Lightweight detection of Panax notoginseng disease based on improved SSD algorithm
-
摘要:目的
针对目前三七Panax notoginseng病害识别模型结构复杂、参数庞大，难以实现在移动设备上部署的问题，提出一种基于SSD(Single shot multibox detector)目标检测的改进模型，以期实现三七病害检测的便捷化、快速化与精准化。
方法基于SSD模型架构，采用轻量化卷积神经网络(MobileNet)替换原始特征提取网络(VGG16)，降低主干网络的参数量与计算量，同时根据人类视觉皮层中群智感受野(pRF)的大小与其视网膜图中偏心率之间的函数关系，构建RFB模块，用该模块替换原SSD模型框架顶部卷积层，从而增强网络深层特征，提高轻量化模型的检测精度与检测速度，实现多尺度三七病害检测。
结果与SSD模型相比，RFB-MobileNet-SSD模型网络参数量和参数计算量分别降低了96.67%和96.10%。在不同天气条件下应用模型对4种不同病害数据进行验证发现，改进模型的准确率提高了4.6个百分点，召回率提高了6.1个百分点，F1精度提高了5.4个百分点，单幅图像检测时间由SSD模型的0.073 s缩短为0.020 s，尺寸仅为SSD模型的54.6%。
结论改进后模型不仅能够满足三七叶片病害实时检测的要求，且更利于在移动设备中部署。此外，RFB-MobileNet-SSD对于小区域病害检测能力表现更优且在复杂环境下抗干扰能力更强，更适合田间环境下的三七病害检测。
Abstract:ObjectiveAiming to address the problem of the current Panax notoginseng disease identification model, with complex structure and large number of parameters, hindering deployment on mobile devices, an improved model based on single shot multibox detector (SSD) target detection is proposed to enable convenient, fast and accurate P. notoginseng disease detection.
MethodBased on the SSD model architecture, the original feature extraction network (VGG16) was replaced by a lightweight convolutional neural network (MobileNet) to reduce the number of parameters and computation amount of the backbone network. Meanwhile, the RFB module was constructed based on the functional relationship between the size of the population wise receptive field (pRF) in human visual cortex and its eccentricity in the retinogram. The top convolutional layer of the original SSD model framework was replaced by the RFB module to enhance the deep features of the network, improve the detection accuracy and detection speed of the lightweight model, and enable multi-scale P. notoginseng disease detection.
ResultCompared with the SSD model, the RFB-MobileNet-SSD model reduced the number of network parameters and the computation amount by 96.67% and 96.10% respectively. The model validation using four different disease data under different weather conditions revealed that the improved model improved the accuracy by 4.6 percentage point, recall by 6.1 percentage point, F1 accuracy by 5.4 percentage point, and the time of single image detection was shortened from 0.073 s of the SSD model to 0.020 s, and the size was only 54.6% of the SSD model.
ConclusionThe improved model not only meets the purpose of real-time detection ofP. notoginseng leaf diseases, but is also more convenient for deployment in mobile devices. Moreover, RFB-MobileNet-SSD shows improved performance for small area disease detection and is more resistant to interference in complex environments, making it more suitable for P. notoginseng disease detection in field environment.
-
Keywords:
- Panax notoginseng /
- Disease identification /
- SSD /
- Image recognition /
- Deep learning /
- Lightweight
-
![]()
图 6 置信度对比图
图中的1、2、3、4分别代表灰霉病、病毒病、白粉病、圆斑病；其余数字表示置信限
Figure 6. Confidence comparison image
1, 2, 3 and 4 in the figure indicate Botrytis cinerea, virus disease, powdery mildew and round spot respectively; The rest numbers represent confidence
![]()
图 7 鲁棒性对比图
图中的1、2、3、4分别代表灰霉病、病毒病、白粉病、圆斑病；其余数字表示置信限
Figure 7. Robustness comparison image
1, 2, 3 and 4 in the figure indicate Botrytis cinerea, virus disease, powdery mildew and round spot respectively; The rest numbers represent confidence
表 1 消融试验性能指标对比
Table 1 Comparison of performance indicators of ablation experiments
网络设置 Network setting 准确率/% Precision 召回率/% Recall F1精度/% F1 accuracy mAP/% 模型尺寸/MB Model size SSD 82.3 75.8 78.9 77.76 126.7 MobileNet-SSD 80.8 72.4 76.4 75.05 69.0 RFB-SSD 87.2 83.1 85.1 83.05 126.8 RFB-MobileNet-SSD 86.9 81.9 84.3 81.94 69.2 
下载: 导出CSV
表 2 不同模型检测性能对比
Table 2 Comparison of detection performance of different models
准确率/% Precision mAP/% 单幅图像检测时间/s Single image detection time 模型尺寸/MB Model size 模型 Model 灰霉病 Botrytis cinerea 病毒病 Virus disease 白粉病 Powdery mildew 圆斑病 Round spot SSD 81.31 71.54 74.01 84.19 77.8 0.073 126.7 MobileNet-SSD 78.71 67.43 71.25 82.81 75.1 0.021 69.0 RFB-SSD 87.55 76.12 79.26 89.27 83.1 0.071 126.8 RFB-MobileNet-SSD 86.05 75.97 77.53 88.21 82.0 0.020 69.2 YOLOv3 84.41 70.76 68.61 86.77 77.6 0.042 112.3 Mask R-CNN 90.48 80.39 83.98 90.29 86.3 2.139 211.8
下载: 导出CSV
-
[1] 蔡琳, 彭鹏. 三七药理作用的研究进展[J]. 山东化工, 2021, 50(3): 70-71. doi: 10.3969/j.issn.1008-021X.2021.03.029 [2] CHUNG A S, PARK K M. Anticancer and antineurode generative effects of ginsenosides[M]//Studies in natural products chemistry. Amsterdam: Elsevier, 2016: 131-158.
