Identification of citrus fruit infected with Huanglongbing based on Mixup algorithm and convolutional neural network
-
摘要:目的
解决传统柑橘黄龙病果实图像识别方法依赖人工设计特征、费时费力、网络模型参数量大、识别准确率低等问题。
方法首先，采集柑橘黄龙病的果实图像，并对其进行翻转、旋转、仿射、高斯扰动等数据扩增；采用Mixup算法建立样本之间的线性关系，增强模型识别数据样本的鲁棒性；然后，迁移Xception网络在ImageNet数据集上的先验知识，提出一种基于Mixup算法和卷积神经网络的柑橘黄龙病果实识别模型−X-ResNeXt模型；最后，采用动量梯度下降优化方法，有效地减缓震荡影响，并且有效地加速模型向局部最优点收敛。
结果采用数据扩增数据集训练的X-ResNeXt模型准确率可以达到91.38%；在进行迁移学习优化后，训练时间减少了432 s，准确率提升为91.97%；结合Mixup混类数据增强进一步训练，模型准确率提升为93.74%；最后，利用动量梯度下降方法进行模型收敛优化，最终模型的准确率达到94.29%，比Inception-V3和Xception网络分别提高了3.98%和1.51%。
结论在数据量较少情况下，降低模型复杂度并迁移已有先验知识，有助于模型性能提升；Mixup混类数据增强方法有利于提高模型识别柑橘黄龙病果实图像样本的适应性，提升柑橘黄龙病果实识别模型性能；X-ResNeXt模型在准确率与召回率指标上优于经典识别模型，可为柑橘黄龙病的高精度、快速无损识别提供参考。
-
关键词:
- 柑橘黄龙病 /
- Mixup算法 /
- 梯度下降 /
- 卷积神经网络 /
- Xception网络
Abstract:ObjectiveThe traditional image recognition method relies on manual design features, is time-consuming and labor-intensive, has large number of network model parameter and has low recognition accuracy rate. The goal was to solve these problems in traditional method for identifying citrus fruit infected with Huanglongbing.
MethodFirstly, we collected the images of citrus fruit with Huanglongbing, and performed data enhancement modes such as flip, rotation, affine, and Gaussian disturbance. Further, we used the Mixup algorithm to establish a linear relationship between samples to enhance the robustness of the model for identifying data samples. Then, we transfered the prior knowledge on the ImageNet data set of Xception network, and proposed a citrus Huanglongbing fruit recognition model of X-ResNeXt model based on Mixup algorithm and convolutional neural network. Finally, the momentum gradient descent optimization method was used to reduce the impact of shocks and effectively accelerate the convergence of the model to the local optimum.
ResultThe accuracy rate of the X-ResNeXt model trained on the data set after data enhancement was 91.38%. After optimization with transfer learning, the training time reduced by 432 s, and the accuracy rate of the model increased to 91.97%. Combined with the enhancement of Mixup mixed data for further training, the accuracy rate of the model improved to 93.74%. Finally, the momentum gradient descent method was used to optimize the model convergence, and the final model accuracy rate reached 94.29%, which was 3.98% and 1.51% higher than Inception-V3 and Xception networks, respectively.
ConclusionIn the case of a small amount of data, reducing the complexity of the model and transfering existing prior knowledge will help to improve the performance of the model. The Mixup mixed data enhancement method is beneficial to improve the adaptability of the model to identify image samples of citrus fruit with Huanglongbing and improve model performance. The X-ResNeXt model is superior to the classic recognition model in terms of accuracy rate and recall rate, and can provide references for the high-precision, rapid and non-destructive identification of citrus Huanglongbing.
-
-
![]()
图 3 不同混合系数(
$ \lambda $ )的Mixup算法进行柑橘黄龙病数据集增强的结果示意图Figure 3. Results of enhancing the citrus Huanglongbing data set by the Mixup algorithm with different mixing coefficient (
$ \lambda $ )表 1 Mixup混类数据增强对模型性能的影响
Table 1 Effects of Mixup mixed data enhancement on model performance
模型大小 Model batchsize λ 准确率/% Accuracy rate 召回率/% Recall rate F1指标 F1 index 32×2 1.0 91.97 90.76 0.914 32×2 0.8 92.17 91.25 0.917 32×2 0.4 93.74 92.58 0.932 32×2 0.2 93.31 92.17 0.927 32×2 0 91.88 90.79 0.913 
下载: 导出CSV
表 2 模型梯度下降优化方法对比试验
Table 2 Comparison test of model optimization methods based on gradient descent
优化方法
Optimization method动量
Momentum准确率/%
Accuracy rate召回率/%
Recall rateF1指标
F1 index随机梯度下降法 Stochastic gradient descent (SGD) 93.53 92.42 0.930 动量梯度下降法 Momentum gradient descent (MGD) 0 93.53 92.42 0.930 0.5 93.74 92.58 0.932 0.9 94.29 93.69 0.939
下载: 导出CSV
-
[1] YAN X J, YUAN H Z, ZHOU X X. Control efficacy of different pesticide formulations and fan-nozzle model on wheat aphids by UAVs[J]. International Journal of Precision Agricultural Aviation, 2020, 3(2): 35-39.
