俞. 龙,黄楚斌,唐劲驰,黄浩宜,周运峰,黄永权,孙佳琪.基于 YOLOX 改进模型的茶叶嫩芽识别方法[J].广东农业科学,2022,49(7):49-56 |
查看全文
HTML
基于 YOLOX 改进模型的茶叶嫩芽识别方法 |
Tea Bud Recognition Method Based onImproved YOLOX Model |
|
DOI:10.16768/j.issn.1004-874X.2022.07.007 |
中文关键词: 茶叶嫩芽识别 深度学习 YOLOX 算法 注意力机制 |
英文关键词: tea bud recognition in-depth learning YOLOX attention mechanism |
基金项目:广东省农业科学院农业优势产业学科团队建设项目(202125TD);广东省现代农业产业技术体系创新团队项目(2021KJ120);广东高校省级重点平台和重大科研项目(2018GkQNCX022);揭阳市科技创新发展专项(210531154613146) |
|
摘要点击次数: 860 |
全文下载次数: 830 |
中文摘要: |
【目的】改善茶叶嫩芽识别困难的问题,提高模型的识别准确率。茶叶嫩芽的识别是实现茶叶自
动化采摘的核心技术之一,而茶叶嫩芽生长的姿态以及采集图像时的拍摄角度等条件都会对茶叶嫩芽的识别带
来困难,造成识别准确率低的问题。【方法】提出一种改进的 YOLOX 茶叶嫩芽检测算法 SS-YOLOX,该方法能
准确地对一芽一叶、一芽二叶等茶叶嫩芽进行识别、分类。该方法通过添加注意力模块(Squeeze and excitation,
SE)提高模型的特征提取能力,改善小目标漏检问题、引入 Soft NMS 算法改善检测框重叠度较高时的打分机
制,提高模型对不同场景下嫩芽的识别能力。【结果】消融试验表明,引入 Soft-NMS 算法、SE 模块均能提
高 YOLOX 模型模型的检测精度,以引入 SE 模块提升较为明显。通过不同嫩芽图像对比验证算法的可行性和
准确性,SS-YOLOX 模型的均值平均精度 mAP 比原 YOLOX 模型提高 2.2%,达到 86.3%,表明经过改进后,
模型的识别能力得到提升。在目标嫩芽数量较多的情况下,SS-YOLOX 模型能有效地降低漏检率和错检率。
【结论】SS-YOLOX 模型能准确识别茶叶嫩芽,且识别效果更好,可为茶叶智能化采摘提供技术基础。 |
英文摘要: |
【Objective】The recognition of tea buds is one of the core technologies to realize automatic tea picking.
The growth posture of tea buds and the shooting angle during collecting images will bring difficulties to the recognition of tea buds, resulting in the problem of low recognition accuracy. The study is conducted to improve the difficulty in tea bud
recognition and improve the recognition accuracy of the model.【Method】The study presents an improved YOLOX tea bud
detection algorithm SS-YOLOX, which can accurately identify and classify tea buds including one bud with one leaf and one
bud with two leaves. In this method, the feature extraction ability of the model is improved by adding the attention module
(SE), the problem of missing detection of small targets is improved, the soft NMS algorithm is introduced to improve the scoring
mechanism when the detection frame overlap is high, and the ability of the model to recognize the buds in different scenes is
improved.【Result】The ablation test of YOLOX model shows that the detection accuracy of the model can be improved by
introducing soft NMS algorithm and SE module, but the improvement result is more obvious by introducing SE module. The
feasibility and accuracy of the algorithm are verified by comparing different bud images. The experimental results show that the
average accuracy mAP of SS-YOLOX model is 2.2% higher than that of the original YOLOX model, reaching 86.3%, indicating
that the recognition ability of the model is improved after the improvement. When the number of target buds is large, SSYOLOX model can effectively reduce the missed detection rate and false detection rate.【Conclusion】Therefore, SS-YOLOX
model can accurately identify tea buds, and the recognition effect is better, which can provide a technical basis for intelligent tea
picking. |
查看/发表评论 下载PDF阅读器 |