摘要
[目的]为了进一步提高稻飞虱图像分类的效率,本文提出一种基于特征优化的稻飞虱图像分类算法。[方法]对采集到的原始昆虫图像进行阈值分割、形态学滤波以及边缘跟踪来获取完整的昆虫彩色背部图像,同时基于该图像集提取昆虫的形态、颜色以及纹理特征66个,结合F-score特征评价方法,筛选出10个特征参数作为最优特征子集,并将其作为BP神经网络的输入特征值。[结果]当采用全部66个特征作为输入特征值时,稻飞虱图像的分类准确率达到96.19%;当采用最优特征子集作为输入特征值时,稻飞虱图像的分类准确率也为96.19%。[结论]以该方法获得的最优特征子集作为稻飞虱图像的特征参数不仅降低特征维度,提高稻飞虱图像分类效率,而且保证分类性能,为实现水稻虫害实时监测预警系统提供了技术支持。
[Objectives] In order to improve the efficiency of the image classification of rice planthopper(RPH),an image classification algorithm based on feature optimization has been proposed in this paper.[Methods] Threshold segmentation,morphological filtering and edge tracking were used to obtain the complete colorful back image of insects,and 66 of insect morphology,color and texture features were extracted based on the image set. With the F-score feature evaluation method,10 feature parameters were selected as the optimal feature subset,which were classified by BP neural network.[Results] When all 66 of features were used as input feature values, the insect image recognition accuracy rate reached 96.19%. What ’ s more,the recognition accuracy rate of the rice planthopper image was also 96.19%,while those input feature values were the optimal feature subset.[Conclusions] The optimal feature subset,presented in this paper,as the feature parameters of RPH image can reduce the feature dimension,improve the classification efficiency of RPH image and maintain the classification accuracy. The research results provide technical support for a real-time monitoring and forecasting system on main crop pests.
作者
陆静
王家亮
朱赛华
何瑞银
LU Jing;WANG Jialiang;ZHU Saihua;HE Rinyin(College of Engineering,Nanjing Agricultural University,Nanjing 210031,China)
出处
《南京农业大学学报》
CAS
CSCD
北大核心
2019年第4期767-774,共8页
Journal of Nanjing Agricultural University
基金
国家重点研发计划项目(2016YFD0300908)
关键词
稻飞虱
图像分类
特征优化
BP神经网络
rice planthopper
image classification
feature optimization
BP neural network