[2] 范国成, 刘波, 吴如健, 等. 中国柑橘黄龙病研究30年[J]. 福建农业学报, 2009, 24(2): 183-190. doi: 10.3969/j.issn.1008-0384.2009.02.019 [3] 谢钟琛, 李健, 施清, 等. 福建省柑橘黄龙病危害及其流行规律研究[J]. 中国农业科学, 2009, 42(11): 3888-3897. doi: 10.3864/j.issn.0578-1752.2009.11.016 [4] 邓小玲, 郑建宝, 梅慧兰, 等. 基于高光谱成像技术的柑橘黄龙病病情诊断及分类[J]. 西北农林科技大学学报(自然科学版), 2013, 41(7): 99-105. [5] 邓小玲, 林亮生, 兰玉彬. 基于调制荧光检测技术的柑橘黄龙病诊断[J]. 华南农业大学学报, 2016, 37(2): 113-116. doi: 10.7671/j.issn.1001-411X.2016.02.018 [6] 马淏, 吉海彦, WON S L. 基于高光谱成像技术应用光谱及纹理特征识别柑橘黄龙病[J]. 光谱学与光谱分析, 2016, 36(7): 2344-2350. [7] 王克健. 柑橘黄龙病光谱检测与柑橘木虱UAV防控技术探索[D]. 重庆: 西南大学, 2018. [8] 肖怀春. 基于高光谱成像技术的柑橘黄龙病诊断方法研究[D]. 南昌: 华东交通大学, 2019. [9] 戴泽翰, 郑正, 黄莉舒, 等. 基于深度卷积神经网络的柑橘黄龙病症状识别[J]. 华南农业大学学报, 2020, 41(4): 111-119. doi: 10.7671/j.issn.1001-411X.201909031 [10] 陈波, 姚林建. 光谱检测技术在柑橘黄龙病诊断中的研究进展[J]. 赣南师范大学学报, 2018, 39(6): 69-72. [11] 黄志强, 李军. 基于卷积神经网络的图像识别研究综述[J]. 汽车工程师, 2020(10): 11-13. doi: 10.3969/j.issn.1674-6546.2020.10.003 [12] KRIZHEVSKY A, SUTSKEVER I, HINTON G E. ImageNet classification with deep convolutional neural networks[J]. Communications of the ACM, 2017, 60(6): 84-90. doi: 10.1145/3065386
[13] 郑二功, 田迎芳, 陈涛. 基于深度学习的无人机影像玉米倒伏区域提取[J]. 河南农业科学, 2018, 47(8): 155-160. [14] 卢宏涛, 张秦川. 深度卷积神经网络在计算机视觉中的应用研究综述[J]. 数据采集与处理, 2016, 31(1): 1-17. [15] ZHANG H, CISSE M, DAUPHIN Y N, et al. Mixup: Beyond empirical risk minimization[EB/OL]. [2020-06-15]. https://arxiv.org/pdf/1710.09412.pdf.
[16] HE K, ZHANG X, REN S, et al. Deep residual learning for image recognition[C]. IEEE Conference on Computer Vision and Pattern Recognition, 2016: 770-778.
[17] XIE S, GIRSHICK R, DOLLÁR P, et al. Aggregated residual transformations for deep neural networks[C]. IEEE Conference on Computer Vision and Pattern Recognition, 2017: 5987-5995.
[18] CHOLLET F. Xception: Deep learning with depth wise separable convolutions[C]. IEEE Conference on Computer Vision and Pattern Recognition, 2017: 1800-1807.
[19] DONAHUE J, JIA Y, VINYALS O, et al. DeCAF: A deep convolutional activation feature for generic visual recognition[J]. Computer Science, 2013. arXiv: 1310.1531.
[20] ALEX K, SUTSKEVER I, HINTON G E. Imagenet classification with deep convolutional neural networks[J]. Communication of the ACM, 2012: 84-90.
[21] SUTSKEVER I, MARTENS G, DAHI G, et al. On the importance of initialization and momentum in deep learning[EB/OL]. [2020-06-15]. http://citeseerx.ist.psu.edu/viewdoc/download;jsessionid=6C87CA259F724D8BDDFA5E3B2C0F10F2?doi=10.1.1.352.8324&rep=rep1&type=pdf.
[22] 胡子昂, 王卫星, 陆健强, 等. 视频图像去雾霾技术研究综述与展望[J]. 电子技术与软件工程, 2015(22): 94-95. [23] 薛月菊, 黄宁, 涂淑琴, 等. 未成熟芒果的改进YOLOv2识别方法[J]. 农业工程学报, 2018, 34(7): 173-179. doi: 10.11975/j.issn.1002-6819.2018.07.022
计量
- 文章访问数: 1556
- HTML全文浏览量: 2
- PDF下载量: 1209
