Image recognition of Pyrausta nubilalis based on optimized convolutional neural network
-
摘要:目的
随着人工智能和大数据技术的不断发展，针对常规玉米虫害识别方法存在的准确率和效率低等问题，本文提出了一种基于改进GoogLeNet卷积神经网络模型的玉米螟虫害图像识别方法。
方法首先通过迁移学习将GoogLeNet的Inception-v4网络结构知识转移到玉米螟Pyrausta nubilalis虫害识别的任务上，构建模型的训练方式；然后通过数据增强技术对玉米螟虫图像进行样本扩充，得到神经网络训练模型的数据集；同时利用Inception模块拥有多尺度卷积核提取多尺度玉米螟虫害分布特征的能力构建网络模型，并在试验过程中对激活函数、梯度下降算法等模型参数进行优化；最后引入批标准化(BN)操作加速优化模型网络训练，并将该模型运用到玉米螟虫害识别中。
结果基于TensorFlow框架下的试验结果表明，优化后的神经网络算法对玉米螟虫害图像平均识别准确率达到了96.44%。
结论基于优化的卷积神经网络识别模型具有更强的鲁棒性和适用性，可为玉米等农作物虫害识别、智能诊断提供参考。
-
关键词:
- 优化卷积神经网络 /
- Inception-v4模型 /
- TensorFlow框架 /
- 图像识别 /
- 玉米螟
Abstract:ObjectiveWith the continuous development of artificial intelligence and big data technology, aiming at solving the problems of low accuracy and low efficiency in conventional identification methods of corn pest, we proposed a corn borer image identification method based on the improved GoogLeNet convolution-neural network model.
MethodFirstly, through migration learning, the structural knowledge of the Inception-v4 network of GoogLeNet was transferred to the task of corn borer (Pyrausta nubilalis) identification, and the training mode of model construction was established. The data set of neural network training model was obtained by expanding the sample of corn borer image through data enhancement technique. At the same time, the Inception module was used to construct the network model with the ability of multi-scale convolution kernel extraction of the distribution characteristics of multi-scale corn borer, and the activation function, gradient descent algorithm and other model parameters were optimized in the experimental process. Finally, batch normalization (BN) operation was performed to accelerate optimizating model network training, and the model was applied in corn borer identification.
ResultExperimental results of TensorFlow framework showed that the average recognition accuracy of the optimized neural network algorithm for corn borer image was 96.44%.
ConclusionThe convolutional neural network recognition model based on optimization has higher robustness and feasibility, which can provide a reference for identification and intelligent diagnosis of plant pests on corn and other crops.
-
-
![]()
图 2 玉米螟虫原图(a)和经过数据增强和预处理的图像(b、c、d)
Figure 2. Original image (a) and images after data enhancement and preprocessing (b，c，d) of corn borers
![]()
图 5 训练迭代次数对识别准确率和损失函数的影响
Figure 5. Effects of iteration number in training on recognition accuracy and loss function
![]()
图 6 数据增强前后训练迭代次数对识别准确率的影响
