枸杞图像目标提取方法研究  

Target Extraction Method of Wolfberry Images

作　　者：马聪 陈学东 周慧 Ma Cong;Chen Xuedong;Zhou Hui(Institute of Agricultural Economy and Information Technology,Ningxia Academy of Agriculture and Forestry Sciences,Yinchuan,Ningxia 750002)

机构地区：[1]宁夏农林科学院农业经济与信息技术研究所,宁夏银川750002

出　　处：《宁夏农林科技》2021年第8期89-94,F0003,共7页Journal of Ningxia Agriculture and Forestry Science and Technology

基　　金：宁夏自然科学基金项目“基于机器视觉技术的黄花菜目标识别与定位方法研究”(2021AAC03257);宁夏农林科学院科技创新引导项目“基于机器视觉及深度学习的枸杞智能精选算法研究”(NKYQ-20-04)。

摘　　要：针对枸杞图像处理及准确分级的需求,本文开展枸杞图像果实目标的有效提取方法研究。直接采用二值化及灰度化方法处理图像,但是无法解决密度较大时枸杞间阴影粘连问题;根据图像目标阴影分布具有方向性的特点,采用Kirsch算子边缘检测方法,通过寻找卷积最大响应确定枸杞干果边界;根据目标与背景颜色差异明显的特点,采用RGB颜色模型识别红色区域部分,通过设置灰度值筛选条件实现枸杞提取。对比应用三种方法处理的图像效果,颜色模型提取目标的方法能准确地提取出单粒、间隙较小和边界相接类型的枸杞图像目标,二值化和边缘检测方法提取范围较小。本文研究结果可为非接触式分级及目标识别等应用场景提供理论依据。In order to meet the requirements of wolfberry image processing and accurate classification,the effective target extraction methods of wolfberry images were studied.The binarization process of images couldn't solve the problem of adhesion shadows in wolfberries when the density was larger.Then according to the directional patterns of image shadow distributions,Kirsch operator was used to detect the edge direction and determine the edge of the dried fruit of wolfberry by searching for the maximum convolution response.Based on obvious differences in target and background colors,RGB color model was used to recognize the red zone and extract wolfberry images by setting gray values screening conditions.Through comparing image effects of the three kinds of processing methods,the extraction method by using the color model could accurately wolfberry images with a single grain boundary,small grain boundary and boundary connect,while the extraction scopes of the other two methods were small.The research results could lay theoretical basis for contactless object classification and object identification and etc..

关 键 词：图像处理 KIRSCH算子 边缘检测 颜色模型 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TP18[自动化与计算机技术—计算机科学与技术]