Fig. 1. Belt noise in image shot by linear array sensor

Fig. 2. Belt noise causes max(z) being too high

Fig. 3. Deviation ratio between M_C of circle target with radius of R and ideal value

Fig. 4. Targets with minimum M_C when 2≤n≤4. (a) M_C=0.3536; (b) M_C=0.3849; (c) M_C=0.3750

Fig. 5. M_C of some typical pixels. (a) M_C=0.4033; (b) M_C=0.6479; (c) M_C=0.8000

Fig. 6. Simulated infrared target in guard band

Fig. 7. Window position shift in low SNR condition. (a) Original image; (b) fusion map of original image and shifted window

Fig. 8. Flowchart of proposed method

Fig. 9. Processing of IR images of birds in clouds. (a1)(a2) Original images; (b1)(b2) heat maps of window distribution; (c1)(c2) target probability maps; (d1)(d2) segmentation results

Fig. 10. Three-dimensional projection of normalized |M_TPM| in Figs. 9(c1) and 9(c2). (a) Projection of Fig. 9(c1); (b) projection of Fig. 9(c2)

Fig. 11. Simulated images and M_TPM for different SNR. (a1) Simulated image with R_SNR=0.2; (b1) simulated image with R_SNR=0.5; (c1) simulated image with R_SNR=1; (d1) simulated image with R_SNR=3; (a2) M_TPM with R_SNR=0.2; (b2) M_TPM with R_SNR=0.5; (c2) M_TPM with R_SNR=1; (d2) M_TPM with R_SNR=3

Fig. 12. Comparison of several target detection methods. (a1)(b1)(c1)(d1) Scenes a,b,c,and d; (a2)(b2)(c2)(d2) NIPPS; (a3)(b3)(c3)(d3) RIPT; (a4)(b4)(c4)(d4) MDTDLMS-RDLCM; (a5)(b5)(c5)(d5) proposed method

Fig. 13. ROC curves of different algorithms