基于CNN-SVM模型的鸡蛋外观品质检测
Egg appearance quality detection based on CNN-SVM model
齐歌 1赵峰 2李婉宁3
作者信息
- 1. 新乡职业技术学院,河南 新乡 453006
- 2. 河南农业大学,河南 郑州 450046
- 3. 河南工业大学,河南 郑州 450001
- 折叠
摘要
[目的]提高鸡蛋外观品质检测的精度,建立CNN-SVM模型的鸡蛋外观品质检测模型.[方法]结合CNN的自适应特征提取功能和SVM的超强泛化分类性能,通过6层卷积神经网络结构处理提取全连接层的特征,采用CNN-SVM混合模型替代传统CNN+softmax,构建一个基于CNN-SVM模型的鸡蛋外观品质检测方法.[结果]与SVM模型、CNN模型和KNN模型相比,CNN-SVM模型在准确率、精确率、召回率和F1 分数方面表现优异,分别为 97.97%,98.10%,98.10%,98.00%.KNN模型在鸡蛋外观品质检测上的精度最低,其准确率、精确率、召回率和F1分数分别为77.46%,79.44%,76.75%,76.90%.[结论]CNN-SVM模型具有很强的鲁棒性和抗噪声能力,可以有效提高鸡蛋外观品质检测的准确性和适用性.
Abstract
[Objective]In order to improve the accuracy of egg appearance quality detection,an egg appearance quality detection model based on CNN-SVM model was established.[Methods]Combined with the adaptive feature extraction capability of CNN and the super-generalization classification capability of SVM,the features of fully connected layers were extracted by six-layer convolutional neural network structure processing,and the CNN-SVM hybrid model was adopted,instead of the traditional CNN+softmax,an egg appearance quality detection method based on CNN-SVM model was proposed.[Results]Compared with SVM model,CNN model and KNN model,CNN-SVM model had better performance in accuracy,precision,recall and F1 score,which were 97.97%,98.10%,98.10%and 98.00%respectively.KNN model had the lowest accuracy in egg appearance quality detection,and its accuracy,precision,recall and F1 fraction are 77.46%,79.44%,76.75%and 76.90%,respectively.[Conclusion]The CNN-SVM model has strong robustness and anti-noise ability,which can effectively improve the accuracy and applicability of egg appearance quality detection..
关键词
卷积神经网络/支持向量机/鸡蛋外观/全连接层Key words
convolutional neural network/support vector machine/egg appearance/full connection layer引用本文复制引用
基金项目
河南省科技攻关项目(21K0437304)
河南省高等教育教学改革研究与实践项目(22SJGLX1308)
出版年
2024
国家科技期刊平台
NSTL
万方数据