基于线性支持向量机的温室害虫智能识别系统  

Intelligent Identification System of Greenhouse Pests Based on Linear Support Vector Machine

作　　者：刘豹 李翌 李峰 鲍煦[1] LIU Bao;LI Yi;LI Feng;BAO Xu(School of Computer Science and Communication Engineering,Jiangsu University,Zhenjiang 212000,China;Amway(China)Plant Research and Development Center Co.,Ltd.,Wuxi 214000,China)

机构地区：[1]江苏大学计算机科学与通信工程学院,江苏镇江212001 [2]安利(中国)植物研发中心有限公司,江苏无锡214000

出　　处：《软件导刊》2023年第12期232-237,共6页Software Guide

基　　金：江苏省六大人才高峰高层次人才计划项目(XYDXX-115);江苏省“333”工程青年人才项目(2022)。

摘　　要：针对温室现场采集的害虫粘虫板图像易受光照不均匀、灯光反射等干扰,从而影响识别精度的问题,基于图像分割算法与线性支持向量机构建一个温室害虫智能识别系统。该系统利用Prewitt边缘检测二值图、Canny边缘检测二值图融合全局阀值分割的方法实现粘虫板图像中害虫区域的精准分割。基于分割的害虫图像人工构建线性支持向量机的训练数据集,并根据飞虱、蓟马特征将单个样本特征提取扩充至12个;基于扩充的训练数据集构建基于线性支持向量机的害虫识别分类器,并使用梯度下降法进行模型训练。实验结果表明,该系统可快速准确地实现粘虫板图像中害虫目标区域的分割,平均准确率为96.3%;针对分割后的图像,该系统可准确实现害虫识别,平均准确率为96.1%,其中飞虱、蓟马的识别准确率分别为95.4%、96.8%。To address the issue of the susceptibility of insect pest and sticky insect board images collected on site in greenhouses to noise such as uneven lighting and light reflection,which affects recognition accuracy,an intelligent recognition system for greenhouse pests is built based on image segmentation algorithms and linear support vector machines.This system utilizes the fusion of Prewitt edge detection binary image and Canny edge detection binary image with global threshold segmentation to achieve precise segmentation of pest areas in sticky insect board images.Artificial construction of a linear support vector machine training dataset based on segmented pest images,and expansion of single sample feature extraction to 12 based on planthopper and thrip features;Construct a linear support vector machine based pest recognition clas⁃sifier based on the expanded training dataset,and use gradient descent method for model training.The experimental results show that this method can quickly and accurately segment the pest target area in the sticky insect board image,with an average accuracy of 96.3%;For seg⁃mented images,this method can accurately identify pests with an average accuracy of 96.1%,with recognition accuracy rates for planthoppers and thrips being 95.4%and 96.8%,respectively.

关 键 词：边缘检测 支持向量机 温室害虫 害虫识别 

分 类 号：TP319[自动化与计算机技术—计算机软件与理论]