| 引用本文： |
-
陈峰练,白琳,张扬虎,张茜,李陶深.基于混合深度学习的蔗糖原料蔗顶芽杂质探测技术[J].广西科学,2021,28(3):242-248. [点击复制]
- CHEN Fenglian,BAI Lin,ZHANG Yanghu,ZHANG Qian,LI Taoshen.Detection of Impurities in Sugarcane Top Buds Based on Hybrid Deep Learning[J].Guangxi Sciences,2021,28(3):242-248. [点击复制]
|
|
| 本文已被：浏览 621次 下载 551次 |
码上扫一扫！ |
| 基于混合深度学习的蔗糖原料蔗顶芽杂质探测技术 |
|
陈峰练1, 白琳1,2, 张扬虎1, 张茜3, 李陶深1,2
|
|
|
| (1.广西大学计算机与电子信息学院, 广西南宁 530004;2.广西高校并行与分布式计算技术重点实验室, 广西南宁 530004;3.广西医科大学第一附属医院, 广西南宁 530021) |
|
| 摘要: |
| 为有效地解决复杂情况下蔗糖原料蔗顶芽杂质的识别问题，本研究对复杂环境下高密集度的原料蔗顶芽杂质的特征进行分析，将特征金字塔网络（Feature Pyramid Network，FPN）和RoIAlign整合到Faster R-CNN和Cascade R-CNN的架构中，再将多种深度网络算法融入到统一的多层架构的深度网络框架中，创新地设计出一种混合深度学习模型。通过实验验证，在原料蔗顶芽杂质的探测任务中，相较于传统的深度网络模型，本研究提出的混合深度学习模型在性能上有较大的提升。表明本研究提出的混合深度学习模型取得较好的平均精度，其识别性能已经可以达到实际应用水准。 |
| 关键词: 深度学习 目标检测 农业智能化 甘蔗杂质探测 蔗糖 智能化生产 |
| DOI：10.13656/j.cnki.gxkx.20210830.010 |
| 投稿时间：2020-12-15 |
| 基金项目:国家自然科学基金项目（61966003）和广西自然科学基金项目（2020GXNSFAA159171）资助。 |
|
| Detection of Impurities in Sugarcane Top Buds Based on Hybrid Deep Learning |
|
CHEN Fenglian1, BAI Lin1,2, ZHANG Yanghu1, ZHANG Qian3, LI Taoshen1,2
|
| (1.School of Computer, Electronics and Information, Guangxi University, Nanning, Guangxi, 530004, China;2.Guangxi Colleges and Universities Key Laboratory of Parallel and Distributed Computing Technology, Nanning, Guangxi, 530004, China;3.The First Affiliated Hospital of Guangxi Medical University, Nanning, Guangxi, 530021, China) |
| Abstract: |
| In order to effectively solve the problem of identifying sugarcane top bud impurities in sucrose raw materials under complex conditions, this study analyzes the characteristics of high-density raw sugarcane impurities in complex environments, and integrates Feature Pyramid Network (FPN) and RoIAlign into the architecture of Faster R-CNN and Cascade R-CNN. Then a variety of deep network algorithms are integrated into the unified multi-layer deep network framework, and a hybrid deep learning model is innovatively designed. The experimental verification shows that the hybrid deep learning model proposed in this study has a great improvement in performance compared with the traditional deep network model in the detection task of sugarcane top bud impurities. It demonstrates that the hybrid deep learning model proposed in this study has achieved good average accuracy, and its recognition performance has reached the level of practical application. |
| Key words: deep learning object detection intelligent agriculture detection of sugarcane impurities sucrose intelligent production |
