基于YOLOv5-SPD改进的杂草识别算法

羊智凡; 李海波

doi:10.12299/jsues.23-0037

基于YOLOv5-SPD改进的杂草识别算法

doi: 10.12299/jsues.23-0037

羊智凡,
李海波^,

上海工程技术大学电子电气工程学院, 上海 201620

详细信息

作者简介:
羊智凡（1997−），男，硕士生，研究方向为计算机视觉。E-mail：2417683036@qq.com

通讯作者:
李海波（1978−），男，讲师，博士，研究方向为计算机视觉、机器学习。E-mail：impart@163.com

中图分类号: S24；TP391.41
计量
- 文章访问数: 30
- HTML全文浏览量: 11
- PDF下载量: 6
- 被引次数: 0
出版历程
- 收稿日期: 2023-02-27
- 刊出日期: 2024-03-30

Improved weed identification algorithm based on YOLOv5-SPD

YANG Zhifan,
LI Haibo^,

School of Electronic and Electrical Engineering, Shanghai University of Engineering Science, Shanghai 201620, China

摘要

摘要: 杂草的精确识别是实现机器代替人工除草的首要前提。初生的杂草目标小，识别难度大。YOLOv5-SPD在小目标识别上有着良好的表现，但在稳健性及准确性上还有待提高。在YOLOv5-SPD基础上加入通道注意力机制可以加强有效特征的权重值，使网络的学习更具有针对性。同时将广义交并比(GIoU)损失函数替换成完全交并比(CIoU)损失函数，可有效解决边框重合关系问题和目标框与预测框的高宽比以及中心点之间的关系，使杂草预测框更加接近真实框。杂草数据集上的试验结果表明，改进后的网络检测精度达到70.3%，准确率达到94.1%，比原来的YOLOv5-SPD分别提高4.7%和2.8%。
- 注意力机制 /
- 边界损失函数 /
- YOLOv5-SPD算法 /
- 杂草识别
Abstract: Accurate identification of weeds is the primary prerequisite for achieving machine replacement of manual weeding. The target of nascent weeds is small, making identification difficult. YOLOv5-SPD has good performance in small target recognition, but its robustness and accuracy still need to be improved. Adding channel attention mechanism on the basis of YOLOv5-SPD can strengthen the weight value of effective features, making the learning of the network more targeted. At the same time, replacing the generalized intersection over union (GIoU) loss function with complete intersection over union (CIoU) can effectively solve the problem of border coincidence, the height width ratio of the target box and the prediction box, and the relationship between the center point, there by making the weed prediction box closer to the real box. The experimental results on the weed dataset show that the improved network detection accuracy reaches 70.3% with an accuracy rate of 94.1% which is 4.7% and 2.8% higher than the original YOLOv5-SPD.
- attention mechanism /
- boundary loss function /
- YOLOv5-SPD algorithm /
- weed identification

HTML全文

图 1 杂草的数据增强

Figure 1. Data enhancement of weeds

下载: 全尺寸图片幻灯片

图 2 数据集目标框分布情况

Figure 2. Distribution of dataset target boxes

下载: 全尺寸图片幻灯片

图 3 YOLOv5-SPD网络结构

Figure 3. YOLOv5-SPD network structure

下载: 全尺寸图片幻灯片

图 4 squeeze-and-excitation注意力机制

Figure 4. squeeze-and-excitation attention mechanism

下载: 全尺寸图片幻灯片

图 5 GIoU图

Figure 5. GIoU map

下载: 全尺寸图片幻灯片

图 6 CIoU图

Figure 6. CIoU map

下载: 全尺寸图片幻灯片

图 7 YOLOv5-SPD-SE网络结构

Figure 7. YOLOv5-SPD-SE network structure

下载: 全尺寸图片幻灯片

图 8 mAP(0.5)对比

Figure 8. mAP (0.5) comparison

下载: 全尺寸图片幻灯片

图 9 损失函数对比

Figure 9. Comparison of loss functions

下载: 全尺寸图片幻灯片

图 10 杂草识别结果

Figure 10. Weed detection results

下载: 全尺寸图片幻灯片

表 1 不同目标检测模型性能对比

Table 1. Performance Comparison of different target detection models

模型	Precision/%	mAP(0.5) /%	Recall /%
Faster R-CNN	83.4	57.9	81.7
CenterNet	87.7	61.3	85.9
YOLOv3	89.5	63.2	87.2
YOLOv5-SPD	91.3	65.6	89.0
本研究改进模型	94.1	70.3	93.0

