基于改进YOLOv7的农作物虫害识别

doi:10.13304/j.nykjdb.2024.0036

摘要/Abstract

摘要：

为提高农作物虫害检测的准确率和效率，提出了一种基于YOLOv7的农作物虫害识别模型。首先，使用信息聚集-分发机制改进YOLOv7的特征融合模块，增强了不同层级之间的特征融合能力；其次，将损失函数替换为最小点距离交并比来计算边界框回归损失，更好地对齐预测框和真实目标框，提高了边界框回归的准确性；最后，通过在SPPCSPC层后添加感受野增强模块，增强了模型对小尺度害虫的识别能力。实验结果表明，改进后的YOLOv7模型平均准确度为80.4%，精确率为85.3%，召回率为75.1%，较改进前分别提升了3.4%、3.2%、2.6%。该模型对农业害虫具有更好的识别效果和鲁棒性，为农业害虫监测与防治提供了更准确和可靠的工具。

关键词: 图像处理, YOLOv7, 农作物虫害, 目标检测

Abstract:

In order to solve the problem of time-consuming and laborious manual detection of crop pests， a crop pest recognition model based on YOLOv7 was proposed in this paper. Firstly， the feature fusion module of YOLOv7 was improved using the information aggregation-distribution mechanism， which enhanced the feature fusion ability between different levels. Secondly， the loss function was replaced by minimum points distance intersection over union to calculate the boundary box regression loss， which better aligned the predicted box and the real target box， and improved the accuracy of the boundary box regression. Finally， the receptive field enhancement module was added after the SPPCSPC layer to enhance the recognition ability of the model to small-scale pests. Experimental results showed that the average accuracy of the improved YOLOv7 model was 80.4%， the precision rate was 85.3%， and the recall rate was 75.1%， which were 3.4%， 3.2% and 2.6% higher than those before improvement. The model had better recognition effect and robustness for agricultural pests， and provided a more accurate and reliable tool for agricultural pest monitoring and control.

Key words: image processing, YOLOv7, crop pests, object detection

中图分类号:

S126

黄诗锐, 王天一, 文韬, 周江龙. 基于改进YOLOv7的农作物虫害识别[J]. 中国农业科技导报, 2024, 26(11): 107-116.

Shirui HUANG, Tianyi WANG, Tao WEN, Jianglong ZHOU. Crop Insect Identification Based on Improved YOLOv7[J]. Journal of Agricultural Science and Technology, 2024, 26(11): 107-116.

