焦俊,王文周,侯金波,孙裴,何屿彤,辜丽川.基于改进残差网络的黑毛猪肉新鲜度识别方法[J].农业机械学报,2019,50(8):364-371.
JIAO Jun,WANG Wenzhou,HOU Jinbo,SUN Pei,HE Yutong,GU Lichuan.Freshness Identification of Iberico Pork Based on Improved Residual Network[J].Transactions of the Chinese Society for Agricultural Machinery,2019,50(8):364-371.
摘要点击次数: 1594
全文下载次数: 720
基于改进残差网络的黑毛猪肉新鲜度识别方法   [下载全文]
Freshness Identification of Iberico Pork Based on Improved Residual Network   [Download Pdf][in English]
投稿时间:2019-01-21  
DOI:10.6041/j.issn.1000-1298.2019.08.040
中文关键词:  猪肉  新鲜度  迁移学习  LReLu-Softplus激活函数  改进残差网络
基金项目:安徽省重点研究与开发计划项目(1804a0702130)、国家自然科学基金项目(31671589、31771679)、安徽省科技重大攻关项目(16030701092)和农业农村部农业电子商务重点实验室开放基金项目(AEC2018010、AEC2018003)
作者单位
焦俊 安徽农业大学 
王文周 安徽农业大学 
侯金波 安徽泓森物联网有限公司 
孙裴 安徽农业大学 
何屿彤 安徽农业大学 
辜丽川 安徽农业大学 
中文摘要:为了提高黑毛猪肉新鲜度的识别准确率,提出基于改进残差网络和迁移学习的黑毛猪肉新鲜度识别方法。首先,根据猪肉的微生物菌体浓度、大肠菌菌体浓度和pH值,结合国家标准,将猪肉新鲜度分为7个类别;然后,将ResNet-50模型用PfidSet数据集训练,使其具有抽取图像特征的能力,利用模型迁移和模型微调对ResNet-50模型进行改进,即用一个3层的自适应网络取代ResNet-50模型的全连接层和分类层,再使用在PfidSet上训练的网络参数初始化改进的ResNet-50模型权重,运用LReLu-Softplus作为自适应网络的激活函数;最后,将改进ResNet-50模型在猪肉样品的图像数据集上学习得到的知识,迁移到黑毛猪肉新鲜度识别任务。选取7类共计23427幅黑毛猪肉图像组成样本集,从样本集中随机选择80%的样本用作训练集、其余20%用作测试集进行测试,试验结果表明,迁移学习能够明显提高模型的收敛速度和识别性能,数据扩充有助于增加数据的多样性,避免出现过拟合现象,在迁移学习和数据扩充方式下的总体识别准确率达到94.5%,是一种高效的猪肉新鲜度识别方法。
JIAO Jun  WANG Wenzhou  HOU Jinbo  SUN Pei  HE Yutong  GU Lichuan
Anhui Agricultural University,Anhui Agricultural University,Anhui Hongsen Internet of Things Co., Ltd.,Anhui Agricultural University,Anhui Agricultural University and Anhui Agricultural University
Key Words:pork  freshness  transfer learning  LReLu-Softplus activation function  improved residual network
Abstract:In order to improve the accuracy of pork freshness identification, a method for pork freshness identification based on improved residual network and transfer learning was proposed. First of all, the pork freshness was classified into seven grades, according to the aerobic plate count, coliform bacteria and pH value of pork combined with national pork food standards(national standards). The ResNet-50 model was trained with the PfidSet data set to have the ability of extracting image features. Then, the ResNet-50 model was improved by using model transferring and model fine tuning in the following ways: firstly, replacing the full connection and classification layers of the ResNet-50 model with a 3 layer adaptive network; next, initializing the improved ResNet-50 model weights by using the network parameters trained on the PfidSet; then using LReLu-Softplus as the activation function of the adaptive network; finally, transferring the knowledge gained by the improved ResNet-50 model on the image data set of the pork sample to the task of Iberico pork freshness identification. A total of 23427 images were selected to form the sample set. Then, 80% of the samples were randomly selected from the sample set to be used as the training set, and the remaining 20% for the test set. The test results showed that transfer learning could significantly improve the convergence speed and classification performance of the model, and data augmentation could increase the diversity of data, avoiding over fitting phenomena. The accuracy of classification in transfer learning and data augmentation could reach as high as 94.5%. Moreover, the test method was an efficient method for classifying pork freshness.

Transactions of the Chinese Society for Agriculture Machinery (CSAM), in charged of China Association for Science and Technology (CAST), sponsored by CSAM and Chinese Academy of Agricultural Mechanization Science(CAAMS), started publication in 1957. It is the earliest interdisciplinary journal in Chinese which combines agricultural and engineering. It always closely grasps the development direction of agriculture engineering disciplines and the published papers represent the highest academic level of agriculture engineering in China. Currently, nearly 8,000 papers have been already published. There are around 3,000 papers contributed to the journal each year, but only around 600 of them will be accepted. Transactions of CSAM focuses on a wide range of agricultural machinery, irrigation, electronics, robotics, agro-products engineering, biological energy, agricultural structures and environment and more. Subjects in Transactions of the CSAM have been embodied by many internationally well-known index systems, such as: EI Compendex, CA, CSA, etc.

   下载PDF阅读器