Adaptive Top-Hat Infrared Small Target Detection Based on Local Contrast

Tengyan Xi; Lihua Yuan; Shupeng Wang

doi:10.3788/LOP222850

Journals >Laser & Optoelectronics Progress >Volume 60 >Issue 16 >Page 1628003 > Article

Laser & Optoelectronics Progress
Vol. 60, Issue 16, 1628003 (2023)

Adaptive Top-Hat Infrared Small Target Detection Based on Local Contrast

Tengyan Xi¹, Lihua Yuan^1,*, and Shupeng Wang²

Author Affiliations

¹Key Laboratory of Nondestructive Testing, Ministry of Education, College of Testing and Optoelectronic Engineering, Nanchang Hangkong University, Nanchang 330063, Jiangxi, China

²China Aviation Development Shenyang Liming Aero Engine Co., Ltd., Shenyang 110000, Liaoning, China

show less

DOI: 10.3788/LOP222850 Cite this Article Set citation alerts

Tengyan Xi, Lihua Yuan, Shupeng Wang. Adaptive Top-Hat Infrared Small Target Detection Based on Local Contrast[J]. Laser & Optoelectronics Progress, 2023, 60(16): 1628003 Copy Citation Text

show less

Fig. 1. Schematic of target area

Download full size

Fig. 2. Original image and LCM map. (a) Original image; (b) LCM map

Download full size

Fig. 3. Schematic of structural elements

Download full size

Fig. 4. Schematic of size setting of structural elements

Download full size

Fig. 5. Overall process of ATHLC method

Download full size

Fig. 6. Schematic of target area

Download full size

Fig. 7. Simulation results of three backgrounds at each stage

Download full size

Fig. 8. Experimental results of similar methods. (a) Original images; (b) 3D gray scale images of original image; (c) Top-Hat; (d) NWTH; (e) PITH; (f) ATHLC

Download full size

Fig. 9. Comparison diagrams of different classes of methods. (a) Original images; (b) ADMD; (c) AADCCD; (d) HBMLCM; (e) RLCM; (f) MPCM; (g) PSTNN; (h) ATHLC

Download full size

Fig. 10. Three dimensional gray scale representation of different classes of methods. (a) Original images; (b) ADMD; (c) AADCCD; (d) HBMLCM; (e) RLCM; (f) MPCM; (g) PSTNN; (h) ATHLC

Download full size

Fig. 11. ROC curves of different methods

Download full size

Fig. 12. PR curves of different methods

Download full size

Sequence No.	Target size	Number of frames	Image size	Average SCR	Image description
Seq1	5 $\times$ 5	30	256 $\times$ 200	0.62	The cloud background，mostly covered by scattered clouds，has a fixed perspective and small targets from left to right.
Seq2	3 $\times$ 3	30	256 $\times$ 256	0.36	The complex ground background，mostly covered by vegetation and mountains，has a sloping dividing line from the moving perspective.
Seq3	3 $\times$ 3	30	256 $\times$ 256	1.46	Complex ground background，partially covered by mountain forest and ground，moving perspective.
Seq4	3 $\times$ 3	30	640 $\times$ 512	0.91	River and building background，the bridge span from top left to bottom right，with a fixed perspective，and the target are from top left to bottom right.
Seq5	1 $\times$ 1	30	640 $\times$ 512	1.08	Sky and architectural background，fixed perspective，target from right to left.
Seq6	3 $\times$ 3	30	640 $\times$ 512	0.26	The sky background，mostly covered by clouds，has a fixed viewing angle and targets from right to left.

Table 1. Sequence image information

Method	Parameter setting
Top-Hat	Structuring type is disk；structuring element size is 5×5
NWTH	R_o=9，R_i=4 for sequences 1-3；R_o=7，R_i=3 for sequences 4-6
PITH	B_o=9，B_i=4，B_e=3 for sequences 1-3；B_o=7，B_i=3，B_e=2 for sequences 4-6
ATHLC	B_o=11，B_i=5，Be=4 for sequences 1-3；B_o=9，B_i=4，B_e=3 for sequences 4-6；scale is 3，5；window size is 3×3；k'=4

