Spectral Polarization Image Target Detection Method Based on Interleaved Sequence Mapping

Abstract

The present invention relates to a method for detecting spectral polarization image targets based on interleaved sequence mapping, which includes the following steps: Step S1: Select a suitable wavelength band according to the spectral polarization characteristics of the target and obtain the polarization image of the corresponding wavelength band; Step S2: Based on the polarization image, calculate the polarization parameters, and then perform differential enhancement and one-dimensional data mapping of the polarization direction based on the interleaved sequence method to obtain the differential enhancement result and the interleaved sequence mapping result; Step S3: Perform image fusion on the differential enhancement result and the interleaved sequence mapping result to obtain a fused image; Step S4: Perform pixel-based target detection on the fused image. The present invention effectively improves the contrast and clarity between the target and the background, thereby improving the accuracy and precision of target detection.

Description

Technical Field

[0001] The present invention belongs to the fields of image fusion technology and computer vision, and particularly relates to a method for detecting targets in spectral polarization images based on interleaved sequence mapping. Background Art

[0002] With the development of sensing technology, more and more information can be extracted from the scene of interest. This includes spatial information captured by cameras, spectral information obtained from spectrometers, and polarization information acquired by polarimeters. Spatial, spectral, and polarization information reveals different characteristics of targets and backgrounds. It has been proven that better detection results can be obtained by using spatial and spectral information (spectral images) or spatial and polarization information (polarization images), but both methods have certain deficiencies. Summary of the Invention

[0003] In view of this, the purpose of the present invention is to provide a method for detecting targets in spectral polarization images based on interleaved sequence mapping, which respectively improves the target contrast of the degree of polarization and polarization angle images by using difference enhancement and interleaved sequence mapping. On the basis of retaining the edge and contour information in the source images, wavelet transform fusion effectively improves the contrast and clarity between the target and the background.

[0004] To achieve the above purpose, the present invention adopts the following technical solutions:

[0005] A method for detecting targets in spectral polarization images based on interleaved sequence mapping, comprising the following steps:

[0006] Step S1: Select a suitable band according to the spectral polarization characteristics of the target and obtain the polarization image of the corresponding band;

[0007] Step S2: Based on the polarization image, calculate polarization parameters, and then perform difference enhancement and one-dimensional data mapping of the polarization direction based on the interleaved sequence method to obtain a difference enhancement result and an interleaved sequence mapping result;

[0008] Step S3: Perform image fusion on the difference enhancement result and the interleaved sequence mapping result to obtain a fused image;

[0009] Step S4: Perform pixel-based target detection on the fused image.

[0010] Further, a polarization camera is used to obtain the polarization image; the polarization camera adopts a polarization filter array camera that can real-time obtain images at four angles of polarization 0°, 45°, 90°, and 135°. The lens uses a lens with a focal length of 35 mm, and a filter of the selected band is installed in front of the lens.

[0011] Further, the specific content of step S2 is as follows:

[0012] 1) Calculate the Stokes parameters S of polarization 0 、S 1 、S 2 :

[0013] 2) Calculate and obtain the new polarization parameter I according to the Stokes parameters d :

[0014]

[0015] 3) First perform mean filtering on the I d parameter, and then perform difference enhancement to obtain the enhanced result I d1 :

[0016]

[0017] where M×N is a small selected area in the image I d , and μ is the mean value of this area;

[0018] 4) Extract the linear polarization intensity I 1 、S 2 in S p , and represent the polarization direction in the form of two-dimensional data (cos2θ, sin2θ):

[0019] I p =S 1 2 +S 2 2

[0020] cos2θ = S 1 / I p

[0021] sin2θ = S 2 / I p

[0022] 5) For the two-dimensional data (X, Y), normalize it to the data (x, y) within the plane region (0, B n )×(0, B n ):

[0023]

[0024] where B is the sequence base and n is the sequence length;

[0025] 6) The sequences x i and y i corresponding to the normalized data (x, y) are the results of separating x and y bit by bit in the B base:

[0026]

[0027] 7) Find x i and y i for the interleaved sequence z i and convert it to a numerical quantity z, which is the mapping result:

