A SAR Imaging Detection Method Based on Local Contrast Enhancement in the Cosine Transform Domain

Abstract

This invention discloses a SAR imaging detection method based on local contrast enhancement in the cosine transform domain, applied to the field of synthetic aperture radar detection and imaging technology. It addresses the problem of poor detection performance in existing technologies when background noise is strong or spatially uneven. This invention enhances the target region using an AC coefficient, then performs threshold segmentation using inverse discrete cosine transform to complete target detection. Experimental verification and analysis show that this method achieves good detection performance and high accuracy even under complex background noise conditions.

Description

Technical Field

[0001] This invention belongs to the field of synthetic aperture radar detection and imaging technology, and specifically relates to a technique in the field of synthetic aperture radar detection and imaging technology. Background Technology

[0002] Synthetic Aperture Radar (SAR) can be used to detect and image ships at sea. Due to the angular reflection of the ship's material and structure, ship targets are generally bright in SAR images, corresponding to large values ​​in the grayscale image. When the sea surface is calm, target ships can be easily detected. However, when the sea surface changes drastically, or when the location of the sea area is different, the background noise of the sea surface will be enhanced and become uneven, making the detection of target ships more difficult.

[0003] Currently, there is little research in the field of radar imaging target detection on target saliency extraction in the compressed domain. Many studies on ship detection using synthetic aperture radar are primarily based on the human visual system. In the paper "Chen C.LP, Li H, Wei Y, et al. A Local Contrast Method for Small Infrared Target Detection[J]. IEEE Transactions on Geoscience and Remote Sensing, vol.52, no.1, pp.574-581, Jan.2014," a detection algorithm based on local contrast measurement was proposed. This algorithm enhances the target and suppresses background noise by measuring the difference between the target area and its surrounding area. However, this method performs poorly when background noise is strong or spatially uneven, and because it processes each sliding window holistically, the detection results cannot well preserve the original target's morphological features. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention proposes a SAR imaging detection method based on local contrast enhancement in the cosine transform domain. This method utilizes the DC and AC coefficients of the discrete cosine transform to extract the target intensity and edge information from the synthetic aperture radar (SAR) imaging results. This information is then used to enhance the target region and suppress background noise, thereby highlighting the target for detection and overcoming the deterioration caused by the intensity and distribution of background noise. Furthermore, by leveraging the enhancement characteristics in the compressed domain, the method maintains the target morphological features before and after processing, resulting in better enhancement effects and accuracy.

[0005] The technical solution adopted in this invention is: a SAR imaging detection method based on local contrast enhancement in the cosine transform domain, comprising:

[0006] S1. Divide the SAR image to be processed into several pixel blocks of equal size, and perform two-dimensional DCT on each pixel block;

[0007] S2. Take the 2D DCT result corresponding to the top left corner pixel of each pixel block as the DC coefficient, and the 2D DCT result corresponding to the other pixels as the AC coefficient.

[0008] S3. A sliding window P of size M×M is used to process the SAR image after two-dimensional DCT processing, where M represents the number of pixel blocks after two-dimensional DCT processing; the gain coefficient of the pixel block located at the center in the current sliding window P is calculated, and the pixel block located at the center in the current sliding window P is enhanced based on the gain coefficient.

[0009] S4. Perform block-based inverse discrete cosine transform on the SAR image after sliding window processing to obtain the saliency map;

[0010] S5. Threshold segmentation is performed on the saliency map to binarize the image; the target morphology features in the radar image are obtained.

[0011] The beneficial effects of this invention are as follows: This invention enhances potential target regions in synthetic aperture radar images using two-dimensional discrete cosine transform (DCT) with AC and DC coefficients. After enhancement, a two-dimensional inverse DCT is performed to obtain a saliency map, and adaptive threshold segmentation is performed based on the image mean and standard deviation to finally obtain the target morphological features in the radar image. Simulation results show that the method of this invention can effectively enhance target regions under strong and complex background noise while maintaining the original morphological features of the target well, which is beneficial for further target detection and recognition. Attached Figure Description

[0012] Figure 1 This is a flowchart of the algorithm of the present invention.

[0013] Figure 2 This is a schematic diagram of the sliding window unit of the present invention.

[0014] Figure 3 This represents the spatial contrast symbolized by each AC coefficient after the discrete cosine transform.

[0015] Figure 4 The images show the original image, saliency map, and imaging result against a uniform clutter background.

[0016] Among them, (a) is the original SAR image, (b) is the saliency image of the present invention, and (c) is the result image of the present invention.

[0017] Figure 5 The images show the original image, saliency map, and imaging result against a non-uniform clutter background.

