| 基于迁移学习的棉花识别 |
| |
| 引用本文： | 王见,田光宝,周勤.基于迁移学习的棉花识别[J].浙江农业学报,2020,32(8):1457. |
| |
| 作者姓名： | 王见 田光宝 周勤 |
| |
| 作者单位： | 1.重庆大学 机械传动国家重点实验室,重庆 400044;2.重庆大学 机械工程学院,重庆 400044 |
| |
| 基金项目： | 重庆市创新重大主题专项(cstc2017rgzn-zdyfX0007); 重庆科技计划(cstc2018jszx-cyztzxX0026) |
| |
| 摘 要： | 提高智能采棉机效率的一个重要途径是实现单个、重叠和遮挡棉花的识别,避免误采摘和漏采摘。针对不同形态棉花的识别,常规的特征提取方法难以达到令人满意的结果,因而采用基于迁移学习的棉花识别方法和基于迁移模型的特征提取与极限学习机(extreme learning machine,ELM)相结合的方法进行棉花识别研究。首先更改AlexNet、GoogleNet、ResNet-50模型分类层和设置相关参数,用训练好的迁移模型对棉花验证集识别,然后利用训练好的迁移模型进行棉花数据集特征提取,再用训练集的特征训练ELM模型,统计不同隐含层神经元个数的ELM模型对棉花的识别准确率。AlexNet、GoogleNet、ResNet-50迁移模型识别率依次为92.03%、93.19%、93.68%;使用特征提取再与ELM结合的方法,准确率比对应迁移模型分别提高了1.97、1.34、1.55百分点。结果表明,迁移模型对小样本棉花识别也有较高准确率,基于特征提取与ELM相结合的方法可进一步提高准确率。
|
| 关 键 词： | 棉花 识别 迁移学习 极限学习机 特征提取 |
| 收稿时间： | 2020-01-16 |
Cotton recognition based on transfer learning |
| |
| WANG Jian,TIAN Guangbao,ZHOU Qin.Cotton recognition based on transfer learning[J].Acta Agriculturae Zhejiangensis,2020,32(8):1457. |
| |
| Authors: | WANG Jian TIAN Guangbao ZHOU Qin |
| |
| Institution: | 1. The State Key Laboratory of Mechanical Transmission, Chongqing University, Chongqing 400044, China;
2. College of Mechanical Engineering, Chongqing University, Chongqing 400044, China |
| |
| Abstract: | An important way to improve the efficiency of the intelligent cotton picker is to realize the identification of single cotton, overlapped cottons, blocked cottons to avoid false picking and missed picking. The conventional feature extraction method is difficult to achieve satisfactory results for the identification of different types of cotton. Methods based on transfer learning and feature extraction based on transfer learning combined with extreme learning machine (ELM) was proposed. Firstly, the classification layer of AlexNet, GoogleNet and ResNet-50 models was changed and the relevant parameters were set, and the trained transfer model was used to identify the cotton validation set; secondly, the trained transfer model was used to extract the characteristics of cotton data set, and then the characteristics of the training set were used to train the ELM model to calculate the accuracy of the ELM model with different hidden layer neurons. The recognition rates of AlexNet, GoogleNet and ResNet-50 transfer models were 92.03%, 93.19% and 93.68%, respectively. The accuracy of the method of feature extraction combined with ELM was 1.97, 1.34 and 1.55 percentage points higher than that of the corresponding transfer model. The results showed that the transfer model could obtain higher accuracy for small sample cotton recognition. The method based on feature extraction and ELM could further improve the accuracy. |
| |
| Keywords: | cotton recognition transfer learning extreme learning machine feature extraction |
| 本文献已被 CNKI 等数据库收录！ |
| 点击此处可从《浙江农业学报》浏览原始摘要信息 |
| 点击此处可从《浙江农业学报》下载免费的PDF全文 |
|
