吉林大学学报(工学版) ›› 2021, Vol. 51 ›› Issue (3): 1131-1139.doi: 10.13229/j.cnki.jdxbgxb20200116

• 农业工程·仿生工程 • 上一篇    

基于深度学习的轻量化田间昆虫识别及分类模型

袁哲明1,2(),袁鸿杰1,2,言雨璇2,李钎2,3,刘双清4,谭泗桥1,2()   

  1. 1.湖南农业大学 湖南省农业大数据分析与决策工程技术研究中心，长沙 410128
    2.湖南农业大学 湖南省农村农业信息化工程技术研究中心，长沙 410128
    3.湖南农业大学 商学院，长沙 410128
    4.湖南农业大学 植物保护学院，长沙 410128
  • 收稿日期:2020-02-25 出版日期:2021-05-01 发布日期:2021-05-07
  • 通讯作者: 谭泗桥 E-mail:zhmyuan@sina.com;tsq@hunau.edu.cn
  • 作者简介:袁哲明（1971-），男，教授，博士生导师. 研究方向：生物信息学，复杂数据分析. E-mail:zhmyuan@sina.com
  • 基金资助:
    国家自然科学基金项目(31772157);中央引导地方科技发展专项资金项目(2019XF5015);湖南省教育厅科学研究项目(17C0757)

Automatic recognition and classification of field insects based on lightweight deep learning model

Zhe-ming YUAN1,2(),Hong-jie YUAN1,2,Yu-xuan YAN2,Qian LI2,3,Shuang-qing LIU4,Si-qiao TAN1,2()   

  1. 1.Hunan Engineering & Technology Research Center for Agricultural Big Data Analysis & Decision-making，Hunan Agricultural University，Changsha 410128，China
    2.Hunan Engineer Research Center for Information Technology in Agriculture，Hunan Agricultural University，Changsha 410128，China
    3.Business School，Hunan Agricultural University，Changsha 410128，China
    4.College of Plant Protection，Hunan Agricultural University，Changsha 410128，China
  • Received:2020-02-25 Online:2021-05-01 Published:2021-05-07
  • Contact: Si-qiao TAN E-mail:zhmyuan@sina.com;tsq@hunau.edu.cn

RICH HTML

 22   

PDF (PC)

335

摘要：

由于田间昆虫环境的复杂性、昆虫类别间样本数量的不均衡性，现有田间昆虫自动识别和分类方法存在误识率高、效率低等缺点。本文基于轻量化深度学习模型提出了新的田间昆虫自动识别和分类算法。首先，对田间昆虫图像进行预处理，将其输入到轻量化算法中进行特征提取，通过多尺度特征融合输出不同尺寸的预测网络。然后，引入联合交并比进行田间昆虫自动识别和分类。最后，与其他算法进行了仿真对比实验，结果表明，本文算法的田间昆虫自动识别和分类正确率高、用时少、鲁棒性强，有效解决了昆虫堆积、背景干扰等问题，可实时、在线识别田间昆虫。

关键词: 模式识别, 深度学习, 目标检测, 昆虫分类, 轻量化算法, 图像预处理

Abstract:

Due to the complexity of the insect environment in the field and the imbalance in the number of samples among insect categories， the existing automatic identification and classification methods for field insects have high misidentification rates and low efficiency. In this paper， a new field insect automatic identification and classification algorithm is developed based on a lightweight deep learning model. First， preprocessing applied to the picture， then those images were input to the lightweight algorithm for feature extraction， and multi-scale feature fusion were adopted to output prediction networks of different sizes； then introduce joint cross-comparison for automatic identification and classification of field insects， and finally compare with the reference The algorithm has been simulated and compared. The results show that the field insect automatic identification and classification of the algorithm in this paper has high accuracy， less time， and strong robustness. It effectively solves the problems of insect accumulation and background interference and can identify field insects in real-time and online.