Figure 6. Effects of iteration number in training on recognition accuracy before and after data enhancement
表 1 改进前后模型各项参数及识别准确率
Table 1 Various parameters and recognition accuracy of the model before and after improvement
模型
Model模型参数 Model parameter 卷积核大小
Convolution kernel size激活函数
Activation function数据量
Data amount平均识别准确率/%
Average recognition accuracy平均单张图片耗时/s
Average time processing a picture改进前
Before improvement3×3、5×5 ReLu 921 89.17 0.47 改进后
After improvement1×1、1×3、1×3 Sigmoid 2 478 96.44 0.39 
下载: 导出CSV
-
[1] 杨虎. 20世纪中国玉米种业发展研究[D]. 南京: 南京农业大学, 2011. [2] 王连霞. 齐齐哈尔市玉米螟发生规律的演变及应用赤眼蜂防治技术的研究[D]. 北京: 中国农业科学院, 2014. [3] 苏一峰, 杜克明, 李颖, 等. 基于物联网平台的小麦病虫害诊断系统设计初探[J]. 中国农业科技导报, 2016, 18(2): 86-94. [4] 汪京京, 张武, 刘连忠, 等. 农作物病虫害图像识别技术的研究综述[J]. 计算机工程与科学, 2014, 36(7): 1363-1370. doi: 10.3969/j.issn.1007-130X.2014.07.026 [5] 王翔宇, 温皓杰, 李鑫星, 等. 农业主要病害检测与预警技术研究进展分析[J]. 农业机械学报, 2016, 47(9): 266-277. doi: 10.6041/j.issn.1000-1298.2016.09.037 [6] SLADOJEVIC S, ARSENOVIC M, ANDERLA A, et al. Deep neural networks based recognition of plant diseases by leaf image classification[J]. Comput Intel Neurosci, 2016: 3289801. doi: 10.1155/2016/3289801
[7] 李松, 魏中浩, 张冰尘, 等. 深度卷积神经网络在迁移学习模式下的SAR目标识别[J]. 中国科学院大学学报, 2018, 35(1): 75-83. [8] 黄双萍, 孙超, 齐龙, 等. 基于深度卷积神经网络的水稻穗瘟病检测方法[J]. 农业工程学报, 2017, 33(20): 169-176. doi: 10.11975/j.issn.1002-6819.2017.20.021 [9] 王春华, 韩栋. 自适应控制深度学习和知识挖掘图像分类[J]. 沈阳工业大学学报, 2018, 40(3): 334-339. doi: 10.7688/j.issn.1000-1646.2018.03.17 [10] 包晓安, 徐海, 张娜, 等. 基于深度学习的语音识别模型及其在智能家居中的应用[J]. 浙江理工大学学报(自然科学版), 2019, 41(2): 217-223. [11] 戴礼荣, 张仕良, 黄智颖. 基于深度学习的语音识别技术现状与展望[J]. 数据采集与处理, 2017, 32(2): 221-231. [12] 王飞, 陈立, 易绵竹, 等. 新技术驱动的自然语言处理进展[J]. 武汉大学学报(工学版), 2018, 51(8): 669-678. [13] 龙满生, 欧阳春娟, 刘欢, 等. 基于卷积神经网络与迁移学习的油茶病害图像识别[J]. 农业工程学报, 2018, 34(18): 194-201. doi: 10.11975/j.issn.1002-6819.2018.18.024 [14] 陈桂芬, 李静, 陈航, 等. 大数据时代人工智能技术在农业领域的研究进展[J]. 吉林农业大学学报, 2018, 40(4): 502-510. [15] 刘立波, 周国民. 基于多层感知神经网络的水稻叶瘟病识别方法[J]. 农业工程学报, 2009, 25(S2): 213-217. [16] TAN W, ZHAO C, WU H. CNN intelligent early warning for apple skin lesion image acquired by infrared video sensors[J]. High Technology Letters, 2016, 22(1): 67-74.
[17] 孙俊, 谭文军, 毛罕平, 等. 基于改进卷积神经网络的多种植物叶片病害识别[J]. 农业工程学报, 2017, 33(19): 209-215. doi: 10.11975/j.issn.1002-6819.2017.19.027 [18] 梁万杰, 曹宏鑫. 基于卷积神经网络的水稻虫害识别[J]. 江苏农业科学, 2017, 45(20): 241-243. [19] 马浚诚, 杜克明, 郑飞翔, 等. 基于卷积神经网络的温室黄瓜病害识别系统[J]. 农业工程学报, 2018, 34(12): 186-192. doi: 10.11975/j.issn.1002-6819.2018.12.022 [20] 刘永波, 雷波, 曹艳, 等. 基于深度卷积神经网络的玉米病害识别[J]. 中国农学通报, 2018, 34(36): 159-164.
计量
- 文章访问数: 18493
- HTML全文浏览量: 12
- PDF下载量: 20813
