基于改进SSD模型的柑橘叶片病害轻量化检测模型

doi:10.3969/j.issn.1004-1524.20230159

浙江农业学报 ›› 2024, Vol. 36 ›› Issue (3): 662-670.DOI: 10.3969/j.issn.1004-1524.20230159

基于改进SSD模型的柑橘叶片病害轻量化检测模型

李大华¹^,²(), 孔舒¹^,²^,^*(), 李栋¹^,², 于晓¹^,²

1.天津理工大学电气工程与自动化学院,天津 300380
2.天津市新能源电力变换传输与智能控制重点实验室,天津 300380

收稿日期:2023-02-14 出版日期:2024-03-25 发布日期:2024-04-09
作者简介:李大华(1978-),男,天津人,硕士,副教授,研究方向为人工智能、智能微电网等。E-mail:18802218638@163.com
通讯作者: *孔舒,E-mail:kongshu0606@163.com
基金资助:
国家自然科学基金(61502340);天津市自然科学基金(18JCQNJC01000);天津理工大学教学基金项目(YB20-05);天津市复杂系统控制理论与应用重点实验室开放基金项目(TJKL-CATCS-201907)

Lightweight detection model of citrus leaf disease based on improved SSD

LI Dahua¹^,²(), KONG Shu¹^,²^,^*(), LI Dong¹^,², YU Xiao¹^,²

1. College of Electrical and Electronic Engineering, Tianjin University of Technology, Tianjin 300380, China
2. Tianjin Key Laboratory of New Energy Power Conversion, Transmission and Intelligent Control, Tianjin 300380, China

Received:2023-02-14 Online:2024-03-25 Published:2024-04-09

摘要/Abstract

摘要：

针对当前目标检测算法存在模型占比大,对柑橘叶片病害检测速度较慢、精度较低等问题,提出了一种基于改进SSD(single shot multibox detector)的柑橘叶片病害轻量化检测方法。引入了轻量化卷积神经网络MobileNetV2作为SSD网络的骨架,以减小模型规模、提高检测速度。引入感受野模块(receptive field block, RFB)来扩大浅层特征感受野,以提高模型对小目标的检测效果。并引入CA(coordinate attention)注意力机制,以强化不同深度的特征信息,进一步提升柑橘叶片病害的识别精度。结果表明,与VGG16-SSD相比,改进模型(MR-CA-SSD)在柑橘叶片病害检测上平均精度均值(mAP)提升4.4百分点,模型占比减小52.3 MB,每秒检测帧数提升3.15。MR-CA-SSD综合性能也优于YOLOv4、CenterNet、Efficientnet-YoloV3等模型。该改进模型可实现对柑橘叶片病害的快速准确诊断,有助于对病害部位及时精准施药。

关键词: 柑橘,叶片病害, 轻量化网络, 感受野模块, 注意力机制

Abstract:

Aiming at the problems of large model proportion, slow detection speed, and low accuracy in the current target detection algorithm for citrus leaf disease, a lightweight detection method based on improved single shot multibox detector (SSD) for citrus leaf disease was proposed. MobileNetV2, a lightweight convolutional neural network, was introduced as the backbone of the SSD network to reduce the model size and improve the detection speed. The RFB (receptive field block) was introduced into the shallow prediction feature map to expand its receptive field, so as to improve the detection effect of the model on small targets. Additionally, the coordinate attention (CA) was introduced to strengthen feature information at different depths, further enhancing the recognition accuracy of citrus leaf disease. The results showed that compared with the VGG16-SSD network, the improved model (MR-CA-SSD) achieved an mean average precision (mAP) increase of 4.4 percentage points in citrus leaf disease detection, reduced the model proportion by 52.3 MB, and improved the frames per seconds by 3.15. The comprehensive performance of MR-CA-SSD also outperformed algorithms such as YOLOv4, CenterNet, and Efficientnet-YoloV3. This improved model could achieve rapid and accurate diagnosis of citrus leaf disease, contributing to timely and precise pesticide application for diseased areas.

Key words: citrus, leaf disease, lightweight network, receptive field block, attention mechanism

中图分类号:

S431

李大华, 孔舒, 李栋, 于晓. 基于改进SSD模型的柑橘叶片病害轻量化检测模型[J]. 浙江农业学报, 2024, 36(3): 662-670.

LI Dahua, KONG Shu, LI Dong, YU Xiao. Lightweight detection model of citrus leaf disease based on improved SSD[J]. Acta Agriculturae Zhejiangensis, 2024, 36(3): 662-670.

图/表 10

图1 SSD网络模型结构 Conv,卷积;s,步长。

Fig.1 SSD network model structure Conv, Convolution; s, Stride.

