| 基于级联卷积神经网络的番茄花期识别检测方法 |
| |
| 引用本文： | 赵春江,文朝武,林森,郭文忠,龙洁花.基于级联卷积神经网络的番茄花期识别检测方法[J].农业工程学报,2020,36(24):143-152. |
| |
| 作者姓名： | 赵春江 文朝武 林森 郭文忠 龙洁花 |
| |
| 作者单位： | 上海海洋大学信息学院,上海 201306;北京农业智能装备技术研究中心,北京 100097;北京农业智能装备技术研究中心,北京 100097 |
| |
| 基金项目： | 宁夏回族自治区重点（重大）研发计划项目（2019BBF02010)；国家自然科学基金项目（31601794）；北京市农林科学院青年基金（QNJJ202027） |
| |
| 摘 要： | 对作物花期状态的准确识别是温室作物执行授粉的前提。为提高花期状态识别的准确度，该研究以温室番茄为例提出了一种基于级联卷积神经网络的番茄花朵花期识别方法。首先采用改进的端到端的特征金字塔网络FS-FPN实现番茄花束的分割，然后采用Prim最小生成树对分割后的花束图片进行花期识别优先级排序，最后将已排序的分割花束图片输入改进的Yolov3网络，实现番茄花朵花期状态的精准识别。在由1600幅包含花蕾期、全开期、谢花期、初果期4类花期状态的番茄花束图像构成的数据集上，所提方法对番茄花朵花期平均检测时间为12.54 ms，平均检测精度分别比Mask R-CNN和SPP-Net提高了3.67%和2.39%，识别错误率比改进前的Yolov3网络降低了1.25%。最终将该方法部署到番茄授粉机器人上，并在大型玻璃温室内进行验证，结果表明，所提方法对番茄花朵各花期的检测精度分别为花蕾期85.71%、全开期95.46%、谢花期62.66%、初果期88.34%，该研究结果可为智能授粉机器人的精准作业提供重要依据。
|
| 关 键 词： | 识别 级联神经网络 小目标检测 授粉机器人 |
| 收稿时间： | 2020/8/31 0:00:00 |
| 修稿时间： | 2020/12/11 0:00:00 |
Tomato florescence recognition and detection method based on cascaded neural network |
| |
| Zhao Chunjiang,Wen Chaowu,Lin Sen,Guo Wenzhong,Long Jiehua.Tomato florescence recognition and detection method based on cascaded neural network[J].Transactions of the Chinese Society of Agricultural Engineering,2020,36(24):143-152. |
| |
| Authors: | Zhao Chunjiang Wen Chaowu Lin Sen Guo Wenzhong Long Jiehua |
| |
| Institution: | 1.College of Information Science, Shanghai Ocean University, Shanghai 201306, China; 2.Beijing Agricultural Intelligent Equipment Technology Research Center, Beijing 100097, China |
| |
| Abstract: | Abstract: Accurate identification of the flowering state of crops is a prerequisite for pollination of greenhouse crops. In order to improve the accuracy of the flowering state recognition, as a key decision for the precise operation of the tomato pollination robot, due to the complex growth environment of tomato flowers, the flowers present small and multi-target distributions, and the same bouquet has the characteristics of flowers of different flowering periods. A single network can realize the recognition of tomato bouquets, but it cannot simultaneously realize the accurate recognition of the flowers in the bouquets with multiple flowering periods, resulting in insufficient flower characteristic information. In response to these problems, this paper proposes a method of cascading two-level neural networks in hopes Realize the research of precise identification of tomato flower blooming period, and explore a new identification method for the precise operation of tomato pollination robot. First, the improved end-to-end feature pyramid network FS-FPN is used to achieve the segmentation of tomato bouquets, and then the Prim minimum spanning tree is used to prioritize the flowering of the segmented bouquet pictures, and finally the sorted segmented bouquet pictures are input to the improved Yolov3 the network realizes accurate identification of the flowering state of tomato flowers. Shooting at 8 am, 12 noon, and 18:00, respectively, and experimented on a data set consisting of 1 600 tomato bouquet images in four types of flowering states: bud, full bloom, declining, and early fruit. The first cascade improved FS-FPN network used to perform multi-scale prediction and pixel segmentation of tomato bouquets can effectively provide multi-scale input for the second-level network flower flowering period recognition to improve the accuracy of flower flowering period detection. The average correct segmentation rate is 98.11%, the over segmentation rate is 3.56%, and the missing segmentation rate is 3.42%. The proposed method uses a multi-scale, multi-input network architecture, and increases the target feature information fusion on the basis of increasing the network operation speed, so that the model recognition rate and accuracy are higher. The average detection accuracy MAP for the flowering period of tomato flowers is 82.79%, and the average detection time is 12.54 ms. The average detection accuracy is 3.67% and 2.39% higher than Mask R-CNN and SPP-Net, respectively, and the recognition error rate is lower than that of the Yolov3 network before the improvement. It''s 1.25%. Finally, the method was deployed on the tomato pollination robot and verified in a large glass greenhouse. The results showed that the detection accuracy of the proposed method for each flowering period of tomato flowers was 85.71% in the bud period, 95.46% in the full bloom period, and 62.66 %in the flowering period, 88.34% of initial fruit period. Especially in the recognition accuracy of the flower bud stage, compared with the improvement of the Yolov3 network before the improvement, it has increased by nearly 7%. Due to the great similarity in the color and shape characteristics of the flowering period and the bud period, the recognition accuracy of the flowering period is low. However, when it is deployed to the facility tomato pollination robot in the later period, the flowers in the flowering period do not need to be pollinated. The research results can provide an important basis for the precise operation of intelligent pollination robots. |
| |
| Keywords: | Tomato cascade network Small target detection pollination robot |
| 本文献已被 万方数据 等数据库收录！ |
| 点击此处可从《农业工程学报》浏览原始摘要信息 |
| 点击此处可从《农业工程学报》下载免费的PDF全文 |
|
