白背飞虱智能识别技术研究

发布时间：2018-04-18 04:08

本文选题：白背飞虱 + 不变矩　；参考：《南京农业大学学报》2016年03期

【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。当使用鐖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。
[Abstract]:[objective] in order to realize the automatic monitoring of rice white back planthopper, an automatic recognition method of white back planthopper based on the features of insect back image was proposed.[methods] 131 insect images of white back planthopper were collected in the field by using a self-made insect image acquisition device. The binary image of the collected insect image was obtained by color (blue component BX 130) threshold segmentation and filter processing.Then a binary image of the back area of a single insect and a gray image of the back area are extracted.By statistical analysis of the size of white back planthopper, a single insect image with obvious non-white back planthopper was removed, and the geometric morphology, color and texture of the insect were extracted by invariant moment and two-dimensional Fourier spectrum data.The 7 invariant moments and 1 脳 l0.Two dimensional Fourier spectrum features are combined as input variables to establish a support vector machine (SVM) based identification model for white-backed planthopper.[results] the size of the single white back planthopper collected by the automatic acquisition device in the field was 1 000 ~ 2 600 pixels.When the recognition model is established by using the Eigenvalue of the invariant moment and the eigenvalue of the 5 脳 5 frequency spectrum, the accuracy of the training set is 95.71, and the accuracy of the test set is 95.00.[conclusion] using invariant moment and two-dimensional Fourier spectrum to extract the geometry, color and texture features of white back planthopper, and establishing the recognition model of support vector machine, the automatic recognition of white back planthopper in the field can be realized.
【作者单位】：南京农业大学工学院/江苏省智能化农业装备重点实验室;
【基金】：江苏省科技厅前瞻性联合研究项目(BY2014095)
【分类号】：S435.112.3;S126

【相似文献】