图2 MR-CA-SSD网络模型结构图 Conv,卷积;CA,坐标注意力机制;RFB,感受野模块。

Fig.2 MR-CA-SSD network model structure Conv, Convolution; CA, Coordinate attention; RFB, Receptive field block.

图3 倒残差块结构

Fig.3 Inverted residual block structure

图4 CA注意力机制结构 Avg Pool,平均池化;BatchNorm,批量归一化;Non-linear,非线性分类;Sigmoid,激活函数;C是通道数,H和W分别是高度和宽度。

Fig.4 Structure of CA attention mechanism Avg Pool, Average pool; BatchNorm, Batch normalization; Non-linear, Non-linear classification; Sigmoid, Activation function; C, channel number; H, Height; W, Width.

图5 RFB模型结构

Fig.5 RFB model structure

图6 RFB-s模型结构

Fig.6 RFB-s model structure

表1 模型改进前后实验结果对比

Table 1 Results comparison before and after improvement of algorithm

模型 Model	mAP@0.5/%	mAP@0.5:0.95/%	F1值F1 score/%			模型大小 Size/MB	每秒检测帧数 Frames per seconds
模型 Model	mAP@0.5/%	mAP@0.5:0.95/%	炭疽病 Anthracnose	黑素病 Melanosis	褐斑病 Brown spot disease	模型大小 Size/MB	每秒检测帧数 Frames per seconds
SSD	92.7	59.4	57.0	94.0	89.0	91.6	54.65
MR-CA-SSD	97.1	66.5	84.0	97.0	86.0	39.3	57.80

图7 MR-CA-SSD模型和原SSD模型的损失曲线和平均精度均值对比

Fig.7 Comparison of loss curve and mean average precision between MR-CA-SSD and original SSD model

图8 MR-CA-SSD模型和原SSD模型的检测效果 Disease-A、Disease-B和Disease-C分别代表柑橘叶片的炭疽病、黑素病和褐斑病。

Fig.8 Detection effect of MR-CA-SSD and original SSD model Disease-A, Disease-B and Disease-C represent anthracnose, melanosis and brown spot disease of citrus leaf, respectively.

表2 不同模型结果对比

Table 2 Comparison of results from different models

模型 Model	mAP@0.5/%	mAP@0.5:0.95/%	模型大小 Size/MB	每秒检测帧数 Frames per seconds
MR-CA-SSD	97.1	66.5	39.3	57.80
Faster R-CNN	88.1	36.3	108.0	11.24
YOLOv4	73.5	27.3	127.0	66.49
CenterNet	94.9	44.1	124.0	55.74
Efficientnet-YoloV3	92.9	51.2	60.7	35.03
Ghostnet-YoloV4	74.4	29.6	42.5	59.70
YoloV4-tiny	87.3	42.8	22.4	141.12