下载: 导出CSV

表 2 采用不同策略改进的网络之间的对比

Table 2. Comparison between networks improved by different strategies

网络模型	SE	GIoU	CIoU	Precision/%	mAP (0.5) /%	Recall /%
YOLOv5-SPD-GIoU	×	√	×	91.3	65.6	89
YOLOv5-SPD-CIoU	×	×	√	93.4	65.8	90
YOLOv5-SPD-SE-GIoU	√	√	×	93.9	66.2	92
YOLOv5-SPD-SE-CIoU	√	×	√	94.1	70.3	93

下载: 导出CSV

参考文献(16)

[1]	HAMUDA E, MC GINLEY B, GLAVIN M, et al. Automatic crop detection under field conditions using the HSV colour space and morphological operations[J] . Computers and Electronics in Agriculture,2017,133:97 − 107. doi: 10.1016/j.compag.2016.11.021
[2]	SHARPE S M, SCHUMANN A W, BOYD N S. Goosegrass detection in strawberry and tomato using a convolutional neural network[J] . Scientific Reports,2020,10(1):9548. doi: 10.1038/s41598-020-66505-9
[3]	CHO S I, LEE D S, JEONG J Y. AE-automation and emerging technologies: Weed-plant discrimination by machine vision and artificial neural network[J] . Biosystems Engineering,2002,83(3):275 − 280. doi: 10.1006/bioe.2002.0117
[4]	姜红花, 张传银, 张昭, 等. 基于Mask R-CNN的玉米田间杂草检测方法[J] . 农业机械学报,2020,51(6):220 − 228, 247.
[5]	孟庆宽, 张漫, 杨晓霞, 等. 基于轻量卷积结合特征信息融合的玉米幼苗与杂草识别[J] . 农业机械学报,2020,51(12):238 − 245. doi: 10.6041/j.issn.1000-1298.2020.12.026
[6]	东辉, 陈鑫凯, 孙浩, 等. 基于改进YOLOv4和图像处理的蔬菜田杂草检测[J] . 图学学报,2022,43(4):559 − 569.
[7]	JIAO L, ZHANG F, LIU F, et al. A survey of deep learning-based object detection[J] . IEEE Access,2019,7:128837 − 128868. doi: 10.1109/ACCESS.2019.2939201
[8]	SHIH K H, CHIU C T, LIN J A, et al. Real-time object detection with reduced region proposal network via multi-feature concatenation[J] . IEEE Transactions on Neural Networks and Learning Systems,2019,31(6):2164 − 2173.
[9]	HE K, GKIOXARI G, Dollár P, et al. Mask R-CNN[J] . IEEE Transactions on Pattern Analysis and Machine Intelligence,2018,42(2):386 − 397.
[10]	REDMON J, DIVVALA S, GIRSHICK R, et al. You only look once: Unified, real-time object detection[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2016: 779−788.
[11]	WANG Y, WANG C, ZHANG H, et al. Automatic ship detection based on RetinaNet using multi-resolution Gaofen-3 imagery[J] . Remote Sensing,2019,11(5):531. doi: 10.3390/rs11050531
[12]	DUAN K , BAI S , XIE L , et al. CenterNet: Keypoint Triplets for Object Detection[C]// Proceedings of International Conference on Computer Vision. Washington: IEEE Press, 2019: 6569−6578.
[13]	SUNKARA R, LUO T. No more strided convolutions or pooling: a new CNN building block for low-resolution images and small objects[C]//Proceedings of European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases. Grenoble: Springer, 2022: 443−459.
[14]	REZATOFIGHI H, TSOI N, GWAK J Y, et al. Generalized intersection over union: A metric and a loss for bounding box regression[C]//Proceedings of 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Long Beach: IEEE, 2019: 658−666.
[15]	ZHENG Z H, WANG P, LIU W, et al. Distance-IoU loss: Faster and better learning for bounding box regression[C]//Proceedings of the AAAI conference on artificial intelligence. Glasgow: AAAI, 2020: 12993−13000.
[16]	JIE H, LI S, GANG S, et al. Squeeze-and-excitation networks[J] . IEEE Transactions on Pattern Analysis and Machine Intelligence,2017,42(8):2011 − 2023.