摘要
针对空间域特征不能全面准确地描述叶片的问题,提出了一种基于复频域纹理特征(Complex frequency domain texture features,CFDTF)的叶片识别算法。首先,对叶片图像进行预处理。其次,对预处理后的图像进行分块,并对每一个图像块进行双树复小波变换(Dual-tree complex wavelet transform,DTCWT),分别计算复频域局部二值模式(Local binary pattern,LBP)和局部相位量化(Local phase quantization,LPQ)特征,得到图像块的特征。接着,串联所有图像块的特征得到整个图像的特征。最后,在Flavia数据库上通过KNN分类器分类识别。结果表明,与传统的颜色、形状、纹理等特征相比,该算法平均识别精度明显提高,达到95.75%。
In the most time,the traditional spatial domain feature cannot describe a leaf completely and accurately.In this study,an algorithm of leaf recognition based on complex frequency domain texture features(CFDTF)was proposed.Firstly,the image of leaf should be preprocessed.Secondly,the preprocessed image could be divided into many blocks,and every image block was processed by the dual-tree complex wavelet transform(DTCWT).Next,the local binary pattern(LBP)and local phase quantization(LPQ)features of image blocks were calculated respectively,and the features of image blocks were obtained.Thirdly,all block features were concatenated to obtain the feature of an entire image.Finally,KNN classifier was used for classification and recongnition in the Flavia dataset.Compared with the traditional color,shape,texture and other features,the average recognition accuracy of our algorithm was significantly improved,reaching 95.75%.
作者
梅星宇
李新华
鲍文霞
张东彦
梁栋
MEI Xing-yu;LI Xin-hua;BAO Wen-xia;ZHANG Dong-yan;LIANG Dong(National Engineering Research Center for Agro-Ecological Big Data Analysis and Application,Anhui University,Hefei 230601,China)
出处
《江苏农业学报》
CSCD
北大核心
2019年第6期1334-1339,共6页
Jiangsu Journal of Agricultural Sciences
基金
国家自然科学基金项目(41771463、61672032)
关键词
植物叶片识别
复频域纹理特征
双树复小波变换
plant leaf recognition
complex frequency domain texture features
dual-tree complex wavelet transform
