Rice Pest Identification Based on Convolutional Neural Network and Transfer Learning

Hongyun Yang; Xiaomei Xiao; Qiong Huang; Guoliang Zheng; Wenlong Yi

doi:10.3788/LOP202259.1615004

Laser & Optoelectronics Progress, Volume. 59, Issue 16, 1615004(2022)

Rice Pest Identification Based on Convolutional Neural Network and Transfer Learning

Hongyun Yang^1、*, Xiaomei Xiao¹, Qiong Huang², Guoliang Zheng¹, and Wenlong Yi¹

Author Affiliations

¹School of Software, Jiangxi Agricultural University, Nanchang 330045, Jiangxi , China

²School of Computer and Information Engineering, Jiangxi Agricultural University, Nanchang 330045, Jiangxi , China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(12)

Fig. 1. Samples of rice pests. (a) Rice weevil; (b) rice planthopper; (c) rice grasshopper; (d) striped rice borer; (e) rice leaf roller; (f) yellow rice borer

Download full size

Fig. 2. Structure of convolutional neural network

Download full size

Fig. 3. VGG16 convolutional neural network model

Download full size

Fig. 4. Flow chart of rice pest algorithm based on CNN and parameter migration

Download full size

Fig. 5. Recognition effect based on the newly learned VGG16. (a) Accuracy curve;(b) loss curve

Download full size

Fig. 6. [in Chinese]

Download full size

Table 1. Rice pest dataset
View table
Table 1. Rice pest dataset
Pest species Number of images
Original After data enhancement
Rice weevil 112 1120
Rice planthopper 122 1220
Rice grasshoppe 107 1070
Striped rice borer 127 1270
Rice leaf roller 106 1060
Yellow rice borer 102 1020

Table 2. Experimental comparison of different top-level design schemes
View table
Table 2. Experimental comparison of different top-level design schemes
Experiment number Accuracy /% Loss Time /s Model size /MB
1 84.70 0.689 9439.03 68.4
2 96.43 0.098 9650 154
3 96.94 0.090 9742 252
4 97.15 0.098 10524 968
5 91.84 0.330 9244 56.2

Table 3. Influence of freezing all convolutional layers on experiment
View table
Table 3. Influence of freezing all convolutional layers on experiment
Model Whether to unfreeze convolution layer Accuracy /% Loss Time /s Model size /MB
VGG16_M No 96.43 0.098 9650 154
Yes 98.98 0.042 10015 172
VGG16_N No 91.84 0.330 9244 56.2
Yes 99.05 0.025 9661 74.2

Table 4. Comparison between improved scheme and original model
View table
Table 4. Comparison between improved scheme and original model
Model Learning method Accuracy /% Loss Time /s Model size /MB
VGG16 New learning 86.52 0.93 20204 1000
VGG16_N Transfer learning 99.05 0.02 9661 74.2

Table 5. Classification performance of proposed model
View table
Table 5. Classification performance of proposed model
Pest species Precision Recall F1
Rice grasshoppe 0.98 0.99 0.98
Rice planthopper 0.99 0.99 0.99
Rice weevil 0.99 0.99 0.99
Striped rice borer 0.99 0.99 0.99
The rice leaf roller 1.00 1.00 1.00
Yellow rice borer 1.00 0.99 0.99

Table 6. Performance comparison of different network models

View table

Table 6. Performance comparison of different network models

Model	Learning method	Accuracy /%	Loss	Time /s	Model size /MB
Alexnet	New learning	94.20	0.3089	5109	444
Resnet34	New learning	98.25	0.0719	15128	173
Resnet50	New learning	96.79	0.1911	398270	349
VGG16_N	Transfer learning	99.05	0.0247	9661	74.2

Tools

Get Citation

Copy Citation Text

Hongyun Yang, Xiaomei Xiao, Qiong Huang, Guoliang Zheng, Wenlong Yi. Rice Pest Identification Based on Convolutional Neural Network and Transfer Learning[J]. Laser & Optoelectronics Progress, 2022, 59(16): 1615004

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites