基于机器视觉的鸡胴体质量分级方法

发布时间：2018-06-04 21:56

本文选题：鸡胴体 + 质量分级　；参考：《农业机械学报》2017年06期

【摘要】：提出一种基于机器视觉技术的鸡胴体质量分级方法。使用数码相机在肉鸡屠宰厂随机采集95幅鸡胴体图像,对采集图像预处理后,提取出鸡胴体投影面积、轮廓长度和胸宽等6个图像特征。然后以这6个特征参数为输入,利用95个样本为训练集,通过回归分析的方法,分别建立预测鸡胴体质量的一元线性回归模型和多元线性回归模型,找出预测质量的最佳模型,最后采集5组共100个样本为验证集,对最佳分级模型进行验证。结果显示,鸡胴体图像的6个特征参数中,基于投影面积的一元线性模型决定系数最大,为0.827;基于投影面积等4个特征量的多元线性模型决定系数最大,为0.880。根据样本数据的学生化残差剔除了8个异常点的数据,修正后的多元线性模型决定系数为0.933,并将其作为最佳模型。利用最佳模型对验证集样本进行质量分级,模型对鸡胴体质量等级判定的平均正确率可达89%。结果表明基于图像特征的鸡胴体自动分级方法是可行的。
[Abstract]:A method of carcass quality classification based on machine vision is proposed. 95 carcass images were collected at random by digital camera in broiler slaughtering plant. After preprocessing, six image features such as carcass projection area, contour length and chest width were extracted. Then, with these six characteristic parameters as input and 95 samples as training set, the univariate linear regression model and multivariate linear regression model for predicting carcass quality of chicken were established by regression analysis, and the best model for predicting carcass quality was found out. Finally, five groups of 100 samples were collected to verify the optimal classification model. The results showed that, among the six characteristic parameters of carcass image, the single linear model based on projection area had the largest determining coefficient (0.827), and the multivariate linear model based on projection area had the largest determination coefficient (0.880). According to the student residual of the sample data, the data of 8 outliers are eliminated, and the decision coefficient of the modified multivariate linear model is 0.933, which is regarded as the best model. The best model was used to grade the samples of the verification set, and the average correct rate of the model for judging the carcass quality grade of chicken was up to 89%. The results show that the automatic classification method based on image features is feasible.
【作者单位】：南京农业大学工学院;
【基金】：公益性行业科研专项(201303083-2)
【分类号】：TP391.41;TS251.7

【相似文献】