Simulation of Rice Disease Recognition Based on Improved Attention Mechanism Embedded in PR-Net Model

doi:10.16182/j.issn1004731x.joss.23-0322

Abstract

Abstract:

Aiming at the low accuracy of existing CNN models in identifying rice leaf diseases, a hybrid convolutional neural network model PRC-Net (parallel residual with coordinate attention network) combining parallel structure and residual structure is proposed. A parallel structure is introduced to improve the receptive field of convolution, and the residual structure is combined to achieve the complete and continuous transmission of feature information. An improved spatial attention mechanism is embedded into the backbone model PR-Net to enhance the degree of aggregation of lesion feature information at different scales. In order to further improve the accuracy of disease identification and reduce the training and reasoning time of the model, the model structure is optimized by changing the weighting method. Simulation results show that, compared to the classification models such as InceptionResNetV2, PRC-Net has fewer training parameters, shorter training time, and higher recognition accuracy, which is superior to the other crop disease identification models.

Key words: rice leaf disease, PRC-Net, convolution neural network, attention mechanism, image recognition

CLC Number:

TP391

Lu Yang, Liu Pengfei, Xu Siyuan, Liu Qiwang, Gu Fuqian, Wang Peng. Simulation of Rice Disease Recognition Based on Improved Attention Mechanism Embedded in PR-Net Model[J]. Journal of System Simulation, 2024, 36(6): 1322-1333.

Figures/Tables 17

Fig. 1

Fig. 2

Table 1

Relevant parameters of PR-Net structure

层名	内核	步长	输出形状
Input			(None, 256, 256, 3)
Conv2D_1	$7 × 7$	2	(None, 128, 128, 128)
MaxPooling2D_1	$3 × 3$	2	(None, 64, 64, 128)
Conv2D_2	$3 × 3$	1	(None, 64, 64, 64)
MaxPooling2D_2	$3 × 3$	2	(None, 32, 32, 64)
PR-Block A			(None, 32, 32, 192)
PR-Block B			(None, 32, 32, 384)
PR-Block C			(None, 32, 32, 1280)
GlobalAveragePooling2D			(None, 1280)
Softmax			(None, 6)

Table 1

Fig. 3

Fig. 4

Fig. 5

Table 2

Fig. 6

Table 3

Table 4

Table 5

Fig. 7

Table 6

Fig. 8

Fig. 9

Table 7

Table 8

References 23

1	Rusinque Leidy, Maleita Carla, Abrantes Isabel, et al. Meloidogyne Graminicola-A Threat to Rice Production: Review Update on Distribution, Biology, Identification, and Management[J]. Biology, 2021, 10(11): 1163.
2	Qiu Jing, Lu Xiaolei, Wang Xingxing, et al. Research on Rice Disease Identification Model Based on Migration Learning in VGG Network[J]. IOP Conference Series: Earth and Environmental Science, 2021, 680(1): 012087.
3	Bashir K, Rehman M, Bari M. Detection and Classification of Rice Diseases: An Automated Approach Using Textural Features[J]. Mehran University Research Journal of Engineering and Technology, 2019, 38(1): 239-250.
4	Saha S, Ahsan S M M. Rice Disease Detection Using Intensity Moments and Random Forest[C]//2021 International Conference on Information and Communication Technology for Sustainable Development (ICICT4SD). Piscataway, NJ, USA: IEEE, 2021: 166-170.
5	Sharma Vikas, Aftab Ahmad Mir, Sarwr Abid. Detection of Rice Disease Using Bayes' Classifier and Minimum Distance Classifier[J]. Journal of Multimedia Information System, 2020, 7(1): 17-24.
6	Simonyan K, Zisserman A. Very Deep Convolutional Networks for Large-scale Image Recognition[EB/OL]. (2015-04-10) [2022-10-28]. .
7	He Kaiming, Zhang Xiangyu, Ren Shaoqing, et al. Deep Residual Learning for Image Recognition[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway, NJ, USA: IEEE, 2016: 770-778.
8	Szegedy C, Vanhoucke V, Ioffe S, et al. Rethinking the Inception Architecture for Computer Vision[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway, NJ, USA: IEEE, 2016: 2818-2826.
9	Szegedy C, Ioffe S, Vanhoucke V, et al. Inception-v4, Inception-ResNet and the Impact of Residual Connections on Learning[C]//Proceedings of the Thirty-first AAAI Conference on Artificial Intelligence. Palo Alto, CA, USA: AAAI Press, 2017: 4278-4284.
10	Niu Zhaoyang, Zhong Guoqiang, Yu Hui. A Review on the Attention Mechanism of Deep Learning[J]. Neurocomputing, 2021, 452: 48-62.
11	Hughes D P, Salathé Marcel. An Open Access Repository of Images on Plant Health to Enable the Development of Mobile Disease Diagnostics[EB/OL]. (2016-04-12) [2023-01-03]. .
12	Prabira Kumar Sethy, Negi Baishalee, Nalini Kanta Barpanda, et al. Measurement of Disease Severity of Rice Crop Using Machine Learning and Computational Intelligence[M]//Sasikumar Gurumoorthy, Bangole Narendra Kumar Rao, Gao Xiaozhi. Cognitive Science and Artificial Intelligence: Advances and Applications. Singapore: Springer Singapore, 2018: 1-11.
13	Shrivastava Vimal K, Pradhan Monoj K. Rice Plant Disease Classification Using Color Features: A Machine Learning Paradigm[J]. Journal of Plant Pathology, 2021, 103(1): 17-26.
14	Mohapatra Subasish, Marandi Chandan, Sahoo Amlan, et al. Rice Leaf Disease Detection and Classification Using a Deep Neural Network[C]//Computing, Communication and Learning. Cham: Springer Nature Switzerland, 2022: 231-243.
15	Chen Zhaoyi, Wu Ruhui, Lin Yiyan, et al. Plant Disease Recognition Model Based on Improved YOLOv5[J]. Agronomy, 2022, 12(2): 365.
16	Chen Junde, Zhang Defu, Zeb Adnan, et al. Identification of Rice Plant Diseases Using Lightweight Attention Networks[J]. Expert Systems with Applications, 2021, 169: 114514.
17	Jaderberg M, Simonyan K, Zisserman A, et al. Spatial Transformer Networks[C]//Proceedings of the 28th International Conference on Neural Information Processing Systems. Cambridge, MA, USA: MIT Press, 2015: 2017-2025.
18	Hou Qibin, Zhou Daquan, Feng Jiashi. Coordinate Attention for Efficient Mobile Network Design[C]//2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway, NJ, USA: IEEE, 2021: 13708-13717.
19	Gulrajani I, Ahmed F, Arjovsky M, et al. Improved Training of Wasserstein GANs[C]//Proceedings of the 31st International Conference on Neural Information Processing Systems. Red Hook, NY, USA: Curran Associates Inc., 2017: 5769-5779.
20	Hu Jie, Shen Li, Sun Gang. Squeeze-and-excitation Networks[C]//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway, NJ, USA: IEEE, 2018: 7132-7141.
21	Woo Sanghyun, Park Jongchan, Young Lee Joon, et al. CBAM: Convolutional Block Attention Module[C]//Computer Vision-ECCV 2018. Cham: Springer International Publishing, 2018: 3-19.
22	Nandhini S, Ashokkumar K. An Automatic Plant Leaf Disease Identification Using DenseNet-121 Architecture with a Mutation-based Henry Gas Solubility Optimization Algorithm[J]. Neural Computing and Applications, 2022, 34(7): 5513-5534.
23	Poornima Singh Thakur, Sheorey Tanuja, Ojha Aparajita. VGG-ICNN: A Lightweight CNN Model for Crop Disease Identification[J]. Multimedia Tools and Applications, 2023, 82(1): 497-520.

