Infrared Aircraft Classification Method with Small Samples Based on Improved Relation Network

Lu Jin; Shijian Liu; Xiao Wang; Fanming Li

doi:10.3788/AOS202040.0811005

Acta Optica Sinica, Volume. 40, Issue 8, 0811005(2020)

Infrared Aircraft Classification Method with Small Samples Based on Improved Relation Network

Lu Jin^1,2,3, Shijian Liu^1,3, Xiao Wang^1,2,3, and Fanming Li^1,3、*

Author Affiliations

¹Shanghai Institute of Technical Physics, Chinese Academy of Sciences, Shanghai 200083, China

²University of Chinese Academy of Sciences, Beijing 100049, China

³Key Laboratory of Infrared System Detection and Imaging Technology, Chinese Academy of Sciences, Shanghai 200083, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(11)

Fig. 1. Overall architecture of infrared aircraft classification learning model with small samples

Download full size

Fig. 2. Architecture of module. (a) Embedding module; (b) relation module

Download full size

Fig. 3. Illustration of few-shot learning datasets under meta learning training mode

Download full size

Fig. 4. Pseudo-code for learning algorithm

Download full size

Fig. 5. Partial examples of three datasets. (a) mini-ImageNet dataset; (b) Infra-object dataset; (c) Infra-aircraft dataset

Download full size

Fig. 6. Test accuracy and loss curves on mini-ImageNet dataset. (a) 5-way 1-shot; (b) 5-way 5-shot

Download full size

Fig. 7. Relationship between test accuracy and adequacy of training samples on mini-ImageNet dataset

Download full size

Fig. 8. Accuracy comparison of ground to air infrared aircraft classification

Download full size

Table 1. Accuracy of convolution kernel size estimation%
View table
Table 1. Accuracy of convolution kernel size estimation%
Convolution kernel size 5-way 1-shot 5-way 5-shot
3×3 81.23 91.28
5×5 81.41 90.92
7×7 79.33 87.07
9×9 75.98 82.30

Table 2. Accuracy of each model on mini-ImageNet dataset%

View table

Table 2. Accuracy of each model on mini-ImageNet dataset%

Model	Fine-tune	5-way 1-shot	5-way 5-shot
Baseline-nearest-neighbor^[19]	N	41.08±0.70	51.04±0.65
Baseline-linear^[20]	Y	42.11±0.71	62.53±0.69
Meta-learner LSTM^[19]	N	43.44±0.77	60.60±0.71
MAML^[21]	Y	48.70±1.84	63.11±0.92
Matching network^[13]	Y	42.40	58.00
Prototypical network^[14]	F	49.42±0.78	68.20±0.66
RelationNet^[15]	F	50.44±0.82	65.32±0.70
Improved relation network	F	54.89±1.02	69.87±0.75

Table 3. Accuracy of model for infrared aircraft classification on different training datasets%

View table

Table 3. Accuracy of model for infrared aircraft classification on different training datasets%

Group	Training dataset	Test dataset	5-way 1-shot	5-way 5-shot	8-way 1-shot	8-way 5-shot
1	Infra-object+Infra-aircraft	Infra-object+Infra-aircraft	86.25±1.25	94.84±0.66	77.82±1.79	91.11±0.63
2	Infra-object	Infra-aircraft	84.37±1.31	93.66±0.76	77.56±1.46	90.58±0.64
3	mini-ImageNet	Infra-aircraft	78.92±2.78	90.76±1.24	74.44±3.28	86.34±1.95
4	Infra-object+mini-ImageNet	Infra-aircraft	82.79±0.75	94.51±0.82	78.47±0.94	89.92±1.02

Tools

Get Citation

Copy Citation Text

Lu Jin, Shijian Liu, Xiao Wang, Fanming Li. Infrared Aircraft Classification Method with Small Samples Based on Improved Relation Network[J]. Acta Optica Sinica, 2020, 40(8): 0811005

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites