摘要
为了寻求快速有效的毛豆内部豆荚螟的检测方法,将高光谱图像技术应用于毛豆内部的豆荚螟无损检测。以225个样本为研究对象,首先采用平均灰度值的方法自动获取毛豆感兴趣区域,然后提取400-1000nm波长范围内共94个波段的能量信息作为特征参数,最后结合支持向量数据描述分类器建立豆荚螟的分类检测模型。研究结果显示,在自动提取的感兴趣区域验证集中,正常样本的分类精度为100%,有虫样本分类精度为75%,验证集的总体分类精度为95.6%,可有效识别出含豆荚螟的毛豆样本。
In order to seek a quick and efficient detection method of edamame,hyperspectral imaging technique was applied to the nondestructive detection of insect-damaged edamame in this study. It was well known that the ROI of the vegetable soybean pods is the position of the beans,A ROI selection approach based on the mean gray values in the horizontal coordinate and vertical coordinate was proposed. In this experiment, hyperspectral transmission images were acquired from normal and insect-damaged vegetable soybeans （225 beans） ,These beans were used as the research samples. First,a region of interest（R-l） of edamame was extracted automatically using the mean gray value method from hyperspectral images. Then,the image power of ROl was extracted as classification feature,which the spectral region covered 400N1000nm and contained 94 wavelengths. At last,support vector data description （SVDD） was used to develop the classification models for the insect-damaged edamame. In the validation set,the results indicated the automatic extracting RQI method based on the mean gray value achieved 100% accuracy for the normal samples,75% accuracy for the insect-damaged samples,and 95.6% overall classification accuracy,which could discriminate insect- damaged edamame.
出处
《食品工业科技》
CAS
CSCD
北大核心
2014年第14期59-63,共5页
Science and Technology of Food Industry
基金
国家自然科学基金项目(61271384
61275155)
国家质检总局科技计划项目(2013KJ58)
江苏省自然科学基金项目(BK2011148)
中国博士后基金项目(2011M500851)
关键词
毛豆
感兴趣区域
能量
支持向量数据描述
edamame
a region of interest
power
support vector data description