类别	训练集数量	测试集数量
合计	10 973	2 863
稻瘟病	1 689	422
白叶枯病	1 439	480
胡麻斑病	1 478	369
纹枯病	1 415	354
东格鲁病	1 329	332
健康叶片	3 623	906

模型	训练准确率/%	训练损失率	测试准确率/%	测试损失率
VGG16	99.61	0.007 5	98.48	0.028 3
ResNet50	99.82	0.054 0	98.27	0.041 0
InceptionV3	99.24	0.009 6	98.13	0.022 2
InceptionResNetV2	99.87	0.002 4	98.51	0.020 1
PR-Net	99.55	0.006 0	99.10	0.013 8
CA + PR-Net	99.62	0.004 9	97.03	0.035 2
Improved CA + PR-Net	99.56	0.004 7	99.27	0.007 6
PRC-Net	99.56	0.005 4	99.65	0.007 3
CA+MobileNetV2	99.36	0.009 7	91.39	0.080 5

模型	指标	0	1	2	3	4	5
VGG16	精确率	0.97	0.97	0.97	0.99	0.99	0.99
VGG16	召回率	0.98	0.98	0.96	0.99	0.99	0.99
ResNet50	精确率	0.98	0.98	0.99	0.97	0.99	0.97
ResNet50	召回率	0.97	0.97	0.96	0.99	0.99	1.00
InceptionV3	精确率	0.97	0.95	1.00	1.00	0.99	1.00
InceptionV3	召回率	0.98	0.98	0.90	0.99	1.00	1.00
Inception-ResNetV2	精确率	0.98	1.00	0.95	1.00	1.00	1.00
Inception-ResNetV2	召回率	0.99	0.97	1.00	0.98	0.99	1.00
PRC-Net	精确率	1.00	0.99	1.00	1.00	1.00	1.00
PRC-Net	召回率	0.99	1.00	1.00	1.00	1.00	1.00

核尺寸	训练准确率/%	训练损失率	测试准确率/%	测试损失率
1	99.56	0.004 7	99.27	0.007 6
2	99.47	0.005 1	98.67	0.011 2
3	99.51	0.005 6	98.53	0.012 6
4	98.95	0.009 3	98.44	0.011 2
5	98.93	0.008 5	98.42	0.012 9

模型	模型尺寸/MB	平均迭代时间/min	总训练时间/min	推理时间/s
VGG16	172	1.42	42.72	0.012
ResNet50	285	1.36	40.72	0.013
InceptionV3	259	1.34	40.13	0.011
Inception-ResNetV2	630	2.93	87.90	0.024
PR-Net	43.8	1.02	30.72	0.008
CA + PR-Net	44.5	1.59	47.25	0.013
Improved CA+PR-Net	86.3	1.64	49.22	0.014
PRC-Net	86.3	1.29	38.72	0.010