图/表 11

参考文献 26

1	农业农村部办公厅关于印发《2021年全国“虫口夺粮”保丰收行动方案》的通知(农办农〔2021〕4号) [J].中华人民共和国农业部公报,2021,2:48-54.
2	邹修国.基于计算机视觉的农作物病虫害识别研究现状[J].计算机系统应用,2011,20(6):238-242.
	ZOU X G.Research status of crop pests recognition over computer vision [J]. Comput. Syst. Appl., 2011,20(6):238-242.
3	周飞燕,金林鹏,董军.卷积神经网络研究综述[J].计算机学报,2017,40(6):1229-1251.
	ZHOU F Y, JIN L P, DONG J.Review of convolutional neural network [J]. Chin. J. Comput., 2017,40(6):1229-1251.
4	高雄,王海超.甘蓝菜青虫害自动识别系统的开发与试验研究:基于机器视觉[J].农机化研究,2015,37(1):205-208, 222.
	GAO X, WANG H C. Research on cabbage rapae pests automatic recognition system based on machine vision [J]. J. Agric. Mech. Res., 2015,37(1):205-208, 222.
5	苏鸿,温国泉,谢玮,等.基于区域卷积神经网络模型的广西柑橘病虫害识别方法研究[J].西南农业学报,2020,33(4):805-810.
	SU H, WEN G Q, XIE W, et al.. Research on Citrus pest and disease recognition method in Guangxi based on regional convolutional neural network model [J]. Southwest China J. Agric. Sci., 2020,33(4):805-810.
6	ALI F, QAYYUM H, IQBAL M J. Faster-PestNet: a lightweight deep learning framework for crop pest detection and classification [J/OL]. IEEE Access, 2023, 11: 104016-104027 [2024-07-11]. .
7	姜敏,沈一鸣,张敬尧,等.基于深度学习的水稻病虫害诊断方法研究[J].洛阳理工学院学报(自然科学版),2019,29(4):78-83.
	JIANG M, SHEN Y M, ZHANG J Y, et al.. Research on rice diseases and pests diagnosis based on deep learning [J]. J. Luoyang Inst. Sci. Technol. (Nat. Sci.), 2019, 29(4):78-83.
8	郭阳,许贝贝,陈桂鹏,等.基于卷积神经网络的水稻虫害识别方法[J].中国农业科技导报,2021,23(11):99-109.
	GUO Y, XU B B, CHEN G P, et al.. Rice insect pest recognition method based on convolutional neural network [J]. J. Agric. Sci. Technol., 2021,23(11):99-109.
9	施杰,林双双,罗建刚,等.基于YOLO v5s改进模型的玉米作物病虫害检测方法[J].江苏农业科学,2023,51(24):175-183.
10	LIKITH S, REDDY B R, REDDY K S. A smart system for detection and classification of pests using YOLO and CNN techniques [C]// Proceedings of International Conference on Computational Performance Evaluation (ComPE). IEEE, 2021: 049-052.
11	WANG C Y, BOCHKOVSKIY A, LIAO H Y M.YOLOv7:trainable bag-of-freebies sets new state-of-the-art for real-time object detectors [EB/OL].(2022-07-06) [2024-07-11]. .
12	WANG C C, HE W, NIE Y, et al.. Gold-YOLO:efficient object detector via gather-and-distribute mechanism [EB/OL]. (2023-09-20) [2024-07-11]. .
13	YU Z P, HUANG H B, CHEN W J, et al.. YOLO-FaceV2:a scale and occlusion aware face detector [EB/OL].(2022-08-03) [2024-07-11] . .
14	MA S L, XU Y, MA S L, et al.. MPDIoU: a loss for efficient and accurate bounding box regression [J/OL]. (2023-07-14) [2024-07-11]. .
15	WU X, ZHAN C, LAI Y K, et al.. Ip102: a large-scale benchmark dataset for insect pest recognition [C]// Proceedings of 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). IEEE, 2019: 8787-8796.
16	农业农村部.中华人民共和国农业农村部公告第654号[J].中华人民共和国农业部公报,2021,2:48-54.
17	YU F, KOLTUN V. Multi-scale context aggregation by dilated convolutions [EB/OL]. (2015-11-23) [2024-07-11]. .
18	VASWANI A, SHAZEER N, PARMAR N, et al.. Attention is all you need [EB/OL]. (2017-06-12) [2024-07-11]. .
19	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 the IEEE/CVF conference on computer vision and pattern recognition. IEEE, 2019: 658-666.
20	ZHENG Z H, WANG P, LIU W, et al.. Distance-IoU loss: faster and better learning for bounding box regression [J]. Proc. AAAI Conf. Artif. Intell., 2020, 34(7): 12993-13000.
21	ZHANG Y F, REN W, ZHANG Z, et al.. Focal and efficient IOU loss for accurate bounding box regression [EB/OL]. (2021-01-20) [2024-07-11]. .
22	GEVORGYAN Z. SIoU loss: more powerful learning for bounding box regression [EB/OL]. (2022-05-25) [2024-07-11]. .
23	LIU W, ANGUELOV D, ERHAN D, et al.. Ssd: Single shot multibox detector [C]// Proceedings of Computer Vision-ECCV 2016: 14th European Conference. Springer International Publishing, 2016: 21-37.
24	GIRSHICK R. Fast r-cnn [C]// Proceedings of the IEEE international conference on computer vision. IEEE, 2015: 1440-1448.
25	WU W T, LIU H, LI L L, et al.. Application of local fully convolutional neural network combined with YOLO v5 algorithm in small target detection of remote sensing image [J/OL]. PLoS One, 2021, 16(10): e0259283 [2024-07-11]. .
26	TALAAT F M, ZAINELDIN H. An improved fire detection approach based on YOLO-v8 for smart cities [J]. Neural Comput. Appl., 2023, 35(28): 20939-20954.

编号 Number	GD	MPDIoU	RFE	准确率 P/%	召回率 R/%	平均准确度均值mAP/%	参数量 Parameter/M
1	—	—	—	82.1	72.5	77.0	37.2
2	√	—	—	83.9	73.1	78.1	40.3
3	—	√	—	84.4	73.6	78.3	37.2
4	—	—	√	83.7	73.5	77.9	38.5
5	√	√	—	84.6	73.2	78.7	40.3
6	√	—	√	85.1	72.9	79.5	41.6
7	—	√	√	84.2	73.8	78.9	38.5
8	√	√	√	85.3	75.1	80.4	41.6

编号 Number	GD	MPDIoU	RFE	准确率 P/%	召回率 R/%	平均准确度均值mAP/%	参数量 Parameter/M
1	—	—	—	82.1	72.5	77.0	37.2
2	√	—	—	83.9	73.1	78.1	40.3
3	—	√	—	84.4	73.6	78.3	37.2
4	—	—	√	83.7	73.5	77.9	38.5
5	√	√	—	84.6	73.2	78.7	40.3
6	√	—	√	85.1	72.9	79.5	41.6
7	—	√	√	84.2	73.8	78.9	38.5
8	√	√	√	85.3	75.1	80.4	41.6

损失函数 Loss	损失函数计算方法 Calculation method for loss function	准确率 P/%	召回率 R/%	平均准确度均值 mAP/%
广义交并比GIoU	基于两框之间的重叠区域 Based on the overlapping area between 2 boxes	82.9	66.5	74.1
距离交并比DIoU	基于两框之间的中心点距离 Based on the distance between the center points between 2 boxes	78.0	69.4	74.4
增强型交并比EIoU	基于两框之间的中心点距离和长宽比 Based on the center point distance and aspect ratio between 2 boxes	80.8	68.5	75.9
形状交并比SIoU	引入方向性 Introduce directionality	83.6	71.4	75.7
最小点距离交并比MPDIoU	最小化两框间的左上和右下点距离 Minimize the distance between the top left and bottom right points of 2 boxes	84.4	73.6	78.3

损失函数 Loss	损失函数计算方法 Calculation method for loss function	准确率 P/%	召回率 R/%	平均准确度均值 mAP/%
广义交并比GIoU	基于两框之间的重叠区域 Based on the overlapping area between 2 boxes	82.9	66.5	74.1
距离交并比DIoU	基于两框之间的中心点距离 Based on the distance between the center points between 2 boxes	78.0	69.4	74.4
增强型交并比EIoU	基于两框之间的中心点距离和长宽比 Based on the center point distance and aspect ratio between 2 boxes	80.8	68.5	75.9
形状交并比SIoU	引入方向性 Introduce directionality	83.6	71.4	75.7
最小点距离交并比MPDIoU	最小化两框间的左上和右下点距离 Minimize the distance between the top left and bottom right points of 2 boxes	84.4	73.6	78.3

模型 Model	准确率 P/%	召回率 R/%	平均准确度均值 mAP/%	参数量 Parameter/M
SSD	77.9	70..4	73.1	24.5
Faster R-CNN	80.3	71.5	75.2	137.8
YOLOv5	79.9	72.8	75.7	7.0
YOLOv8n	82.6	71.9	76.6	3.0
YOLOv7	82.1	72.5	77	37.2
Ours	85.3	75.1	80.4	41.6