A dark and weak target detection method based on multi-stage noise suppression

Abstract

The application provides a dark and weak target detection method based on multi-stage noise suppression. In view of the problem that the signal-to-noise ratio of the target and the contrast between the target and the background are low, and the target occupies a small number of pixels in the whole image, and the target detection is difficult, the dark and weak target is detected by using the steps of background elimination, threshold segmentation, large block noise elimination, small particle noise elimination and target repair. The application has the beneficial effects that the method can successfully detect the target in the case that the contrast between the target and the background is low in a single frame image, and solves the problem of target energy leakage.

Description

Technical Field

[0001] This invention belongs to the field of image processing, specifically relating to a method for detecting dim targets based on multi-level noise suppression. Background Technology

[0002] For detection systems, when the distance between the target and the detector reaches tens of kilometers, the imaged target exhibits characteristics of being "weak" and "small" due to the influence of external factors such as atmospheric disturbance, optical scattering, and diffraction. "Weak" primarily refers to the target's low signal-to-noise ratio and low contrast with the background; "small" mainly refers to the target occupying a small number of pixels in the entire image. Low signal-to-noise ratio images often show weak targets in various scenes, including the sky, ground, and sea, making them difficult to distinguish amidst natural clutter, resulting in high false alarm rates and low detection rates for target detection.

[0003] The main challenges in detecting dimly lit targets are as follows:

[0004] • Low target contrast results in limited features available for target detection. At long distances, the target signal strength received by the detector is limited, leading to a very low signal-to-noise ratio. The target imaging area is small, sometimes only a few pixels, thus lacking contour and shape features, and providing relatively little usable information.

[0005] • In complex backgrounds, targets are often submerged in the background or noise. Distant targets are usually affected by strong backgrounds, such as a strong sky background during the day.

[0006] Current literature contains numerous methods for extracting faint targets. For example, grayscale thresholding, including global and local thresholding, is effective for targets with high signal-to-noise ratios (SNR), but struggles with targets with low SNR and low contrast. Accumulation methods, which overlay multiple frames to enhance target information, are unsuitable for fast-moving targets. Other literature utilizes the differences and singularities in structural features such as local image intensity, gradient, and divergence between the target and the background / noise for faint target detection; however, these methods are computationally intensive and unsuitable for real-time detection and tracking. Summary of the Invention

[0007] The technical problem this invention aims to solve is to address the impact of strong sky backgrounds on the detection of weak targets, and to provide a weak target detection method based on multi-level noise suppression. This method utilizes steps such as background removal, threshold segmentation, large-block noise removal, small-particle noise removal, and target restoration to detect weak targets. Actual measurements show that the method achieves good detection performance on target images against a strong sky background.

[0008] The technical solution adopted in this invention is: a method for detecting weak targets in dark environments based on multi-level noise suppression, the specific implementation steps of which are as follows:

[0009] Step 1: Perform median filtering on the original image, and then subtract the filtered image from the original image to obtain the background suppression map;

[0010] Step 2: Calculate the global threshold of the background suppression map, and subtract the threshold from the background suppression map to obtain the noise-suppressed map and the segmented bright area map;

[0011] Step 3: Calculate the area of ​​each connected region in the segmented bright area map, set the pixels of connected regions with an area greater than the area threshold to zero, and then multiply them by the noise-suppressed map to obtain the result map.

[0012] Step 4: Perform binary segmentation on the result image from Step 3 to obtain a secondary segmented bright area image;

[0013] Step 5: Using the secondary segmentation bright area image as a seed, repair the target using the watershed algorithm.

[0014] Furthermore, the global threshold for calculating the background suppression map in step 2 can be implemented using the Otsu algorithm.

[0015] The advantages of this invention compared to the prior art are as follows:

[0016] (1) This invention solves the problem of target energy leakage;

[0017] (2) Even if the contrast between the target and the background is low in a single frame image, the target can still be successfully detected.

[0018] (3) Even if the target is stuck to the background, the target edge information can be successfully restored. Attached Figure Description

[0019] Figure 1 This is the original image of a faint target collected in the embodiment;

[0020] Figure 2 It is a 3D image of the original image of the faint target in the embodiment;

[0021] Figure 3 This is the background suppression image after median filtering in the embodiment;

[0022] Figure 4 This is the noise reduction image after subtracting the threshold in the embodiment;

[0023] Figure 5 This is the segmented bright area map after threshold segmentation in the embodiment;

[0024] Figure 6 This is a binary image after removing large blocks of noise in the embodiment;

[0025] Figure 7 This is a result image after eliminating large blocks of noise in the embodiment;

[0026] Figure 8 This is a secondary segmentation bright area image after eliminating small particle noise in the embodiment;

[0027] Figure 9 This is the target binary image repaired using the watershed algorithm in the embodiment;

[0028] Figure 10 This is a target detection result image repaired by the watershed algorithm in the embodiment;

[0029] Figure 11 This is the target detection result image marked in the original image in the embodiment;

[0030] Figure 12 This is a flowchart of the low-light target detection method based on multi-level noise suppression according to the present invention. Detailed Implementation

[0031] The technical solution provided by the present invention will be described in detail below with reference to specific embodiments. These embodiments are only for explaining the present invention and are not intended to limit it. The scope of protection of the invention should include all the contents of the claims. All similar implementations that achieve the same effect through substantially the same method fall within the scope of protection of the present invention. Through the following embodiments, those skilled in the art can implement all the contents of the claims of the present invention.

