融合改进卷积神经网络和层次SVM的鸡蛋外观检测  

Egg appearance detection based on improved CNN and hierarchical SVM

作　　者：姚万鹏[1] 张凌晓[2] 赵肖峰 王飞成 YAO Wanpeng;ZHANG Lingxiao;ZHAO Xiaofeng;WANG Feicheng(Henan Economics and Management School,Nanyang,Henan 473000,China;Nanyang Institute of Technology,Nanyang,Henan 473000,China;Henan University of Technology,Zhengzhou,Henan 450001,China;Henan Agricultural University,Zhengzhou,Henan 450002,China)

机构地区：[1]河南省经济管理学校,河南南阳473000 [2]南阳理工学院,河南南阳473000 [3]河南工业大学,河南郑州450001 [4]河南农业大学,河南郑州450002

出　　处：《食品与机械》2025年第1期158-164,共7页Food and Machinery

基　　金：河南省高等学校重点项目(编号:24B520027);河南省科技攻关项目(编号:222102110045)。

摘　　要：[目的]实现鸡蛋精细化分类和提高鸡蛋外观检测的准确率。[方法]提出一种融合改进卷积神经网络和层次SVM的鸡蛋外观检测方案。(1)采用鸡蛋机器视觉图像采集设备获取不同方位、不同外观鸡蛋图像,并运用图像增强技术扩充鸡蛋图像数据库。(2)设计改进的浣熊优化算法(coati optimization algorithm,COA)和FCM聚类算法,在此基础上对卷积神经网络(convolutional neural network,CNN)模型结构和超参数进行优化,以提升CNN泛化能力。运用优化后的CNN深度学习鸡蛋图像数据库,从而实现鸡蛋外观图像特征的有效提取。(3)建立层次支持向量机鸡蛋外观分类工具,最终实现对鸡蛋外观的准确检测分类。[结果]所提鸡蛋外观检测方案的检测准确率提高了1.74%~4.31%,检测时间降低了21.68%~53.51%。[结论]所提方法能够有效实现对鸡蛋的在线实时精细化分类。[Objective]To achieve fine classification of eggs and improve the accuracy of egg appearance detection.[Methods]An egg appearance detection scheme based on improved convolutional neural network(CNN)and hierarchical support vector machine(SVM)was proposed.①Egg images with different orientations and appearances were captured using an egg machine vision image acquisition device,and image enhancement techniques were applied to expand the egg image database.②An improved Coati optimization algorithm(COA)and fuzzy C-means(FCM)clustering algorithm were designed,based on which the structure and hyperparameters of the CNN model were optimized to enhance its generalization ability.The optimized CNN was then used for deep learning on the egg image database to effectively extract features from egg appearance images.③A hierarchical SVM was established for fine classification of egg appearance,ultimately achieving accurate detection and classification of egg appearance.[Results]The detection accuracy of the proposed egg appearance detection scheme improved by 1.74%~4.31%,and the detection time was reduced by 21.68%~53.51%.[Conclusion]The proposed method effectively enables online real-time fine classification of eggs.

关 键 词：鸡蛋外观 卷积神经网络 浣熊优化算法 支持向量机 特征提取 

分 类 号：TS253.7[轻工技术与工程—农产品加工及贮藏工程] TP18[轻工技术与工程—食品科学与工程] TP391.41[自动化与计算机技术—控制理论与控制工程]