التفاصيل البيبلوغرافية
| العنوان: |
基于 YOLOv8-FECA 的密集场景下小麦赤霉病孢子目标检测. (Chinese) |
| Alternate Title: |
Detection of wheat scab spores in dense scene based on YOLOv8-FECA. (English) |
| المؤلفون: |
张东彦, 高 玥, 程 涛, 胡根生, 杨 雪, 乔红波, 郭 伟, 谷春艳 |
| المصدر: |
Transactions of the Chinese Society of Agricultural Engineering; Nov2024, Vol. 40 Issue 21, p127-136, 10p |
| مصطلحات موضوعية: |
PLANT diseases, FOOD safety, INFECTIOUS disease transmission, INFORMATION networks, SPORES |
| Abstract (English): |
Wheat scab is one of the most serious diseases that threaten global wheat production. The primary causal agents can be from the fungi Fusarium graminearum and Fusarium asiaticum. The Fusarium spores can be disseminated in the transmission and infection of wheat scab. Consequently, the early and precise identification of pathogen spore concentrations is of paramount importance to prompt the detection of wheat scab, particularly for the high wheat yield and quality, as well as food safety. Conventional techniques, such as microscopic observation and medium isolation, have posed a great challenge to the rapid response and the transmission of large-scale crop disease, due to time-consuming and sophisticated technical expertise. Furthermore, there are missed and erroneous detections of densely distributed small targets in scab spore images. In this study, a small target detection model, YOLOv8-FECA was developed to overcome this challenge. Firstly, a small target detection layer with a feature map size of 160×160 was added to the original YOLOv8 baseline model. The network was then improved to capture the semantic information of small targets, in order to enhance the accuracy of feature description. Secondly, a focus attention mechanism, FECA, was designed to combine the CBAM and ECA modules. This module was then added at the neck end of the network to reduce information loss during downsampling. Ultimately, Wise-IoU Loss and DFL Loss were integrated as the regression loss for the bounding box, in order to enhance the convergence and the precision of bounding box estimation. The experimental results demonstrate that the small target detection layer was added to introduce the focus attention mechanism, compared with the original YOLOv8 baseline model. Wise-IoU Loss and DFL were also introduced after Loss improvement. The average detection accuracy of the new YOLOv8-FECA on spore data set mAP@0.5 and mAP@0.5:0.95 increased by 4.3 percentage points and 6.3 percentage points, respectively, compared with the benchmark model Yolov8n. In terms of performance, the YOLOv8-FECA model was improved in both accuracy and recall, compared with the YOLOv8n. The improved model demonstrated an elevated degree of accuracy and comprehensiveness in identifying the wheat scab spores, with an increase of 2.6 percentage points in Precision and 3.8 percentage points in Recall. There was a decrease in the frame rate per second (FPS) of YOLOv8-FECA, compared with YOLOv8n (from 134 to 106). Nevertheless, a high detection speed was achieved among performance indicators. The YOLOv8-FECA successfully improved the detection accuracy during real-time detection, where mAP @ 0.5 reached 96.8 %. The robustness of the model was verified as well. The results demonstrated that the accurate detection of spores was achieved even in challenging scenarios, such as the high spore density and low light conditions. Moreover, compared with the prevalent target models, including YOLOv5s, YOLOv7-tiny, SSD network, and the exemplar two-stage Faster-R-CNN, YOLOv8-FECA exhibited superior performance, in terms of mAP. The new model was superior in detecting small targets, such as scab spores. This finding can provide technical support to the automatic detection of wheat scab spores in the field, especially for the early warning of scab. [ABSTRACT FROM AUTHOR] |
| Abstract (Chinese): |
针对小麦赤霉病孢子图像中密集分布小目标的漏检错检问题, 该研究设计了一种针对该场景下小目标检测模型 YOLOv8-FECA. 以 YOLOv8 为基准模型, 先添加新的小目标检测层, 以此增强网络对更小目标语义信息的捕捉以及提 高其特征描述的准确性; 其次, 构建融合特征的焦点注意力机制 (focal efficient channel attention module, FECA) 以实 现对小目标和密集分布目标的关注; 再次, 引入 Wise-IoU Loss 与 DFL Loss 结合作为边界框的回归损失, 提高模型的收 敛能力以及对边界框预测的准确性; 最后, 在不同密集场景和不同光线环境下验证了模型的鲁棒性. 结果表明, 优化后 模型 YOLOv8-FECA 相比 YOLOv8 在孢子数据集中的平均检测精度均值 mAP@0.5 提高了 4.3 个百分点, 达到 96.8%, 且改进模型的检测效率达到实时检测要求. 该研究设计的 YOLOv8-FECA 网络模型在保持高性能的同时能够有效提升密 集场景小目标检测的精度, 为其他作物病害孢子的检测提供了思路. [ABSTRACT FROM AUTHOR] |
|
Copyright of Transactions of the Chinese Society of Agricultural Engineering is the property of Chinese Society of Agricultural Engineering and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.) |
| قاعدة البيانات: |
Complementary Index |