文献类型： 会议论文

第一作者： Yang Ding

作者： Yang Ding ¹ ; Xin Zhao ¹ ; Dongyan Zhang ¹ ; Dong Liang ¹ ; Zhicun Wang ¹ ; Shangjing Xi ¹ ; Shizhou Du ² ;

作者机构： 1.National Engineering Research Center for Agro-Ecological Big Data Analysis & Application, Anhui University

2.Institute of Crops, Anhui Academy of Agricultural Sciences

关键词： Rice lodging;UAV;Texture features;SVM;Kappa coefficient

会议名称： IEEE International Geoscience and Remote Sensing Symposium

主办单位：

页码： 9228-9231

摘要： Lodging is one of the important factors causing the decrease of rice yield and quality. For obtaining the lodging area of rice accurately, evaluating the disaster situation correctly, and providing technical support for agricultural insurance and production decision-making, this paper used the consumption-grade unmanned aerial vehicle DJI Phantom 4 Advanced to acquire high-resolution visible images at 100 m flight altitude, to study the extraction of rice lodging features and area based on machine vision technology. Firstly, through calculating four texture features such as the correlation, energy, contrast and homogeneity, it was found that the difference between lodging and normal rice was the most obvious by the texture image of homogeneity. Furthermore, on the basis of homogeneity texture, the YCBCR transform was applied to enhance texture image, namely as the R1G1B1 image. We found that the separability of samples was as high as 1.9 for G1 Band from the R1G1B1 image. The G1 Band was chosen as the most effective feature to distinguish lodging area from rice field. On this basis, the texture feature T of G1 band was obtained by choosing the texture of the deficit moment and setting the size of the sub-window to 29 ×29_. Finally, the color feature, color feature plus texture feature T were used to classify by support vector machine (SVM), and the classification accuracy were 87.33% and 92.96%, respectively. The corresponding Kappa coefficients were respectively 0.658 and 0.817. The results showed that the combination of texture features can improve the recognition accuracy of visible image for lodging rice. Meanwhile, compared with the traditional K-means and maximum likelihood method, the SVM-based classification was the best. The above results provide valuable reference for the extraction of other crops' lodging area based on optical sensor.

分类号： TP7-53