摘要
为了降低葡萄果园的管理成本,及时发现并预防葡萄病害,文章提出了一种基于改进YOLOv4模型的葡萄叶片病害检测与识别算法。该算法对传统YOLOv4模型进行了改进,针对细粒度、多尺度的葡萄叶片早期疾病检测优化了检测速度和准确性,并应用于真实环境中的实时检测。在检测时间为18.31 ms时,该检测模型的平均准确率(mAP)和F1得分分别达到90.4%和94.8%。总体检测结果表明,当前算法的性能显著优于现有的检测模型,精度提高了7.8%,F1分数提高了6.6%。该模型可作为一种检测复杂现实情景下葡萄叶片病害的有效方法。
In order to reduce the management cost of grape orchards and timely detect and prevent grape diseases,this paper proposes a grape leaf disease detection and recognition algorithm based on an improved YOLOv4 model.The traditional YOLOv4 model has been improved to optimize detection speed and accuracy for fine-grained and multi-scale early disease detection of grape leaves,and applied to real-time detection in real environments.At a detection time of 18.31 ms,the average accuracy(mAP)and F1-score of the detection model reached 90.4%and 94.8%,respectively.The overall detection results indicate that the current algorithm performs significantly better than existing detection models,with an accuracy improvement of 7.8%and an F1-score improvement of 6.6%.This model can serve as an effective method for detecting grape leaf diseases in complex real-world scenarios.
作者
金彬
解祥新
Jin Bin;Xie Xiangxin(Computer and Information Engineering Department,Nantong Institute of Technology,Nantong 226002,China)
出处
《无线互联科技》
2023年第18期129-132,共4页
Wireless Internet Technology
关键词
实时目标检测
葡萄叶病
卷积神经网络
计算机视觉
real-time object detection
grape leaf disease
convolutional neural network
computer vision