Key words: pattern recognition, deep learning, object detection, insect classification, lightweight algorithm, image preprocessing

中图分类号: 

  • TP181

图1

轻量化昆虫检测与分类算法架构"

图2

深度可分离卷积原理"

表1

特征提取网络参数"

类型/步长卷积核尺寸输出尺寸
Conv/s23*3*3*32304*240*32
Conv_dw/s13*3*32304*240*32
Conv_dw/s11*1*32*64304*240*64
Conv_dw/s23*3*64152*120*64
Conv_dw/s11*1*64*128152*120*128
Conv_dw/s13*3*128152*120*128
Conv_dw/s11*1*128*128152*120*128
Conv_dw/s23*3*12876*60*128
Conv_dw/s11*1*128*25676*60*256
Conv_dw/s13*3*25676*60*256
Conv_dw/s11*1*256*25676*60*256
Conv_dw/s23*3*25638*30*256
Conv_dw/s11*1*256*51238*30*512
5×Conv_dw/s13*3*51238*30*512
Conv_dw/s11*1*512*51238*30*512
Conv_dw/s23*3*51219*15*512
Conv_dw/s11*1*512*102419*15*1024
Conv_dw/s13*3*102419*15*1024
Conv_dw/s11*1*1024*102419*15*1024

图3

IoU和GIoU的预测结果对比"

图4

部分昆虫训练集样本"

图5

各分类样本样本"

图6

训练损失函数曲线"

表2

本文目标检测模型对各测试集中各分类昆虫的预测效果"

昆虫类别AP/%
均值70.98
Chilo suppressalis70.69
Lepidoptera64.96
Hydrophilidae latreille87.38
Coleoptera71.02
Gryllotalpidae orientalis87.71
Naranga aenescens53.82
Kirkaldyia deyrollei62.04
Cicadellidae70.21

图7

不同场景下的模型检测结果"

图8

模型特征提取层可视化"

表3

不同种模型对昆虫的平均识别精度、检测速度和模型尺寸"

模型mAP/%检查速度/ms参数数量/ 百万模型体积/MB
YOLOv369.275361236
YOLOv3-TINY52.5132933
Faster R-CNN68.19325136532
本文70.9852830

图9

本文算法与YOLOv3算法预测结果对比"