[0032] This embodiment uses an image acquired in a low-light target detection experiment as an example, such as... Figure 1 As shown. It can be seen that due to the strong background light, the target is barely visible. The 3D representation of the original image is as follows. Figure 2 As shown, the target is almost completely obscured by the background, thus hindering target detection. Figure 12 As shown, the specific implementation steps of the low-light target detection method based on multi-level noise suppression of the present invention are as follows.

[0033] Step 1: Perform median filtering on the original image, and then subtract the filtered image from the original image to obtain the background suppression map.

[0034] This step is as follows: Figure 1 Taking the dim target image shown as an example, a median filter is applied to it, and then the filtered image is subtracted from the original image to obtain a background suppression image, as shown. Figure 3 As shown.

[0035] Step 2: Calculate the global threshold of the background suppression map, subtract the threshold from the background suppression map to obtain the noise-suppressed map, and perform binary segmentation on the background suppression map to obtain the segmented bright area map.

[0036] This example uses the Otsu algorithm to calculate the global threshold of the background suppression map in step 1, and then subtracts this threshold from the background suppression map to obtain the denoised map, as shown below. Figure 4As shown, the background suppression map is binary segmented with this threshold to obtain a segmented bright area map, as Figure 5 shown.

[0037] In this embodiment, the Otsu algorithm is used to calculate the global threshold. Let {0, 1, 2, …, L-1} represent the L different gray levels in the image, and n i represent the number of pixels with gray level i. Then the probability p i of the gray level i in the image is: where n is the total number of pixels in the image.

[0038] Assume that the threshold th is selected, 0 < th < L-1. The image is thresholded into two classes C1 and C2. Among them, C1 consists of all pixels with gray values within [0, th], and C2 consists of all pixels with gray values within [th+1, L-1]. Then the probability P1 that a pixel is classified into C1 is: The probability P2 that it is classified into C2 is: The average gray value m1 of the pixels assigned to C1 is: Similarly, the average gray value m2 of the pixels assigned to C2 is: The average gray value m G of the entire image is: Finally, the between-class variance is:

[0039]

[0040] As can be seen from formula (1), the greater the difference between the two class means m1 and m2, the greater the between-class variance . It shows that the between-class variance is a measure of the separability between classes. Therefore, the optimal threshold th * needs to be obtained so that the between-class variance is the largest, that is:

[0041]

[0042] Step 3: Calculate the area of each connected component in the segmented bright area map, set the pixel points of the connected components larger than the area threshold to zero, and then multiply it by the noise suppression map to obtain the result map.

[0043] In this step, the area of each connected component in the segmented bright area map is calculated, and the pixel points of the connected components larger than the area threshold are set to zero to obtain a binary map after eliminating large block noises, as Figure 6 shown. Then, it is used as a mask to multiply the noise suppression map obtained in step 2 to obtain a result map after eliminating large block noises, as Figure 7 shown.

[0044] Step 4: Calculate the global threshold of the result map in step 3 and perform binary segmentation on the noise suppression map to obtain a secondarily segmented bright area map.

[0045] This embodiment uses the Otsu algorithm to calculate the result image of step 3 (i.e., Figure 7 The global threshold of ) is used, and binary segmentation is performed to obtain a secondary segmented bright area map, such as Figure 8 As shown.

[0046] Step 5: Using the secondary segmentation bright area image as a seed, repair the target using the watershed algorithm.

[0047] This embodiment uses the secondary segmented bright area image obtained in step 4 as a seed, and employs the watershed algorithm, using the segmented bright area image as a mask, to find the watershed on the background suppression image, thus obtaining the repaired target binary image, as shown below. Figure 9 As shown, the target is multiplied by the background suppression map from step 1 using the target as a mask to obtain the repaired target, as shown. Figure 10 As shown, the target is marked with a box in the original image, such as... Figure 11 As shown.

[0048] The parts of this invention not described in detail are well-known to those skilled in the art.

Claims

1. A method for detecting dim and weak targets based on multi-stage noise suppression, characterized in that, The implementation steps are as follows: Step 1, median filtering is performed on the original image, and then the filtered image is subtracted from the original image to obtain a background suppression image; Step 2, the global threshold of the background suppression image is calculated, the background suppression image is subtracted from the threshold to obtain a noise suppression image, and the background suppression image is binarized to obtain a segmented bright area image; Step 3, the area of each connected domain in the segmented bright area image is calculated, the pixel points of the connected domain larger than the area threshold are set to zero, and then the result image is multiplied by the noise suppression image to obtain a result image; Step 4, the result image of step 3 is binarized to obtain a secondary segmented bright area image; Step 5, the secondary segmented bright area image is used as a seed to repair the target by using the watershed algorithm.

2. The method according to claim 1, wherein the method is characterized by: The global threshold of the background suppression image in step 2 can be realized by using the Otsu algorithm.

Images