An improved RPCA infrared small target detection method based on local signal-to-clutter ratio

Abstract

The application discloses an improved RPCA infrared small target detection method based on a local signal-to-clutter ratio. Firstly, a robust principal component analysis is performed on an infrared image containing a small target, and the image is decomposed into a background matrix and a target matrix. The background matrix captures the low rank of the image and describes the background information of the image, and the target matrix utilizes sparsity and effectively represents target information in the image. A local window is introduced in the target matrix, and a local signal-to-clutter ratio of the image in the window is calculated. The local signal-to-clutter ratio considers the difference between the target and the background, and is helpful to distinguish the real target from false alarms. By comparing the local signal-to-clutter ratio difference between the small target and the background in the infrared image, the real target can be accurately identified. The robust principal component analysis can effectively separate the target and the background in the image, thereby improving the detection rate of the small target.

Description

Technical Field

[0001] This invention relates to the fields of image processing, target detection, and image segmentation, and specifically to an improved RPCA infrared small target detection method based on local signal-to-noise ratio. Background Technology

[0002] Local Signal-to-Clutter Ratio (LSCR), as a method for evaluating the relationship between signal and clutter, is of great significance in infrared small target detection. Its main purpose is to enhance the difference between the target and the background, thereby improving detection efficiency. This technique calculates the LSCR within a local region of the target matrix by introducing a local window. This helps to effectively distinguish targets and background in an image, thus enhancing detection accuracy and reliability, and improving detection speed. However, if the window slides element-wise, it can significantly increase computation time, especially with large image sizes.

[0003] Robust Principal Component Analysis (RPCA), a matrix dimensionality reduction and decomposition technique applied in image processing, has significant theoretical and practical value. In practical applications, RPCA effectively separates image content into two parts—background and target—by representing an image as a combination of a background matrix and a target matrix. The background matrix carries common information in the image, while the target matrix presents local non-common structures. This decomposition method captures different components in the image while exhibiting robustness to noise and outliers. However, the processing results of RPCA may be affected by strong noise. The presence of noise can disrupt the true structure of the data, complicating the decomposition process and reducing accuracy. Therefore, by using the local signal-to-noise ratio, a more accurate target representation can be provided for RPCA decomposition, effectively improving the detection rate of small infrared targets. Summary of the Invention

[0004] The purpose of this invention is to provide an improved RPCA infrared small target detection method based on local signal-to-noise ratio, which can effectively suppress background interference and improve the success rate of target detection in infrared small target detection.

[0005] The technical solution adopted in this invention is: an improved RPCA infrared small target detection method based on local signal-to-clutter ratio, comprising the following steps:

[0006] Step (1): Obtain a single-frame original image of a small infrared target;

[0007] Step (2): Perform RPCA on the original image to obtain the background matrix and the target matrix;

[0008] Step (3): Use the iterative threshold method to segment the target matrix, calculate all 8-connected connected components in the segmented target matrix, and remove connected components whose number of pixels does not conform to the definition of infrared small target;

[0009] Step (4): Construct a local window for each remaining connected component after removal, and calculate the signal-to-noise ratio (SNR) of the local window, i.e., the local SNR.

[0010] Step (5): Compare the signal-to-noise ratio of each local window to obtain the true target.

[0011] Furthermore, in step (1), the method is used to process infrared images where the target area is usually no more than 25 pixels, and this is referred to as the original image.

[0012] Furthermore, in step (2), the original image is subjected to RPCA using the Accelerated Proximal Gradient Algorithm to obtain a background matrix and a target matrix.

[0013] Further, in step (3), the global threshold of non-zero elements in the target matrix is ​​calculated by the iterative threshold algorithm. This threshold is used for image segmentation to remove the influence of noise and background clutter. After image segmentation based on the threshold, the number of pixels contained in each 8-connected connected component in the image is calculated. Based on the definition of infrared small targets in step (1), connected components with a non-compliant number of pixels are removed.

[0014] Further, in step (4), for the connected components remaining after being removed in step (3), the size of the connected component with the largest number of pixels is calculated and denoted as m×n. For each connected component remaining after being removed in step (3), a local window with a fixed size of (2m+1)×(2n+1) is created, and the center point of each remaining connected component coincides with the center point of the corresponding local window. The local signal-to-noise ratio of each window is calculated.

[0015] Furthermore, in step (5), since the local window containing the real target carries target information, the calculated signal-to-noise ratio should be much higher than that of the local window that does not contain the real target, and the real target is obtained by using this characteristic.

