摘要
快速准确地评估鱼类的新鲜度,对鱼类品质智能监控和保证食用者安全具有重要意义。目前基于图像的鱼类新鲜度评估方法中,基于鱼鳃特征的分析需去除鳃盖、对鱼体具有侵入性,利用其他部位则评估准确率较低。针对上述问题,提出了一种基于颜色纹理特征融合线性判别分析(CHGLDA)的鱼类新鲜度无损检测方法。首先,对采集的鱼体图像进行标注、图像缩放、颜色空间转换等预处理操作;然后,融合鱼体头部图像中提取的颜色直方图特征和灰度共生矩阵(GLCM)特征,并通过LDA进行特征降维;最后,利用K最近邻(KNN)算法对鱼的新鲜度进行分类。提出的GHGLDA新鲜度检测方法解决了提取的鱼体图像特征质量低致使分类性能差的问题。在真实的鲫鱼数据集上进行实验,其精确率、召回率、F1分数和准确率均为1。与颜色直方图、颜色矩、GLCM等特征相比,该方法在KNN、随机森林(RF)、人工神经网络(ANN)及轻量级梯度提升机(LightGBM)分类器上各评价指标的性能均有提高,其中KNN的评估时间最优,为0.01 s。
Quick and accurate assessment of fish freshness is of great significance for intelligent quality monitoring and ensuring the safety of consumers.In the current fish freshness evaluation method based on visual images,the study of fish gills needs to remove the gill cover,which is invasive to the fish body,and the analysis of other parts has a low evaluation accuracy.To solve the above problems,a fish freshness classification method based on color histogram&grey-level co-occurrence matrixlinear discriminant analysis(CHGLDA)was proposed.Firstly,preprocessing operations such as labeling,image zooming and color space conversion were performed on the collected fish images.Secondly,the extracted color histogram features and grey-level co-occurrence matrix(GLCM)features were fused to constitute the features,and the feature dimension was reduced by LDA.Finally,K-nearest neighbor(KNN)algorithm was used to classify fish freshness.The CHGLDA method proposed solved the problem of poor classification performance caused by the low quality of the extracted fish image features.The experiment was carried out on a real crucian data set,and the index values of precision,recall,F1-score and accuracy were all 1.Compared with color histogram features,color moment,GLCM features,etc.,this method improved the performance of each evaluation index on KNN,RF,ANN,and LightGBM classifiers.Among them,the evaluation time of KNN was the best,which was 0.01 s.Experimental results showed that this method can achieve accurate and non-destructive evaluation of fish freshness,and it was feasible for actual production monitoring.
作者
段青玲
徐晓玲
李道亮
李文升
刘春红
DUAN Qingling;XU Xiaoling;LI Daoliang;LI Wensheng;LIU Chunhong(College of Information and Electrical Engineering,China Agricultural University,Beijing 100083,China;National Innovation Center for Digital Fishery,China Agricultural University,Beijing 100083,China;Laizhou Mingbo Aquatic Products Co.,Ltd.,Laizhou 261400,China)
出处
《农业机械学报》
EI
CAS
CSCD
北大核心
2021年第10期385-393,共9页
Transactions of the Chinese Society for Agricultural Machinery
基金
山东省重大科技创新工程项目(2019JZZY010703)
宁波市公益事业科技项目(202002N3034)
江苏省农业科技自主创新资金项目(CX(19)1003)。