[0018] Among them, (a) is the original SAR image, (b) is the saliency image of the present invention, and (c) is the result image of the present invention.

[0019] Figure 6 The images show the original image, saliency map, and imaging result against a background of strong clutter.

[0020] Among them, (a) is the original SAR image, (b) is the saliency image of the present invention, and (c) is the result image of the present invention.

[0021] Figure 7 To enhance the suppression effect of the algorithm of this invention;

[0022] Among them, (a) is the original SAR image, (b) is the saliency image, (c) is the original 3D SAR image, and (d) is the 3D saliency image.

[0023] Figure 8 for Figure 5 The original image, the actual image, and the image result of a ship target;

[0024] Among them, (a) is the original SAR image of the ship target, (b) is the image with correct data annotation, and (c) is the target shape after processing. Detailed Implementation

[0025] To facilitate understanding of the technical content of this invention by those skilled in the art, the following description, in conjunction with the accompanying drawings, further illustrates the invention.

[0026] This invention primarily employs simulation experiments to verify the effectiveness of the proposed synthetic aperture radar imaging and detection algorithm based on local contrast enhancement in the cosine transform domain. All steps and conclusions were verified correctly using the MATLAB 2021a platform on a Windows 10 operating system. To facilitate understanding of the technical content of this invention by those skilled in the art, the following description, in conjunction with accompanying drawings and tables, further elaborates on the invention.

[0027] like Figure 1 As shown, a SAR imaging detection method based on local contrast enhancement in the cosine transform domain according to the present invention includes the following steps:

[0028] Step 1: Synthetic Aperture Radar Image Block Discrete Cosine Transform

[0029] To better demonstrate the effectiveness of the algorithm in handling different background clutter conditions, experiments were conducted on SAR images with uniform clutter, non-uniform clutter, and strong clutter, respectively. Figure 4 (a) Figure 5 (a) Figure 6 As shown in (a), its data information is listed in Table 1.

[0030] Table 1. Experimental Image Feature Parameters Table

[0031] Experimental images Number of ships Background noise pixel size Figure 4 (a) 15 uniform 1000×1000 Figure 5 (a) 7 Uneven 1000×1000 Figure 6 (a) 8 strong 1000×1000

[0032] In general image processing, sliding windows are frequently used. This involves setting up a window that slides across the pixels of an image, processing each pixel as it slides. For example... Figure 2 As shown, when P slides through the radar image, if the features of the central cell C are different from those of its neighboring cell blocks, C can be enhanced to make the target stand out.

[0033] This invention combines the ideas of sliding window and discrete cosine transform, both of which are block-based processing methods, and performs sliding window processing in the compressed domain after dividing the image into blocks and performing two-dimensional discrete cosine transform.

[0034] The international standard for image compression, JPEG, uses 8×8 pixel blocks for DCT transformation. This local transformation helps preserve local image details during compression, avoiding information loss that might occur with global transformations. Simultaneously, it provides sufficient resolution in the frequency domain to represent image details while maintaining high efficiency during encoding. This invention also uses 8×8 pixel blocks to preserve local image details and improve computational efficiency.

[0035] As explained above, the 1000×1000 pixel experimental image is divided into 8×8 pixel blocks, and a two-dimensional discrete cosine transform (DCT) is performed on each pixel block. The formula for the two-dimensional DCT is as follows:

[0036]

[0037] Where the constant coefficient C u C v The size depends on the position in the DCT coefficient block.

[0038]

[0039] f(i,j) represents the pixel value of an 8×8 pixel block, and its position (i,j) ranges from 0 to 7; F uv This is the calculation result, representing the DCT coefficients after 2D DCT. The subscripts u and v represent the positions of the DCT coefficients within a single 8×8 DCT coefficient block, with values ​​ranging from 0 to 7. For example, in the 8×8 DCT coefficient block obtained from the 2D DCT of each pixel block, the top-left corner F... 00 The DC coefficient represents the average strength of the entire piece; the rest, F... uvThe AC coefficient is used. Unlike the classic LCM algorithm, to ensure the subsequent IDCT, the central unit C of the image after block DCT cannot be changed to a single value, but rather the whole unit is scaled. At the same time, the scaling degree of the central unit C is determined by its own DC coefficient and the AC coefficients of its surrounding neighboring units. The specific scaling is shown in formula (5).

[0040] In this embodiment, a sliding window P is formed by every 9 adjacent pixel blocks, i.e., M=3, where the central unit is C and the neighboring units are Cells. The sliding window slides across the image to cover each pixel with a step size of L pixels, while continuously processing the pixels of the current central unit. If the features of the central unit C are different from those of the neighboring units, it is enhanced to make the target stand out.

