基于图像处理的花生荚果品种识别方法研究

为实现品种鉴定与真伪识别的自动化,基于图像识别的方法,采用扫描仪采集了20个品种,每个品种100颗花生果正面和2个侧面的图像,分别获取每幅图像的形态、颜色和纹理三大类共50个特征,并对这些特征进行主分量分析(PCA)优化,针对优化和没有优化的特征,搭建了人工神经网络识别模型和支持向量机模型,并采用两种模型进行品种识别,结果表明,采集的特征经PCA优化后表现出更强的识别性能,SVM较神经网络识别效果总体上得到提高,并且识别效果稳定。品种的数量对识别效果有影响,在通常情况下可根据品种的数量来确定特征的数量,可以进一步提高效率,对20个品种,需要选择超过15个特征。颜色类特征比形态类和纹理类特征具有更好的识别效果,经过不同类别的特征组合后,整体上识别性能达到90%以上,基本可以推广到实际生产中使用。

Study on variety identification of peanut pods based on image processing

In order to realize the automation of peanuts variety identification and recognition,based on image recognition method,we have obtained the 20 varieties images of peanut pods by scanner.Each pod includes one front and two side images of 100 peanuts respectively.For each image,we have acquired 50 characteristics including shape,color and texture categories and PCA optimization is conducted on these characteristics.Directed at the characteristics optimized by PCA and none,we construct the artificial neural network models and support vector machine model to identify different species.The results show that the acquisition features optimized by PCA show stronger recognition performance and SVM has higher recognition effect and more stability than neural network.The number of species affects the identification results.Under normal circumstances,we can determine the number of species by the number of features to improve the recognition efficiency.For 20 varieties,selecting more than 15 features is more appropriate.Features of color have better recognition results than texture and morphological character.Combining the characteristics of different categories,the overall recognition performance can reach more than 90%,which basically can be extended to actual production use.