of the Im-SegNet model is 92.1％,the accuracy rate is 91.8％,and the recall rate is 84.9％.The three indicators are higher than the extraction results of the classic SegNet model,indicating that the Im-SegNet model is more suitable for extracting the spatial distribution information of winter wheat from high-resolution images.
关键词
全卷积神经网络/遥感影像/贝叶斯/冬小麦Key words
full convolutional neural network/remote sensing image/bayesian/winter wheat引用本文复制引用
基金项目
科技创新2030-"新一代人工智能"重大项目(2022ZD0119500)
山东省气象局引导类项目(2021SDYD33)
泰安市科技创新发展项目(2020NS062)
出版年
2024
NETL
NSTL
维普
万方数据