[0028]

[0029]

[0030] Furthermore, the specific steps of step S3 are as follows:

[0031] 1) Use wavelet transform to decompose both the difference enhancement result and the interleaved sequence mapping result into 1 low-frequency component and 3 high-frequency components;

[0032] 2) Perform average weighted fusion on the high-frequency components:

[0033] ch = (ch1 + ch2) / 2

[0034] cv = (cv1 + cv2) / 2

[0035] cd = (cd1 + cd2) / 2

[0036] where ch, ch1, and ch2 are the horizontal components of the fused image, the difference enhancement result, and the interleaved sequence mapping result respectively, cv, cv1, and cv2 are the vertical components of the fused image, the difference enhancement result, and the interleaved sequence mapping result respectively, and cd, cd1, and cd2 are the diagonal components of the fused image, the difference enhancement result, and the interleaved sequence mapping result respectively;

[0037] 3) Perform energy weighted fusion on the low-frequency components:

[0038]

[0039]

[0040] where ca, ca1, and ca2 are the low-frequency components of the fused image, the difference enhancement result, and the interleaved sequence mapping result respectively, cak(n) is the gray value at position n on the corresponding low-frequency component, k = 1, 2;

[0041] 4) Use inverse wavelet transform to obtain the preliminary fused image F 1 ;

[0042] 5) Perform dilation and erosion on the interleaved sequence mapping result respectively.

[0043] 6) Perform weighted fusion on the result and the preliminary fused image to obtain the final fused image:

[0044] F = F 1 + F d+F e

[0045] where F is the fusion result, and F d is the dilation result, and F e is the erosion result.

[0046] Further, the specific step S4 is as follows: performing threshold segmentation on the fused image by using adaptive threshold segmentation, performing morphological processing on the segmentation result, and then mapping the detection result back to the target image.

[0047] The present invention has the following beneficial effects compared with the prior art:

[0048] The present invention modifies the polarization degree parameter and enhances its difference, improves the saliency of the target while suppressing its background; uses the interleaved sequence mapping instead of the polarization angle mapping, reducing a large amount of background noise; after wavelet transform, the low-frequency coefficients are weighted and averaged, and the high-frequency coefficients are selected based on the criterion of the maximum energy of the selected region for fusion, focusing on fusing the high-frequency part of the image, and on the basis of retaining the edge and contour information in the source image, can effectively improve the contrast and clarity between the target and the background; at the same time, the background noise is further reduced through secondary fusion, improving the adaptability of the algorithm to complex environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 is the flowchart of the method of the present invention;

[0050] Figure 2 is the polarization degree image in an embodiment of the present invention;

[0051] Figure 3 is the polarization I d parameter difference enhancement result;

[0052] Figure 4 is the polarization angle image in an embodiment of the present invention;

[0053] Figure 5 is the polarization interleaved sequence mapping result in an embodiment of the present invention;

[0054] Figure 6 is the preliminary fusion image in an embodiment of the present invention;

[0055] Figure 7 is the final fusion image in an embodiment of the present invention;

[0056] Figure 8 is the target detection image in an embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0057] The present invention will be further described below with reference to the drawings and embodiments.

[0058] Please refer to Figure 1 , the present invention provides a method for detecting targets in spectral polarization images based on interleaved sequence mapping, including the following steps:

[0059] Step S1: Select a suitable band according to the spectral polarization characteristics of the target and obtain the polarization image of the corresponding band;

[0060] Step S2: Based on the polarization image, calculate the polarization parameters, and then perform differential enhancement and one-dimensional data mapping of the polarization direction based on the interleaved sequence method to obtain the differential enhancement result and the interleaved sequence mapping result;

[0061] Step S3: Perform image fusion on the differential enhancement result and the interleaved sequence mapping result to obtain a fused image;

[0062] Step S4: Perform pixel-based target detection on the fused image.

[0063] In this embodiment, the polarization camera selects a polarization filter array camera (BLACKFLY SBFS-U3-51S5PC, FLIR, CANADA) that can acquire images at four angles of polarization 0°, 45°, 90°, and 135° in real time. The lens uses a lens with a focal length of 35mm (VIS-NIR, #67-716, EDMUND), and a filter with a wavelength of 750nm is installed in front of the lens. The camouflage target is a military truck model covered with fake leaves.

[0064] In this embodiment, the band determined in step S1 is the band with a relatively high contrast in the polarization parameters of the camouflage target, that is, the band of the filter in front of the camera lens.

[0065] In this embodiment, the new polarization parameter I d is:

[0066]

[0067] I d The difference in polarization degree between the smooth area and the rough area in the image is relatively large. In order to more clearly show this difference, we perform differential enhancement on it:

[0068]

[0069] where M×N is a small area selected in the image I d . In the present invention, the size of the selected area is 3×3, and μ is the mean value of this area. The result is as Figure 3 shown. Compared with the polarization degree map of Figure 2 , it suppresses the background of the target, highlights the target, but also misidentifies some background as the target.

[0070] Meanwhile, perform an interleaved sequence mapping on the S 1 , S 2 parameters. Extract the linear polarization intensity I 1 , S 2 from S p and represent the polarization direction in the form of two-dimensional data (cos2θ, sin 2θ):

[0071] I p = S 1 2 + S 2 2

[0072] cos2θ = S 1 / I p

[0073] sin2θ = S 2 / I p

[0074] After that, normalize the two-dimensional data (X, Y) to the data (x, y) within the plane region (0, B n ) × (0, B n ):

[0075]

[0076] where B is the sequence radix and n is the sequence length. Then, separate the x and y bit by bit in the B - radix to obtain the sequences x i and y i corresponding to the normalized data (x, y):

[0077]

[0078] Next, find the interleaved sequence z i of x i and y i and convert it to a numerical quantity z, which is the mapping result:

[0079]

[0080]

[0081] The interleaved sequence mapping result is as shown in Figure 5 . Compared with the polarization angle mapping result of Figure 4 , it removes a large amount of noise while ensuring the highlighting of the target.

[0082] The results of difference enhancement and interleaved sequence mapping both highlight the target, but there is a lot of noise in both, and the noise overlap between the two is relatively low. Therefore, to improve the target detection performance of the image, we use wavelet transform to decompose both the difference enhancement result and the interleaved sequence mapping result into one low-frequency component and three high-frequency components.

[0083] Perform average weighted fusion on the high-frequency components:

[0084] ch = (ch1 + ch2) / 2

[0085] cv = (cv1 + cv2) / 2

[0086] cd = (cd1 + cd2) / 2

[0087] Among them, ch, ch1, and ch2 are the horizontal components of the fused image, the difference enhancement result, and the interleaved sequence mapping result respectively. cv, cv1, and cv2 are the vertical components of the fused image, the difference enhancement result, and the interleaved sequence mapping result respectively. cd, cd1, and cd2 are the diagonal components of the fused image, the difference enhancement result, and the interleaved sequence mapping result respectively.

[0088] Perform energy weighted fusion on the low-frequency components:

[0089]

[0090]

[0091] Among them, ca, ca1, and ca2 are the low-frequency components of the fused image, the difference enhancement result, and the interleaved sequence mapping result respectively. cak(n) is the gray value at position n on the corresponding low-frequency component, k = 1, 2. Then use inverse wavelet transform to obtain the preliminary fused image F as shown in Figure 6 shown. 1 . However, the background noise in the preliminary fused image F 1 is still relatively high. To further reduce the background noise, perform secondary fusion based on the interleaved sequence mapping result. Dilate and erode the interleaved sequence mapping result obtained in step S2 respectively. Finally, perform weighted fusion on the result and the preliminary fused image to obtain the final fused image:

[0092] F = F 1 + F d + F e

[0093] where F d is the dilation result, F e is the erosion result, and F is the fusion result, as shown in Figure 7 shown.

[0094] To demonstrate the fusion effect of this embodiment, the fused image is subjected to adaptive threshold segmentation, followed by image morphological processing to segment the camouflaged target. The results are as Figure 8 shown. The fused image successfully highlights the camouflaged targets hidden in the grass and fallen leaves, proving that the fusion algorithm of the present invention realizes the complementarity of spectral and polarization information and improves the detection performance of camouflaged targets.