[3] 夏鹏国, 崔秀明, 韦美膛, 等. 三七的生物学特性研究进展[J]. 中药材, 2012, 35(5): 831-835. doi: 10.13863/j.issn1001-4454.2012.05.027 [4] 李宏伟, 郭富贵, 刘洋, 等. 三七病虫害研究现状与展望[J]. 草业科学, 2019, 36(5): 1415-1427. [5] 蒋妮, 覃柳燕, 叶云峰. 三七病害研究进展[J]. 南方农业学报, 2011, 42(9): 1070-1074. doi: 10.3969/j.issn.2095-1191.2011.09.010 [6] 杨涛, 陈昱君, 段承俐, 等. 三七黑斑病抗性人工鉴定方法的建立[J]. 云南农业大学学报, 2006, 21(5): 549-553. [7] 王志敏, 皮自聪, 罗万东, 等. 三七圆斑病和黑斑病及其防治[J]. 农业与技术, 2016, 36(1): 49-51. [8] 李欣. 三七主要病害病原菌的快速检测及叶部病害病原菌的鉴定[D]. 昆明: 昆明理工大学, 2020. [9] 彭勤, 高士, 彭佳红. 基于 SVM 的豆类作物病害识别研究[J]. 电脑知识与技术, 2019, 15(24): 210-211. [10] 谭秦红. 基于无人机图像处理的大豆叶片病害识别准确率研究[J]. 河南农业科学, 2021, 50(3): 174-180. doi: 10.15933/j.cnki.1004-3268.2021.03.023 [11] FELZENSZWALB P, MCALLESTER D, RAMANAN D. A discriminatively trained, multiscale, deformable part model[C]//2008 IEEE Conference on Computer Vision and Pattern Recognition. Anchorage, AK, USA: IEEE, 2008: 1-8.
[12] BI C, CHEN G F. Bayesian networks modeling for crop diseases[C]//International Conference on Computer and Computing Technologies in Agriculture. Berlin, Heidelberg: Springer, 2010: 312-320.
[13] 李旭冬, 叶茂, 李涛. 基于卷积神经网络的目标检测研究综述[J]. 计算机应用研究, 2017, 34(10): 2881-2886. doi: 10.3969/j.issn.1001-3695.2017.10.001 [14] LECUN Y, BENGIO Y, HINTON G. Deep learning[J]. Nature, 2015, 521(7553): 436-444. doi: 10.1038/nature14539
[15] 林建吾, 张欣, 陈孝玉龙, 等. 基于轻量化卷积神经网络的番茄病害图像识别[J]. 无线电工程, 2022, 52(8): 1347-1353. doi: 10.3969/j.issn.1003-3106.2022.08.007 [16] 刘合兵, 鲁笛, 席磊. 基于MobileNetV2和迁移学习的玉米病害识别研究[J]. 河南农业大学学报, 2022, 56(6): 1041-1051. doi: 10.16445/j.cnki.1000-2340.20220421.001 [17] 周巧黎, 马丽, 曹丽英, 等. 基于改进轻量级卷积神经网络MobileNetV3的番茄叶片病害识别[J]. 智慧农业(中英文), 2022, 4(1): 47-56. [18] 王远志, 施子珍, 张艳红. 改进的Faster R-CNN用于苹果叶部病害检测[J]. 安庆师范大学学报(自然科学版), 2022, 28(2): 26-30. [19] 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. [20] 李就好, 林乐坚, 田凯, 等. 改进Faster R-CNN的田间苦瓜叶部病害检测[J]. 农业工程学报, 2020, 36(12): 179-185. doi: 10.11975/j.issn.1002-6819.2020.12.022 [21] 王林柏, 张博, 姚竟发, 等. 基于卷积神经网络马铃薯叶片病害识别和病斑检测[J]. 中国农机化学报, 2021, 42(11): 122-129. doi: 10.13733/j.jcam.issn.2095-5553.2021.11.19 [22] 张文静, 孙秀朋, 乔永亮, 等. 基于InceptionV3的烟草病害识别[J]. 中国烟草学报, 2021, 27(5): 61-70. doi: 10.16472/j.chinatobacco.2021.T0061 [23] 周长建, 宋佳, 向文胜. 基于人工智能的作物病害识别研究进展[J]. 植物保护学报, 2022, 49(1): 316-324. doi: 10.13802/j.cnki.zwbhxb.2022.2022803 [24] LIU W, ANGUELOV D, ERHAN D, et al. SSD: Single shot multibox detector[C]//European Conference on Computer Vision. Cham: Springer, 2016: 21-37.