参考文献 18

[1]	ÇETINER H. Citrus disease detection and classification using based on convolution deep neural network[J]. Microprocessors and Microsystems, 2022, 95: 104687.
[2]	TYAGI A C. Towards a second green revolution[J]. Irrigation and Drainage, 2016, 65(4): 388-389.
[3]	NEUPANE K, BAYSAL-GUREL F. Automatic identification and monitoring of plant diseases using unmanned aerial vehicles: a review[J]. Remote Sensing, 2021, 13(19): 3841.
[4]	DANANJAYAN S, TANG Y, ZHUANG J J, et al. Assessment of state-of-the-art deep learning based citrus disease detection techniques using annotated optical leaf images[J]. Computers and Electronics in Agriculture, 2022, 193: 106658.
[5]	BARMAN U, CHOUDHURY R D, SAHU D, et al. Comparison of convolution neural networks for smartphone image based real time classification of citrus leaf disease[J]. Computers and Electronics in Agriculture, 2020, 177: 105661.
[6]	樊湘鹏, 许燕, 周建平, 等. 基于迁移学习和改进CNN的葡萄叶部病害检测系统[J]. 农业工程学报, 2021, 37(6): 151-159.
	FAN X P, XU Y, ZHOU J P, et al. Detection system for grape leaf diseases based on transfer learning and updated CNN[J]. Transactions of the Chinese Society of Agricultural Engineering, 2021, 37(6): 151-159. (in Chinese with English abstract)
[7]	李鑫然, 李书琴, 刘斌. 基于改进Faster R_CNN的苹果叶片病害检测模型[J]. 计算机工程, 2021, 47(11): 298-304.
	LI X R, LI S Q, LIU B. Apple leaf diseases detection model based on improved Faster R_CNN[J]. Computer Engineering, 2021, 47(11): 298-304. (in Chinese with English abstract)
[8]	SHARIF M, KHAN M A, IQBAL Z, et al. Detection and classification of citrus diseases in agriculture based on optimized weighted segmentation and feature selection[J]. Computers and Electronics in Agriculture, 2018, 150(C): 220-234.
[9]	王云露, 吴杰芳, 兰鹏, 等. 基于改进Faster R-CNN的苹果叶部病害识别方法[J]. 林业工程学报, 2022, 7(1): 153-159.
	WANG Y L, WU J F, LAN P, et al. Apple disease identification using improved Faster R-CNN[J]. Journal of Forestry Engineering, 2022, 7(1): 153-159. (in Chinese with English abstract)
[10]	李辉, 严康华, 景浩, 等. 基于改进SSD的苹果叶部病理检测识别[J]. 传感器与微系统, 2022, 41(10): 134-137.
	LI H, YAN K H, JING H, et al. Apple leaf pathology detection and recognition based on improved SSD[J]. Transducer and Microsystem Technologies, 2022, 41(10): 134-137. (in Chinese with English abstract)
[11]	陈家栋, 雷斌. 基于改进SSD轻量化的交通路口目标检测[J]. 传感器与微系统, 2022, 41(10): 117-121.
	CHEN J D, LEI B. Traffic intersection target detection based on improved SSD lightweight[J]. Transducer and Microsystem Technologies, 2022, 41(10): 117-121. (in Chinese with English abstract)
[12]	LIU W, ANGUELOV D, ERHAN D, et al. SSD: single shot MultiBox detector[M]//Computer Vision-ECCV 2016. Cham: Springer International Publishing, 2016: 21-37.
[13]	汝承印, 张仕海, 张子淼, 等. 基于轻量级MobileNet-SSD和MobileNetV2-DeeplabV3+的绝缘子故障识别方法[J]. 高电压技术, 2022, 48(9): 3670-3679.
	RU C Y, ZHANG S H, ZHANG Z M, et al. Fault identification method for high voltage power grid insulator based on lightweight MobileNet-SSD and MobileNetV2-DeeplabV3+network[J]. High Voltage Engineering, 2022, 48(9): 3670-3679. (in Chinese with English abstract)
[14]	吴天成, 王晓荃, 蔡艺军, 等. 基于特征融合的轻量级SSD目标检测方法[J]. 液晶与显示, 2021, 36(10): 1437-1444.
	WU T C, WANG X Q, CAI Y J, et al. Lightweight SSD object detection method based on feature fusion[J]. Chinese Journal of Liquid Crystals and Displays, 2021, 36(10): 1437-1444. (in Chinese with English abstract)
[15]	SANDLER M, HOWARD A, ZHU M L, et al. MobileNetV2: inverted residuals and linear bottlenecks[C]// 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. June 18-23, 2018. Salt Lake City, UT. IEEE, 2018: 4510-4520.
[16]	HOU Q B, ZHOU D Q, FENG J S. Coordinate attention for efficient mobile network design[C]// 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). June 20-25, 2021. Nashville, TN, USA. IEEE, 2021: 13713-13722.
[17]	LIU S T, HUANG D, WANG Y H. Receptive field block net for accurate and fast object detection[C]// European Conference on Computer Vision. Cham: Springer, 2018: 404-419.
[18]	张岚, 邢博闻, 李彩, 等. 采用改进SSD网络的海参目标检测算法[J]. 农业工程学报, 2022, 38(8): 297-303.
	ZHANG L, XING B W, LI C, et al. Algorithm for detecting sea cucumbers based on improved SSD[J]. Transactions of the Chinese Society of Agricultural Engineering, 2022, 38(8): 297-303. (in Chinese with English abstract)

基于改进SSD模型的柑橘叶片病害轻量化检测模型

Lightweight detection model of citrus leaf disease based on improved SSD

RichHTML

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

图/表 10

参考文献 18

相关文章 5

编辑推荐

Metrics

本文评价

[1]	李颀, 李煜哲. 弱光条件下猪舍清洗目标检测[J]. 浙江农业学报, 2023, 35(9): 2240-2249.
[2]	李荣鹏, 买买提·沙吾提, 盛艳芳, 何旭刚. 基于CA-MobileNet-V2的核桃病害识别与应用[J]. 浙江农业学报, 2023, 35(12): 2977-2987.
[3]	王英允, 龙燕, 杨智优, 黄铝文. 基于改进U-Net网络的苹果叶部病害语义分割方法[J]. 浙江农业学报, 2023, 35(11): 2731-2741.
[4]	周品志, 裴悦琨, 魏冉, 张永飞, 谷宇. 基于YOLOV4模型的果园樱桃实时检测研究[J]. 浙江农业学报, 2022, 34(11): 2522-2532.
[5]	张宁, 吴华瑞, 韩笑, 缪祎晟. 基于多尺度和注意力机制的番茄病害识别方法[J]. 浙江农业学报, 2021, 33(7): 1329-1338.