[0016] The advantages of this invention compared to the prior art are:

[0017] (1) This invention introduces local signal-to-noise ratio into the RPCA target detection algorithm, which significantly improves the target detection rate and robustness of the method.

[0018] (2) The present invention introduces an iterative thresholding method to automatically calculate the global threshold of the image, eliminating the need to manually set the threshold coefficient and increasing the convenience of the method.

[0019] (3) The present invention optimizes the selection logic of the local window and strengthens the comparison of the local signal-to-noise ratio between the real target and the false target, thereby selecting the real target more accurately.

[0020] (4) This invention proposes a threshold calculation method based on local signal-to-noise ratio, which can automatically and effectively segment and obtain the real target based on the local signal-to-noise ratio. Attached Figure Description

[0021] Figure 1 This is a flowchart of the present invention;

[0022] Figure 2 The detection results are from different algorithms;

[0023] Figure 3 The algorithmic SCRG value is the original image. Detailed Implementation

[0024] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0025] like Figure 1 As shown, the target detection method process is represented by a flowchart.

[0026] The specific steps of the improved RPCA infrared small target detection method based on local signal-to-clutter ratio described in this invention are as follows:

[0027] Step (1): Obtain a single-frame original image of a small infrared target;

[0028] Step (2): Perform RPCA on the original image to obtain the background matrix and the target matrix;

[0029] According to the theory of Principal Component Analysis (PCA), the original image D can be decomposed as follows:

[0030] D = B + T

[0031] Where B is the background matrix with low rank, and T is the target matrix with sparsity.

[0032] According to RPCA theory, the above equation can be expressed as a convex optimization problem:

[0033] min||B|| * +λ||T||1 stD=B+T

[0034] Among them, ||·|| *Let ||·||1 represent the nuclear norm, ||·||1 represent the l1 norm, and λ (λ>0) be the regularization coefficient. The convex optimization problem described above is solved using the accelerated proximal gradient algorithm to obtain the background matrix B and the objective matrix T.

[0035] Step (3): Use the iterative threshold method to segment the target matrix, calculate all 8-connected connected components in the segmented target matrix, and remove connected components whose number of pixels does not conform to the definition of infrared small target;

[0036] For the target matrix T, first calculate the initial threshold T0 of T:

[0037]

[0038] Among them, T max and T min These are the maximum and minimum values ​​of the target matrix T, respectively. Based on T0, T can be divided into two parts: less than T0 and greater than T0, denoted as t1 and t2 respectively.

[0039] t1 = T (T < T0)

[0040] t2 = T (T > T0)

[0041]

[0042] The `mean(·)` operator represents the averaging operation, where `t0` is the expected average of the two parts. This process is iterated repeatedly, continuously updating `t`. n Value (n≥0), when |t n+1 -t n When the value is less than the set threshold (the threshold value in this method is 0.01), the global threshold th is taken as t. n+1 ,Right now:

[0043] th = t n+1

[0044] TL = T (T > th)

[0045] TL is the target matrix after global thresholding. This allows for the initial removal of non-target pixels, reducing the computational cost of local signal-to-noise ratio and improving method efficiency. Calculate the 8-connected components in TL. Calculate the number of pixels contained in each connected component. According to the definition in step (1), connected components containing more than 25 pixels are considered non-real target connected components.

[0046] Step (4): Construct a local window for each remaining connected component after removal, and calculate the signal-to-noise ratio (SNR) of the local window, i.e., the local SNR.

[0047] After removing non-true target connected components, calculate the size of the remaining connected component containing the most pixels, denoted as m×n. Simultaneously, create a local window of fixed size (2m+1)×(2n+1) for each remaining connected component, ensuring the center point of each remaining connected component coincides with the center point of the corresponding local window. Within the local window, everything except the remaining connected components is considered local background. Calculate the signal-to-noise ratio of the local window:

[0048]

[0049] Where LSCR represents the calculated signal-to-noise ratio of the local window, μ tar μ represents the mean gray value of the remaining connected components. back σ represents the average gray level of a local background region. back This represents the standard deviation of grayscale values ​​in a local background area.

[0050] Step (5): Compare the signal-to-noise ratio of each local window to obtain the true target.

[0051] Because the amount of information in a local window containing the real target is much greater than the amount of information in a local window not containing the real target, the difference in their local signal-to-noise ratios is significant. Calculate the mean and standard deviation of the signal-to-noise ratio for all remaining connected components.