[0041] Step 2: Define the gain coefficient and calculate the saliency plot.

[0042] According to formula (1), the AC coefficient can be calculated:

[0043]

[0044] It can be seen that the AC coefficient F 10 The value depends on the strength difference between the upper and lower parts of the unit block, and the other four AC coefficients are similar, such as... Figure 3 As shown. When F 10 A large absolute value indicates that there is a clear upper and lower boundary within the unit block; a small absolute value indicates that the intensity within the unit block tends to be uniform.

[0045] As derived above, the absolute value of the AC coefficients can indicate whether there are obvious boundaries within the DCT block. To simplify the calculation, only F is considered. 10 ,F 01 ,F 20 ,F 02 The four AC coefficients can symbolize the spatial distribution under different conditions, such as Figure 3 As the sliding window P slides block by block from left to right and from top to bottom on the image after DCT, the gain coefficient K of the current central cell is calculated, defined as follows:

[0046]

[0047] Where DC is the DC coefficient of the central unit C, and dividing by the window size equals its mean μ. C Cell i,j For the cell located at (i,j) in a 3×3 sliding window, F k The corresponding F in this unit 10 ,F 01 ,F 20 ,F 02 Four AC coefficients.

[0048] As the sliding window P slides block by block across the DCT-generated image from left to right and top to bottom, the central cell C is scaled as a whole, and the enhancement value at each central cell is defined as...

[0049] C E =K×C (5)

[0050] Equation (5) represents C E The K value is determined by both the DC and AC coefficients. DC reflects the intensity of the central cell itself, while AC reflects whether there is a clear boundary around the central cell. When the central cell is located in the background, the K value is small regardless of the strength of the background clutter; when it is located in or around the target, the K value is large, which can enhance the target area.

[0051] Calculate the enhancement value C of the central unit of the entire image. E Then, a block-based inverse discrete cosine transform is performed on it, transforming it back to the image domain to obtain the saliency map, as shown below. Figure 4 (b) Figure 5 (b) Figure 6 (b) After this enhancement process, regardless of the uniformity or intensity of the background clutter, it has been essentially completely suppressed. The target enhancement effect in the saliency map can be observed more intuitively in the 3D image, such as... Figure 7 .

[0052] Step 3: Thresholding and Binarizing the Image

[0053] After obtaining the saliency map, the prominent regions in the scene are likely the targets. To highlight the target detection results, thresholding is performed on the entire image to binarize it. An adaptive threshold is used.

[0054] T = μ + k th ×σ (6)

[0055] Where μ and σ are the mean and standard deviation of the significance plot, respectively, and k th Let k be a fixed empirical coefficient. In the three images of the experiment, k is set... th A value of 10 yields excellent segmentation results. The binarization result is as follows: Figure 4 (c) Figure 5 (c) Figure 6 .(c).

[0056] To more objectively measure the performance of the algorithm of this invention, Tables 2-4 present various metrics and compare them with other algorithms. LCM is the local contrast measurement algorithm, DCTASD is the DCT domain saliency detection algorithm, and MVWIE is the multi-scale variance-weighted image entropy algorithm.

[0057] Table 2. Advantage values ​​and variances of each algorithm

[0058] image LCM DCTASD MVWIE Algorithm of this invention Figure 4 0.405 1 0.933 1 Figure 5 <0.100 0.636 1 0.875 Figure 6 0.091 0.625 0.889 0.875 FOM variance 0.0377 0.0455 0.0031 0.0052

[0059] Table 3. Signal-to-noise ratio gain of each algorithm

[0060]

[0061] Table 4 Background suppression factors for each algorithm

[0062] image LCM DCTASD MVWIE Algorithm of this invention Figure 4 0.61 0.57 1.71 2.18 Figure 5 0.67 0.49 1.96 8.59 Figure 6 0.69 0.60 1.01 2.42

[0063] Table 2 lists the Figure of Merit (FOM) values ​​for each algorithm. FOM can be used to evaluate detection performance; it is the ratio of the number of correctly detected targets to the sum of actual and false alarm targets. A higher FOM value indicates a higher detection rate and a lower false alarm rate. FOM is defined as follows:

[0064]

[0065] Where, N tt N represents the number of targets that were correctly detected. fa N represents the number of false alarms. gt This represents the number of real targets present in the image. The algorithm of this invention boasts a high FOM value and achieves good detection results under various background environments. Furthermore, variance analysis of the FOM value reveals the robustness of the algorithm; the FOM variance of this invention's algorithm remains at a low level, indicating its strong robustness in target detection under different conditions.

