基于YOLOv5改进模型的杂交稻芽种快速分级检测

钟海敏; 马旭; 李泽华; 王曦成; 刘赛赛; 刘伟文; 王承恩; 林泳达

doi:10.7671/j.issn.1001-411X.202209015

基于YOLOv5改进模型的杂交稻芽种快速分级检测

Rapid grading detection on hybrid rice bud seeds based on improved YOLOv5 model

摘要

摘要:
目的提高杂交稻种子活力分级检测精度和速度。
方法提出了一种基于YOLOv5改进模型(YOLOv5-I)的杂交稻芽种快速分级检测方法，该方法引入SE (Squeeze-and-excitation)注意力机制模块以提高目标通道的特征提取能力，并采用CIoU损失函数策略以提高模型的收敛速度。
结果 YOLOv5-I算法能有效实现杂交稻芽种快速分级检测，检测精度和准确率高，检测速度快。在测试集上，YOLOv5-I算法目标检测的平均精度为97.52%，平均检测时间为3.745 ms，模型占用内存空间小，仅为13.7 MB；YOLOv5-I算法的检测精度和速度均优于YOLOv5s、Faster-RCNN、YOLOv4和SSD模型。
结论 YOLOv5-I算法优于现有的算法，提升了检测精度和速度，能够满足杂交稻芽种分级检测的实用要求。

Abstract:
Objective In order to improve the grading detection accuracy and speed of hybrid rice seed vigor.
Method A rapid grading detection method for hybrid rice bud seeds named YOLOv5-I model, which was an improved model based on YOLOv5, was proposed. The feature extraction ability of the target channel of YOLOv5-I model was improved by introducing the SE (Squeeze-and-excitation) attention mechanism module, and a CIoU loss function strategy was adopted to improve the convergence speed of this model.
Result The YOLOv5-I algorithm effectively achieved the rapid grading detection of hybrid rice bud seeds, with high detection accuracy and speed. In the test set, the average accuracy of the YOLOv5-I model was 97.52%, the average detection time of each image was 3.745 ms, and the memory space occupied by the YOLOv5-I model was small with 13.7 MB. The detection accuracy and speed of YOLOv5-I model was better than those of YOLOv5s, Faster-RCNN, YOLOv4 and SSD models.
Conclusion The YOLOv5-I algorithm is better than existing algorithms, improves detection accuracy and speed, and can meet the practical requirement for grading detection of hybrid rice bud seeds.

HTML全文

参考文献(24)

施引文献

资源附件(0)