基于显著性检测的害虫图像自动分割算法研究  

Research on Pest Image Automatic Segmentation Algorithm Based on Saliency Detection

作　　者：陈向东 邓江洪 Chen Xiangdong;Deng Jianghong(School of Information Engineering,Huanghuai University,Zhumadian 463000;School of Animation,Huanghuai University,Zhumadian 463000)

机构地区：[1]黄淮学院信息工程学院,驻马店463000 [2]黄淮学院动画学院,驻马店463000

出　　处：《中国粮油学报》2021年第1期181-186,共6页Journal of the Chinese Cereals and Oils Association

基　　金：河南省科技厅科技攻关(122102210549,132102210423)。

摘　　要：图像分割技术应用于农田检测中能够区分不同种类及程度的虫害,极大程度的节省时间并提高劳动效率。首先建立了害虫显著性图像的分割模型,从DCT系数中提取颜色、亮度、纹理和深度的特征,并基于图像块之间存在空间距离,采用高斯模型加权来估计图像显著性,然后设计了一种融合算法,对这些选定的图像特征进行检测分类。研究结果表明,所提的分割算法不仅能够对田间图像进行无监督分类,还能实现对害虫的高效、精准分类与在线监测,有利于不同类型杀虫剂的实时选择。Image segmentation technology can be applied to farmland detection to distinguish between different types and degrees of pests,which could be able to save time and improve labor efficiency significantly.Firstly,a segmentation model of pest saliency image was established,and the features of color,brightness,texture and depth were extracted from DCT coefficients.Based on the spatial distance between image blocks,Gaussian model weighting was used to estimate the image saliency.At the same time,a kind of image saliency was estimated.A fusion algorithm detects and classifies these selected image features.The research results showed that the proposed segmentation algorithm could perform the unsupervised for the field image.Besides,it could also realize efficient and accurate classification and online monitoring of pests,beneficial to the real-time selection of different types of pesticides.

关 键 词：田间害虫 图像分割 显著性检测 纹理分析 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]