Table 2. Parameter settings of comparison methods

Sequence No.	Top-Hat SCRG/BSF/CG	NWTH SCRG/BSF/CG	PITH SCRG/BSF/CG	ATHLC SCRG/BSF/CG
Seq1	15.66/1.08/8.53	Inf/10.00/4.83	Inf/11.78/4.59	Inf/Inf/16.24
Seq2	16.04/0.88/9.12	345.99/5.66/9.01	434.45/6.76/8.90	Inf/Inf/29.73
Seq3	3.60/1.23/2.32	23.97/10.83/1.97	27.45/12.25/1.95	Inf/Inf/6.78
Seq4	4.16/1.57/1.70	Inf/31.79/1.87	Inf/34.45/1.87	Inf/300.58/9.90
Seq5	5.31/2.99/3.59	Inf/50.40/3.59	Inf/55.09/3.60	Inf/1550.09/19.79
Seq6	179.92/9.46/52.28	Inf/134.77/40.12	Inf/149.33/40.26	Inf/310.57/573.06

Table 3. SCRG, BSF, and CG of similar methods

Method	Parameter setting
ADMD	Scale is 3，5，7，9；window size is 3 $\times$ 3
AADCDD	Scale is 3，5，7，9；window size is 3 $\times$ 3
HBMLCM	Scale is 3，5，7，9；window size is 15 $\times$ 15
RLCM	Scale is 3； $k_{1}$ =2，5，9； $k_{2}$ =4，9，16
MPCM	Scale is 3，5，7，9；mean filtering size is 3 $\times$ 3
PSTNN	Patch size is 40 $\times$ 40；sliding step is 40； $λ = \frac{1}{\sqrt[]{m a x (m, n)}}$
ATHLC	B_o=11，B_i=5，B_e=4 for sequences 1-3；B_o=9，B_i=4，B_e=3 for sequences 4-6；scale is 3，5；window size is 3 $\times$ 3；k′=4

Table 4. Parameter settings for different classes of methods

Method	Seq1	Seq2	Seq3	Seq4	Seq5	Seq6
ADMD	224.59	Inf	30.71	Inf	Inf	3816.40
AADCDD	255.53	Inf	59.11	Inf	Inf	Inf
HBMLCM	89.14	120.94	29.66	49.78	82.70	1384.73
RLCM	112.66	48.78	27.98	120.97	Inf	Inf
MPCM	3.35	1.15	1.28	0.17	1.14	34.28
PSTNN	Inf	Inf	Inf	Inf	Inf	Inf
ATHLC	Inf	Inf	Inf	Inf	Inf	Inf

Table 5. SCRG of different classes of methods

Method	Seq1	Seq2	Seq3	Seq4	Seq5	Seq6
ADMD	15.16	4.53	5.02	13.23	22.03	28.93
AADCDD	9.55	4.91	4.82	7.30	26.88	116.93
HBMLCM	24.31	3.18	3.41	8.15	13.36	107.05
RLCM	4.08	1.73	4.78	7.06	12.78	25.71
MPCM	2.10	3.26	4.04	4.21	18.17	8.07
PSTNN	Inf	4.08	3.35	15.02	33.92	130.54
ATHLC	Inf	Inf	Inf	300.58	1550.09	310.57

Table 6. BSF of different classes of methods

Method	Seq1	Seq2	Seq3	Seq4	Seq5	Seq6
ADMD	2.63	3.12	2.39	0.67	2.99	182.73
AADCDD	3.71	1.88	1.20	0.19	0.88	36.62
HBMLCM	3.82	5.51	2.62	0.77	2.83	317.84
RLCM	17.86	26.84	9.28	9.60	18.37	231.67
MPCM	2.30	0.67	0.94	0.30	0.78	34.86
PSTNN	6.21	6.43	1.89	1.49	3.07	45.57
ATHLC	16.24	29.73	6.78	9.90	19.79	573.06

Table 7. CG of different classes of methods

Method	Seq1	Seq2	Seq3	Seq4	Seq5	Seq6
ADMD	0.0253	0.0152	0.0130	0.0380	0.0370	0.0358
AADCDD	0.0260	0.0271	0.0265	0.0939	0.0940	0.0923
HBMLCM	0.0153	0.0147	0.0131	0.0543	0.0556	0.0470
RLCM	0.9791	1.3668	1.3906	7.1697	7.5835	6.7185
MPCM	0.0335	0.0423	0.0446	0.1742	0.1789	0.1779
PSTNN	0.0423	0.1697	0.1883	0.9150	0.7672	0.5253
ATHLC	0.0242	0.0367	0.0327	0.1331	0.1335	0.1289

Table 8. Running time of different classes of methods

Tengyan Xi, Lihua Yuan, Shupeng Wang. Adaptive Top-Hat Infrared Small Target Detection Based on Local Contrast[J]. Laser & Optoelectronics Progress, 2023, 60(16): 1628003

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information

微信扫一扫：分享

微信扫一扫：分享