[0095] The above are only the preferred embodiments of the present invention. All equivalent changes and modifications made according to the scope of the patent application of the present invention shall fall within the scope of the present invention.

Claims

1. A method for target detection of spectral polarization images based on interleaved sequence mapping, characterized in that, it includes the following steps: Step S1: Select a suitable band according to the spectral polarization characteristics of the target and obtain the polarization image of the corresponding band; Step S2: Based on the polarization image, calculate the polarization parameters, and then perform differential enhancement and one-dimensional data mapping of the polarization direction based on the interleaved sequence method to obtain the differential enhancement result and the interleaved sequence mapping result; Step S3: Use wavelet transform to decompose both the differential enhancement result and the interleaved sequence mapping result into low-frequency components and high-frequency components, perform average weighted fusion on the high-frequency components, perform energy weighted fusion on the low-frequency components, and perform image fusion to obtain a fused image; Step S4: Perform pixel-based target detection on the fused image; The specific content of step S2 is: 1) Calculate the Stokes parameters S 0 、S 1 、S 2； 2) Calculate the new polarization parameter I based on the Stokes parameters d : 3) First, perform mean filtering on the I d parameter, and then perform difference enhancement to obtain the enhanced result I d1 : where M×N is a small region selected in the image I d and μ is the mean value of this region; 4) Extract the linear polarization intensity I 1 from S 2 and represent the polarization direction in the form of two-dimensional data (cos2θ, sin2θ): p ​ I p = S 1 2 + S 2 2 cis2θ = S 1 / I p sin2θ = S 2 / I p 5) For two-dimensional data (X, Y), it is normalized to the data (x, y) within the planar region (0, B n ) × (0, B n ): where B is the sequence base and n is the sequence length; 6) The sequences x i and y i corresponding to the normalized data (x, y) are the results of separating x and y bit by bit in base B: 7) Find x i and y i for the interleaved sequence z i and convert it to a numerical quantity z, which is the mapping result:

2. The method for target detection of spectral polarization images based on interleaved sequence mapping according to claim 1, characterized in that , a polarization camera is used to obtain the polarization image; the polarization camera uses a polarization filter array camera that can obtain images at four angles of polarization 0°, 45°, 90°, and 135° in real time, and the lens uses a lens with a focal length of 35mm, and a filter of the selected band is installed in front of the lens.

3. The method for target detection of spectral polarization images based on interleaved sequence mapping according to claim 1, characterized in that , the specific content of step S3 is: 1) Use wavelet transform to decompose both the differential enhancement result and the interleaved sequence mapping result into 1 low-frequency component and 3 high-frequency components; 2) Perform average weighted fusion on the high-frequency components: ch = (ch1 + ch2) / 2 cv = (cv1 + cv2) / 2 cd = (cd1 + cd2) / 2 where ch, ch1, and ch2 are the horizontal components of the fused image, the differential enhancement result, and the interleaved sequence mapping result respectively, cv, cv1, and cv2 are the vertical components of the fused image, the differential enhancement result, and the interleaved sequence mapping result respectively, and cd, cd1, and cd2 are the diagonal components of the fused image, the differential enhancement result, and the interleaved sequence mapping result respectively; 3) Perform energy weighted fusion on the low-frequency components: where ca, ca1, and ca2 are the low-frequency components of the fused image, the differential enhancement result, and the interleaved sequence mapping result respectively, cak(n) is the gray value at n on the corresponding low-frequency component, and k = 1, 2; 4) Inverse wavelet transform to obtain the preliminary fused image F 1； 5) Perform dilation and erosion on the interleaved sequence mapping result respectively; 6) Perform weighted fusion on the result and the preliminary fused image to obtain the final fused image: F = F 1 + F d + F e Among them, F is the fusion result, F d is the dilation result, F e is the erosion result.

4. The method for target detection of spectral polarization images based on interleaved sequence mapping according to claim 1, characterized in that , the specific content of step S4 is: Use adaptive threshold segmentation to perform threshold segmentation on the fused image, perform morphological processing on the segmentation result, and then map the detection result back to the target image.

Images