基于BP神经网络模型的果实蝇自动分类系统  

A Study on the Automatic Classification System for Fruit Flies Based on BP Neural Network Model

作　　者：彭莹琼[1,2] 廖牧鑫 张永红[4] 黄丽莉[2] 殷华[1] 唐建军[3] 邓泓[1] PENG Ying-qiong LIAO Mu-xin ZHANG Yong-hong HUANG Li-li YIN Hua TANG Jian-jun DENG Hong(Software College, Jiangxi Agricultural University, Nanchang 3300451, China Jiangxi Entry- exit In- spection and Quarantine Inspection and Quarantine Technology Center, Nanehang 330038, China College of Computer and Information Engineering, Jiangxi Agricultural University, Nanchang 330045, China Jiangxi Vo- cational College of Mechanical ＆ Electrical Technology, Nanchang 330013, China)

机构地区：[1]江西农业大学软件学院,江西南昌330045 [2]江西出入境检验检疫局综合技术中心,江西南昌330038 [3]江西农业大学计算机与信息工程学院,江西南昌330045 [4]江西机电职业技术学院,江西南昌330013

出　　处：《江西农业大学学报》2016年第6期1205-1210,共6页Acta Agriculturae Universitatis Jiangxiensis

基　　金：国家质检公益性行业科研专项支助项目(201410080);江西省科技支撑计划项目(0123BBF60177)~~

摘　　要：果实蝇属昆虫危害水果和蔬菜,造成产量下降,影响对外贸易,对其识别是检疫工作中的重要部分,现有的人工辨识方法受时间、知识等因素影响不能准确、有效辨识。提出了一种基于BP神经网络模型的果实蝇分类方法,采用几何形态测量学中的标记点法对果蝇翅进行特征提取,通过方差分析确定了用于果蝇鉴定的11个主特征,建立3层BP神经网络模型,结合Levenberg-Marquardt BP训练函数对数据集进行训练,得到完整的可用于果实蝇分类的BP神经网络。实验表明,该方法能够对实蝇进行有效的辨识,对桔小实蝇、瓜实蝇、具条实蝇和南亚果实蝇等高风险果实蝇辨识的准确率分别是90.0%、93.3%、90.0%和96.7%,总体准确率为92.5%,具有良好的应用前景。Fruit fly genus insects hazard fruits and vegetables,resulting in production decline and undermine foreign trade.Its recognition is an important part of the quarantine work. Artificial identification methods restricted by factors of time,knowledge,etc,cannot accurately and effectively identify the insects.This paper introduces a system for fruit fly classification based on a BP neural network model,using geometric morphometric method in marker on Drosophila extracts the wing features,identifies 11 main features by analysis of variance,and establishes the three layer BP neural network model. Combined with Levenberg- Marquardt BP training function the data sets were trained to give a complete BP neural network,which can be used to classify the fruit fly. Experiments show that the method can effectively identify fruit flies,melon fly,stripped fly and the accuracy of identification are 90.0%,93.3%,90.0% and 96.7%,respectively the overall accuracy is 92.5%.This method hsa a good prospect.

关 键 词：BP神经网络 果实蝇分类 鉴定 图像识别 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]