1 刘海启. 以精准农业驱动农业现代化加速现代农业数字化转型[J]. 中国农业资源与区划,2019,40(1):1-6, 73.
Liu Hai-qi. Driving agricultural modernization with precision agriculture and accelerating digital transformation of modern agriculture[J]. Chinese Journal of Agricultural Resources and Regional Planning, 2019,40(1):1-6, 73.
2 聂娟,孙瑞志,邓雪峰,等. 基于数据世系管理的精准农业不确定性复杂事件处理[J].农业机械学报, 2016, 47(5):245-253.
Nie Juan, Sun Rui-zhi, Deng Xue-feng, et al. Uncertainty complex event processing of precision agriculture based on data lineage management[J]. Transactions of the Chinese Society for Agricultural Machinery, 2016,47(5):245-253.
3 封洪强,姚青. 农业害虫自动识别与监测技术[J].植物保护, 2018,44(5):127-133, 198.
Feng Hong-qiang, Yao Qing. Automatic identification and monitoring technology of agricultural pests[J]. Plant Protection, 2018, 44(5):127-133, 198.
4 张永玲,姜梦洲,俞佩仕,等. 基于多特征融合和稀疏表示的农业害虫图像识别方法[J]. 中国农业科学, 2018, 51(11):2084-2093.
Zhang Yong-ling, Jiang Meng-zhou, Yu Pei-shi, et al. Image recognition of agricultural pests based on multi feature fusion and sparse representation [J]. China Agricultural Science, 2018,51(11):2084-2093.
5 马子骥,卢浩,董艳茹. 双通道单图像超分辨率卷积神经网络[J].吉林大学学报:工学版,2019,49(6):2089-2097.
Ma Zi-ji, Lu Hao, Dong Yan-ru. Dual channel single image super-resolution convolution neural network[J]. Journal of Jilin University(Engineering and Technology Edition), 2019,49(6): 2089-2097.
6 周爱明,马鹏鹏,席天宇,等. 基于深度学习的蝴蝶科级标本图像自动识别[J]. 昆虫学报,2017,60(11):1339-1348.
Zhou Ai-ming, Ma Peng-peng, Xi Tian-yu, et al. Automatic image recognition of butterflies based on deep learning[J]. Acta Entomology, 2017,60(11):1339-1348.
7 Cheng Xi, Zhang You-hua, Chen Ju-qiong, et al. Pest identification via deep residual learning in complex background[J]. Computers and Electronics in Agriculture,2017,141:351-356.
8 张银松,赵银娣,袁慕策. 基于改进Faster-RCNN算法的粘虫板图像昆虫识别与计数[J]. 中国农业大学学报, 2019,24(5):115-122.
Zhang Yin-song, Zhao Yin-di, Yuan Mu-ce. Insect recognition and counting based on improved FAST-RCNN model in the image of armyworm board[J]. Journal of China Agricultural University, 2019,24(5):115-122.
9 郭继昌,吴洁,郭春乐,等. 基于残差连接卷积神经网络的图像超分辨率重构[J]. 吉林大学学报:工学版, 2019, 49(5):1726-1734.
Guo Ji-chang, Wu Jie, Guo Chun-le, et al. Image super-resolution reconstruction based on residual connected convolutional neural network[J]. Journal of Jilin University(Engineering and Technology Edition), 2019,49(5):1726-1734.
10 Huang J, Rathod V, Sun C, et al. Speed/accuracy trade-offs for modern convolutional object detectors[C]∥IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Honolulu, HI, 2017, 3296-3297.
11 王毅,刘波,熊龙烨,等. 基于深度学习的果园道路导航线生成算法研究[J]. 湖南农业大学学报:自然科学版,2019,45(6):674-678.
Wang Yi, Liu Bo, Xiong Long-ye, et al. Research on algorithm of orchard road navigation line generation based on deep learning[J]. Journal of Hunan Agricultural University(Natural Sciences), 2019,45(6):674-678.
12 吴天舒,张志佳,刘云鹏,等. 基于改进SSD的轻量化小目标检测算法[J]. 红外与激光工程,2018,47(7):47-53.
Wu Tian-shu, Zhang Zhi-jia, Liu Yun-peng, et al. Lightweight small target detection algorithm based on improved SSD [J]. Infrared and Laser Engineering, 2018,47(7): 47-53.