[0052]

[0053] Among them, th LSCR It is the signal-to-noise ratio threshold used to segment and obtain the true target, μ LSCR It is the mean of the confidence-to-noise ratio, σ LSCR It is the standard deviation of the signal-to-noise ratio. For all remaining connected components obtained in step (3), if the local signal-to-noise ratio of a certain remaining connected component is greater than th... LSCR If the connected component is true, then the connected component is considered the real target.

[0054] To demonstrate the effectiveness and detection capability of this method, three infrared small target images with complex backgrounds were selected for verification and compared with other commonly used target detection methods. The other three comparison algorithms were: Local Contrast Measure (LCM), Top-Hat filtering, and RPCA. The detection comparison results are as follows: Figure 2 As shown in the figure, column (a) is the original image, column (b) is the result of the LCM algorithm, column (c) is the result of the Top-Hat algorithm, column (d) is the result of the RPCA algorithm, and column (e) is the result of this method.

[0055] To quantitatively compare the merits of different algorithms, a signal-to-clutter ratio (SCRG) gain is introduced:

[0056]

[0057] The signal-to-clutter ratio (SCR) is expressed as follows:

[0058]

[0059] Where, μ t μ is the mean gray value of the target region. b σ is the average grayscale value of the background area. b represents the standard deviation of the grayscale values ​​in the background region. SCRG indicates the enhancement effect of the algorithm on the target; the larger the SCRG value, the better the detection effect of the algorithm.

[0060] Figure 3 The SCRG values ​​in the table show the performance of each algorithm on the target algorithm. Figure 2 (a) Detection results for three original images. Combined with... Figure 2 and Figure 3 This demonstrates that the proposed method has good detection capability for small infrared targets against complex backgrounds.

[0061] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications or substitutions that can be conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of the present invention. Contents not described in detail in this specification are prior art known to those skilled in the art.

Claims

1. An improved RPCA infrared small target detection method based on local signal-to-clutter ratio, characterized in that: This method is based on local signal-to-noise ratio and low-rank sparsity theory, and includes the following steps: Step (1): Obtain a single-frame original image of a small infrared target; Step (2): Perform RPCA on the original image to obtain the background matrix and the target matrix; Step (3): Use the iterative threshold method to segment the target matrix, calculate all 8-connected connected components in the segmented target matrix, and remove connected components whose pixel count does not conform to the definition of infrared small target; Step (4): Construct a local window for each remaining connected component after removal, and calculate the signal-to-noise ratio (SNR) of the local window, i.e., the local SNR; After removing non-real target connected components, calculate the size of the connected component with the largest number of pixels among the remaining connected components, denoted as m×n. Simultaneously, create a local window of fixed size (2m+1)×(2n+1) for each remaining connected component, ensuring that the center point of each remaining connected component coincides with the center point of the corresponding local window. Within the local window, all areas except the remaining connected components are local background regions. Calculate the signal-to-noise ratio of the local window: in, This represents the calculated signal-to-noise ratio of the local window. This represents the average gray value of the remaining connected components. This represents the average gray level of a local background area. This represents the standard deviation of gray levels in a local background area; Step (5): Compare the signal-to-noise ratio of each local window to obtain the true target.

2. The improved RPCA infrared small target detection method based on local signal-to-clutter ratio according to claim 1, characterized in that: In step (1), the method is used to process infrared images where the target area is no more than 25 pixels, and this is called the original image.

3. The improved RPCA infrared small target detection method based on local signal-to-clutter ratio according to claim 1, characterized in that: In step (2), the original image is subjected to RPCA using the Accelerated Proximal Gradient Algorithm to obtain a background matrix and a target matrix.

4. The improved RPCA infrared small target detection method based on local signal-to-clutter ratio according to claim 1, characterized in that: In step (3), the global threshold of non-zero elements in the target matrix is ​​calculated by the iterative threshold algorithm. This threshold is used for image segmentation to remove the influence of noise and background clutter. After image segmentation based on the threshold, the number of pixels contained in each 8-connected connected component in the image is calculated. The original image of a single-frame infrared small target is obtained according to step (1), and connected components with inconsistent pixel counts are removed.

5. The improved RPCA infrared small target detection method based on local signal-to-clutter ratio according to claim 1, characterized in that: In step (5), since the local window containing the real target carries target information, the calculated signal-to-noise ratio is higher than that of the local window that does not contain the real target. The real target is obtained by using this characteristic.

Images