| 本文已被：浏览 1201次 下载 1191次 |
码上扫一扫！ |
| 基于深度卷积网络的育肥猪体重估测 |
|
张建龙1,2,冀横溢1,2,滕光辉1,2,3*
|
|
|
| (1.中国农业大学 水利与土木工程学院, 北京 100083;(2.农业农村部设施农业工程重点实验室, 北京 10083;(3.北京市畜禽健康养殖环境工程技术研究中心, 北京 100083) |
|
| 摘要: |
| 为快速、无应激、准确地获取育肥猪体重数据，采用深度卷积网络对育肥猪体重进行了估测。结果表明:1)在改造后的Xception、MobileNetV2、DenseNet201和ResNet152V2 4 种模型中，DenseNet201模型体重估测效果最好，在验证集上估测的相关系数为0.993 9，均方根误差为1.85 kg，平均绝对误差为1.10 kg，平均相对误差1.57%，被选为本研究所用的育肥猪体重估测模型；2)在测试数据上考察了该模型的泛化效果，其估测的相关系数为0.976 7，均方根误差为2.75 kg，平均绝对误差为2.10 kg，平均相对误差3.03%，效果良好；3)该模型的平均估测时间为0.16 s，其处理速度远快于传统方法，更适合用于育肥猪分群系统、母猪饲喂站等对猪只体重获取速度要求严格的场合。综上，深度卷积网络模型可用于快速估测育肥猪体重，为猪场的自动化、智能化和无人化管理提供依据。 |
| 关键词: 育肥猪 体重估测 深度学习 卷积神经网络 |
| DOI：10.11841/j.issn.1007-4333.2021.08.11 |
| 投稿时间：2020-10-17 |
| 基金项目:“十三五”国家重点研发计划(2016YFD0700204)；重庆市技术创新与应用发展专项重点项目(cstc2019jscx-gksbX0093) |
|
| Weight estimation of fattening pigs based on deep convolutional network |
|
ZHANG Jianlong1,2,JI Hengyi1,2,TENG Guanghui1,2,3*
|
| (1.College of Water Resources and Civil Engineering, China Agricultural University, Beijing 100083, China;2.Key Laboratory of Agricultural Engineering in Structure and Environment of Ministry of Agriculture and Rural Affairs, Beijing 100083, China;3.Beijing Engineering Research Center on Animal Healthy Environment, Beijing 100083, China) |
| Abstract: |
| In order to obtain the weight data of fattening pigs quickly, accurately and without stress, deep convolutional networks were applied to estimate the weight of fattening pigs. The results showed that: 1)Among the transformed Xception, MobileNetV2, DenseNet201 and ResNet152V2, the pre-trained DenseNet201 model had the best weight estimation effect. On the validation set, the estimated correlation coefficient of the model was 0. 9939, the root mean square error was 1. 85 kg, the mean absolute error was 1. 10 kg, and the mean relative error was 1. 57%. The Densenet201 model was then selected as the fattening pig weight estimation model in this study; 2)The generalization effect of the model was investigated based on the test data. The estimated correlation coefficient of 0. 976 7, the root mean square error of 2. 75 kg, the mean absolute error of 2. 10 kg, and the mean relative error of 3. 03% showed that the performance of the model was good. 3)The mean estimated time of pre-trained DenseNet201 was 0. 16 s, which was much faster than that of the traditional methods. Therefore, the Densenet201 model was more suitable for sorting systems, sow feeding stations and other occasions where the speed of pig weight acquisition was strict. In conclusion, the deep convolution network can be used to quickly estimate the weight of fattening pigs and provides basis for automated, intelligent and unmanned management of pig farms. |
| Key words: fattening pigs weight estimation deep learning convolutional neural network |
