摘要
在小麦不完善粒识别中,高光谱图像的光谱特征信息与高分辨率图像的空间结构信息对不同类别小麦的识别有着各自的优势。单纯地利用一种图像源进行小麦识别,无法解决单种数据源的信息局限性。首先通过将高光谱图像进行波段选择和分段主成分分析(PCA)数据降维,然后与高分辨率图像进行配准融合,用新的融合图像作为数据源来进行小麦分类识别。最后新的数据源在结合特征金字塔改进的VGG卷积网络识别算法中,平均识别率相较于高光谱图像和高分辨率图像分别提高6.08%以及3.34%。新数据源有效地融合两种信息源识别小麦的优势,提升识别准确率,进一步推进小麦不完善粒检测技术的发展。
In the recognition of imperfect grain of wheat,the spectral feature information of hyperspectral image and the spatial structure information of high-resolution image have their own advantages for the identification of different types of wheat.Simply using an image source for wheat identification does not address the information limitations of a single data source.The hyperspectral image is band-selected and the data is dimension-reduced by segmented Principal Component Analysis(PCA).Then the high-resolution image is used for registration and fusion,and the new fused image is used as the data source for wheat classification and recognition.Finally,the new data source in the improved VGG convolution network identification algorithm combined with feature pyramid,the average recognition rate is increased by 6.08%and 3.34%compared with hyperspectral image and high resolution image respectively.The proposed method effectively combines the advantag⁃es of two kinds of information sources on wheat classification,and improves the recognition accuracy.It is of great significance for the detec⁃tion and identification of imperfect grain of wheat intelligence.
作者
郝传铭
卿粼波
何小海
李晓亮
HAO Chuan-ming;QING Lin-bo;HE Xiao-hai;LI Xiao-liang(College of Electronic Information,Sichuan University,Chengdu 610065;Chengdu Institute of Grain Storage Science,Chengdu 610091)
出处
《现代计算机》
2019年第36期44-48,共5页
Modern Computer
关键词
高光谱图像融合
主成分分析(PCA)
卷积网络
小麦不完善粒识别
Hyperspectral Image Fusion
Principal Component Analysis(PCA)
Convolutional Network
Recognition of Imperfect Grain of Wheat