13 于长东,毕晓君,韩阳,等. 基于轻量化深度学习模型的粒子图像测速研究[J]. 光学学报, 2020, 40(1):1-15.
Yu Chang-dong, Bi Xiao-jun, Han Yang, et al. Particle image velocimetry based on lightweight deep learning model[J]. Acta optica Sinica, 2020, 40(1):1-15.
14 吴天舒,张志佳,刘云鹏,等. 结合YOLO检测和语义分割的驾驶员安全带检测[J]. 计算机辅助设计与图形学学报,2019,31(1):126-131.
Wu Tian-shu, Zhang Zhi-jia, Liu Yun-peng, et al. Driver safety belt detection combined with YOLO detection and semantic segmentation[J]. Journal of Computer-Aided Design & Computer Graphics, 2019,31(1):126-131.
15 杨观赐,杨静,苏志东,等. 改进的YOLO特征提取算法及其在服务机器人隐私情境检测中的应用[J]. 自动化学报,2018,44(12):2238-2249.
Yang Guan-ci, Yang Jing, Su Zhi-dong, et al. Improved Yolo feature extraction algorithm and its application in privacy situation detection of service robots[J]. Acta Automatica Sinica, 2018,44(12):2238-2249.
16 彭明霞,夏俊芳,彭辉. 融合FPN的Faster R-CNN复杂背景下棉田杂草高效识别方法[J]. 农业工程学报,2019,35(20):202-209.
Peng Ming-xia, Xia Jun-fang, Peng Hui. Fast R-CNN based on FPN for efficient weed identification in cotton field[J]. Journal of Agricultural Engineering, 2019, 35(20): 202-209.
17 Fu Yi-hao, Aldrich C. Froth image analysis by use of transfer learning and convolutional neural networks[J]. Minerals Engineering, 2018, 115:68-78.
18 Ren S, He K, Girshick R, et al. Faster R-CNN: towards real-time object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2017, 39(6):1137-1149.
19 Peng Chao, Xiao Te-te, Li Ze-ming, et al. MegDet: a large mini-batch object detector[C]∥IEEE Conference on Computer Vision and Pattern Recognition, UTAH, USA, 2018:6181-6189.
[1] 宋震,李俊良,刘贵强. 基于深度学习和限幅模糊的变转速液压动力源恒流量预测方法[J]. 吉林大学学报(工学版), 2021, 51(3): 1106-1110.
[2] 李锦青,周健,底晓强. 基于循环生成对抗网络的学习型光学图像加密方案[J]. 吉林大学学报(工学版), 2021, 51(3): 1060-1066.
[3] 徐涛,马克,刘才华. 基于深度学习的行人多目标跟踪方法[J]. 吉林大学学报(工学版), 2021, 51(1): 27-38.
[4] 刘富,刘璐,侯涛,刘云. 基于优化MSR的夜间道路图像增强方法[J]. 吉林大学学报(工学版), 2021, 51(1): 323-330.
[5] 赵宏伟,刘晓涵,张媛,范丽丽,龙曼丽,臧雪柏. 基于关键点注意力和通道注意力的服装分类算法[J]. 吉林大学学报(工学版), 2020, 50(5): 1765-1770.
[6] 谌华,郭伟,闫敬文,卓文浩,吴良斌. 基于深度学习的SAR图像道路识别新方法[J]. 吉林大学学报(工学版), 2020, 50(5): 1778-1787.
[7] 郜峰利,陶敏,李雪妍,何昕,杨帆,王卓,宋俊峰,佟丹. 基于深度学习的CT影像脑卒中精准分割[J]. 吉林大学学报(工学版), 2020, 50(2): 678-684.
[8] 车翔玖,刘华罗,邵庆彬. 基于Fast RCNN改进的布匹瑕疵识别算法[J]. 吉林大学学报(工学版), 2019, 49(6): 2038-2044.
[9] 徐谦,李颖,王刚. 基于深度学习的行人和车辆检测[J]. 吉林大学学报(工学版), 2019, 49(5): 1661-1667.
[10] 翟凤文,党建武,王阳萍,金静,罗维薇. 基于扩展轮廓的快速仿射不变特征提取[J]. 吉林大学学报(工学版), 2019, 49(4): 1345-1356.
[11] 郭立民,陈鑫,陈涛. 基于AlexNet模型的雷达信号调制类型识别[J]. 吉林大学学报(工学版), 2019, 49(3): 1000-1008.
[12] 黄勇,杨德运,乔赛,慕振国. 高分辨合成孔径雷达图像的耦合传统恒虚警目标检测[J]. 吉林大学学报(工学版), 2018, 48(6): 1904-1909.
[13] 刘富, 兰旭腾, 侯涛, 康冰, 刘云, 林彩霞. 基于优化k-mer频率的宏基因组聚类方法[J]. 吉林大学学报(工学版), 2018, 48(5): 1593-1599.
[14] 杨超宇, 李策, 梁胤程, 杨峰. 基于改进粒子滤波的煤矿视频监控模糊目标检测[J]. 吉林大学学报(工学版), 2017, 47(6): 1976-1985.
[15] 姜宏, 李垠, 吕巍. 基于线性收缩的大阵列MIMO雷达目标盲检测[J]. 吉林大学学报(工学版), 2017, 47(3): 973-980.
Viewed
Full text


Abstract

Cited
  Shared   
  Discussed   
No Suggested Reading articles found!
版权所有 © 《吉林大学学报(工学版)》编辑部
地址：长春市人民大街5988号　电话：+86-431-85095297　传真：+86-431-85094074　E-mail: xbgxb@jlu.edu.cn
本系统由北京玛格泰克科技发展有限公司设计开发