摘要
[目的]为了实现水稻白背飞虱的自动监测,提出一种利用昆虫背部图像特征的白背飞虱自动识别方法。[方法]用自制的野外昆虫图像采集装置,在野外环境下,采集131张白背飞虱昆虫图像,通过颜色（蓝色分量B=130）阈值分割、滤波处理后,获取所采集昆虫图像二值化图,然后提取出单个昆虫背部区域二值化图和背部区域灰度图。通过对白背飞虱的大小统计分析的方法,剔去明显非白背飞虱的单个昆虫图像,再运用不变矩和二维傅里叶频谱数据提取昆虫几何形态、颜色和纹理共88个特征值,将7个不变矩和l×l（l=1,2,…,9）二维傅里叶频谱特征进行组合后作为输入变量,建立基于支持向量机的白背飞虱识别模型。[结果]自动采集装置在野外环境下采集的单个白背飞虱大小为1 000~2 600个像素点。当使用1~5不变矩特征值和5×5频谱特征值建立识别模型,训练集准确率达到95.71%,测试集准确率达到95.00%。[结论]使用不变矩和二维傅里叶频谱提取白背飞虱的几何形态、颜色和纹理特征,并建立支持向量机的识别模型,可以实现田间白背飞虱的自动识别。
[Objectives]An identification method of Sogatella furcifera based on the image features of the insect back was proposed for automatic monitoring of the Sogatella furcifera. [Methods]One hundred and thirty-one images of the Sogatella furcifera were captured in the field by using a automatic acquisition device which was developed by ourselves. After filtering,these images were segmented according to the threshold value（ Blue component,B = 130） and the banalized images were obtained. Then the banalized and the gray images of back regional of the individual insects were extracted. After the insects which were obviously different from the Sogatella furcifera in the size were removed from the images by the statistics analysis of the size of Sogatella furcifera images,the moment invariant and 2-d Fourier spectrum data were used to extract eighty-eight characteristic parameters related to the geometry,color and texture of Sogatella furcifera back images. The seven parameters and the l×l（ l = 1,2,…,9） 2-d logarithm spectrum data which were extracted from the Sogatella furcifera back images were combined into a feature vector. Finally,based on the feature vector,a classification model of Sogatella furcifera was developed using the support vector technology. [Results]The size of individual Sogatella furcifera is 1 000 to 2 600 pixels which were captured in the field by using a automatic acquisition device. When the 1-5moment invariant and the 5× 5 logarithm spectrum data were used to develop classification model,the accuracy of training sets was 95.71%,and the accuracy of test sets was 95.00%. [Conclusions]The proposed method can be used to identify Sogatella furcifera automatically when the moment invariant and 2-d Fourier spectrum data were used to extract characteristic parameters related to the geometry,color and texture of Sogatella furcifera back images,and a classification model of Sogatella furcifera was developed using the support vector technology.
出处
《南京农业大学学报》
CAS
CSCD
北大核心
2016年第3期519-526,共8页
Journal of Nanjing Agricultural University
基金
江苏省科技厅前瞻性联合研究项目(BY2014095)
关键词
白背飞虱
不变矩
二维傅里叶频谱
识别
Sogatella furcifera
moment invariant
2-d Fourier spectrum
identification