[0066] Target enhancement and background suppression capabilities are metrics for evaluating algorithm effectiveness. Tables 3 and 4 present the signal-to-noise ratio gain (SCRG, describing the algorithm's target enhancement capability) and background suppression factor (BSF, describing the algorithm's background suppression capability) for different algorithms, defined as follows:

[0067]

[0068] Where, σ in and σ out These are the standard deviations of the original image and the saliency map, respectively; SCR out and SCR in The signal-to-noise ratios of the saliency map and the original image are defined as follows:

[0069]

[0070] In the formula, μ t μ represents the average pixel value of a single target region. b and σ b These represent the average pixel value and standard deviation of the target neighborhood background, respectively.

[0071] Table 3 selected each Figures 4-6 The signal-to-noise ratio gain of the five ship target regions was analyzed, and Table 4 calculated the background suppression factor of the overall image before and after processing. As can be seen from Tables 3 and 4, the method of this invention has the highest SCRG and BSF values, indicating that it has better target enhancement and background suppression capabilities.

[0072] At the same time, the algorithm of this invention can effectively preserve the original morphological characteristics of the ship target while enhancing it. For example... Figure 8 As shown, the selected Figure 5 A ship target area is shown in Table 5. The correct data annotation map (actual shape) of this target is basically similar to the target shape processed by this invention. Figure 8 The measurement index value corresponding to the detection result of the target. Defined as follows:

[0073]

[0074] Wherein, TP represents the number of true positive targets, TN represents the number of true negative targets, FP represents the number of false positive targets, and FN represents the number of false negative targets. Precision can characterize, to a certain extent, the effectiveness of the detected targets in maintaining their original morphological characteristics. Compared with other algorithms, the method proposed in this invention shows better performance in terms of precision, accuracy, false negative rate, and false alarm rate.

[0075] Table 5. Algorithm Pairs Figure 8 Detection index value of the target

[0076]

[0077] Those skilled in the art will recognize that the embodiments described herein are intended to help the reader understand the principles of the invention, and should be understood that the scope of protection of the invention is not limited to such specific statements and embodiments. Various modifications and variations can be made to the invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the invention should be included within the scope of the claims of the invention.

Claims

1. A SAR imaging detection method based on local contrast enhancement in the cosine transform domain, characterized in that, include: S1. Divide the SAR image to be processed into several pixel blocks of equal size, and perform two-dimensional DCT on each pixel block; S2. Take the 2D DCT result corresponding to the top left corner pixel of each pixel block as the DC coefficient, and the 2D DCT result corresponding to the other pixels as the AC coefficient. S3. A sliding window P of size M×M is used to process the SAR image after two-dimensional DCT processing, where M represents the number of pixel blocks after two-dimensional DCT processing. By calculating the gain coefficient of the pixel block located at the center in the current sliding window P, the pixel block located at the center in the current sliding window P is enhanced based on the gain coefficient. S4. Perform block-based inverse discrete cosine transform on the SAR image after sliding window processing to obtain the saliency map; S5. Threshold segmentation is performed on the saliency map to binarize the image; the target morphology features in the radar image are obtained.

2. The SAR imaging detection method based on local contrast enhancement in the cosine transform domain according to claim 1, characterized in that, The expression for the gain coefficient is: Where DC is the DC coefficient of the central pixel block in the current sliding window P, μ C Cell is the average value of the central pixel block in the current sliding window P. i,j Let F be the pixel block located at (i,j) in a sliding window P of size M×M, where i,j=0,2,…,M-1. k Corresponding Cell i,j The exchange coefficient in the equation.

3. The SAR imaging detection method based on local contrast enhancement in the cosine transform domain according to claim 2, characterized in that, The formula for calculating the DCT coefficients in each pixel block is: Among them, F uv This represents the DCT coefficient of the pixel located at (u,v) in the pixel block, where u,v = 0, 1, 2, ..., L-1, and L represents the size of the pixel block. u C v f(i,j) is a constant coefficient, representing the pixel value at position (i,j) in the L×L pixel block, with the position (i,j) ranging from 0 to L-1.

4. The SAR imaging detection method based on local contrast enhancement in the cosine transform domain according to claim 3, characterized in that, constant coefficient C u C v The value can be:

5. The SAR imaging detection method based on local contrast enhancement in the cosine transform domain according to claim 4, characterized in that, In step S5, an adaptive threshold is used, expressed as follows: T=μ+k th ×s Where μ and σ are the mean and standard deviation of the significance plot, respectively, and k th It is a fixed empirical coefficient.

Images