果实蝇属昆虫图像识别系统的开发与测试  

Developing and testing of image identification system for Bactrocera spp

作　　者：汪露[1] 黄丽莉[2] 杨慧勇[1,3] 高灵旺[1] 

机构地区：[1]中国农业大学农学与生物技术学院,北京100193 [2]江西出入境检验检疫局 [3]太原市星火技术发展中心

出　　处：《植物检疫》2013年第5期29-36,共8页Plant Quarantine

基　　金：国家质检总局科研计划项目(2009IK281)

摘　　要：为实现双翅目实蝇科果实蝇属昆虫图像识别,为出入境实蝇检疫鉴定、监测及初步分类工作提供帮助,本研究设计开发了"果实蝇属昆虫图像识别系统"。该系统以实蝇翅地标点间的欧氏距离为分类特征,采用随机森林算法,包含图像输入、图像预处理、图像标记和特征提取、分类器、结果显示、鉴定报告打印6个模块。系统测试与应用结果表明:该系统可实现试验果实蝇属昆虫种类的分类鉴定,识别效果好,稳定性高,对进行测试试验的果实蝇属5种实蝇共438个样本的平均识别准确率达到92.9%。系统具有良好的可扩展性,可通过向训练集中添加物种的方式,实现对更多实蝇种类的识别。该研究是实蝇图像识别研究领域的又一进展,在理论方法和实际应用方面对其他昆虫图像识别研究也有借鉴作用。To provide help to the identification,monitoring and preliminary classification of fruit flies in entry-exit inspection and quarantine,this research designed and developed 'Image Identification System for Bactrocera spp.' for automatic identification of Bactrocera spp.(Diptera Tephritidae).The system used the Euclidean distance between the landmarks of the fruit fly wing image as the classification feature and random forests as the main algorithm,which consisted of six modules including image input,image preprocessing,image landmark and feature extraction,classifier,outcome display and appraisal report print.Testing and application results show that: the system has good efficiency and high stability in classification and identification of fruit flies,and the average recognition accuracy of test samples,which contain 438 samples belong to 5 Bactrocera spp.,is 92.9 %.The system also shows good scalability.By adding more species to the training set,this system can achieve identification in more fruit fly species.This study is the latest study in fruit fly image recognition,and it also gives a reference about theoretical and practical application to image recognition study of other insects.

关 键 词：果实蝇 图像识别系统 地标点 随机森林 

分 类 号：S433[农业科学—农业昆虫与害虫防治] S41-3[农业科学—植物保护]