An underwater target laser range-gated search imaging method based on target recognition

Abstract

The present invention provides an underwater target laser range-gated search imaging method based on target recognition. Based on the laser range-gated imaging technology, the imaging strategy is further optimized. First, the target distance search range is divided into multiple ultra-wide distance slice images for rough search. The deep learning target recognition method is used to determine whether there is a target in the ultra-wide distance slice images. When the target appears, equal-step fine scanning is performed within the ultra-wide distance slice image where the target is searched. According to the brightness and contrast of the target ROI area in the rough search result, the target existing in the image is accurately determined, and the specific position of the target is determined, while ensuring the target discrimination accuracy and improving the search efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of laser range-gated imaging, and particularly to an underwater target laser range-gated search imaging method based on target recognition. Background Art

[0002] The laser range-gated imaging technology uses pulsed laser as the active light source and utilizes the time-of-flight gating principle. The detector only receives the optical signal reflected at the target distance, effectively shielding the backscattered signal between the detector and the target. It has the advantages of high signal-to-noise ratio, long detection distance, good imaging effect, etc., and is widely used in fields such as security monitoring, long-distance reconnaissance, and underwater detection. However, the laser range-gated imaging technology belongs to a fixed-distance imaging technology, which requires knowing in advance the distance from the target to be measured to the detector. By adjusting the gating time parameter, it detects at a fixed distance near the target, and the search efficiency is low, bringing great difficulties to practical engineering applications. Especially when the distance of the target is unknown, or multiple targets in the detection range are at different distances, it is necessary to perform range-gated imaging search at different distances in sequence, and the detection real-time performance is poor, and the imaging parameter adjustment is cumbersome.

[0003] In view of the above disadvantages of the range-gated imaging technology, the Ramics airborne laser mine countermeasure system in the United States uses multiple detectors to simultaneously perform range-gated imaging at different distances within the detection range. Although this method is direct and effective, it is costly. Some domestic scholars have proposed a full-gating imaging method based on high-repetition-rate pulsed laser. Although this scheme expands the detection range of range gating, since range-gated imaging is performed at different distances, it will inevitably introduce backscattering, resulting in the dispersion of laser energy, sacrificing a certain detection distance and imaging frame rate, and at the same time, the target distance cannot be obtained.

[0004] Therefore, how to give full play to the advantages of laser range-gated imaging and quickly and accurately search for targets is the difficulty in current underwater target detection. Summary of the Invention

[0005] In order to overcome the disadvantages and deficiencies of the prior art, the present invention provides an underwater target laser range-gated search imaging method based on target recognition, which searches for unknown targets within the detection range by means of regional segmentation, target recognition, and step-by-step search, and has the advantages of large search range, high accuracy, and good real-time performance.

[0006] The present invention provides an underwater target laser range-gated search imaging method based on target recognition, including the following steps:

[0007] Step 1: Divide the target distance search range L into K ultra-wide distance slice images for range-gated imaging, and the gating delay parameters are sequentially set to The gating window width parameters are all set to Among them, L is the target search range, K is the number of ultra-wide distance slice images, and R ref is the refractive index of the medium, and c is the propagation speed of light in the medium;

[0008] Step 2: Use the pre-trained target recognition model to perform target recognition on the ultra-wide distance slice images, and determine whether there is a target to be measured in the ultra-wide distance slice images. If there is a target to be measured, go to Step 3 to perform a subdivision search on the ultra-wide distance slice image where the target to be measured is located. If there is no target to be measured, repeat Steps 1 and 2;

[0009] Step 3: Subdivide the ultra-wide distance slice image with the target to be measured into n narrow distance slice images for narrow distance slice subdivision scanning;

[0010] Step 4: Calculate the average gray value of the ROI region in each narrow distance slice image according to the ROI region obtained from the target recognition result in Step 2 and the standard deviation where P ROI (x, y) is the gray value of each pixel point in the ROI region, and N is the number of pixels;

[0011] Step 5: According to the change curve of I mean , use the peak search method to determine all narrow distance slice images that satisfy I mean (j) > I mean (j - 1), I mean (j) > I mean (j + 1), j = 1, 2, 3... n, and further compare I std (j - 1), I std (j), and I std (j + 1) to determine the narrow distance slice image with the largest standard deviation as the gated image of the target to be measured.

[0012] Furthermore, the target recognition model is a YOLOv4-tiny network model based on the DarkNet framework.

[0013] Furthermore, before Step 2, it also includes:

[0014] Divide different target distance search ranges into several ultra-wide distance slice images respectively and perform target annotation to obtain a training data set of the target to be measured;

[0015] Use the training data set of the target to be measured to train the target recognition model to be trained to obtain a trained target recognition model.

[0016] After adopting the above technical solution, the present invention has at least the following beneficial effects:

[0017] 1. Optimized the target search imaging strategy, applied the deep learning target recognition method in the "coarse scan" within an ultra-wide distance range, and improved the search efficiency.

[0018] 2. Combined the deep learning target recognition and the brightness / contrast target discrimination method, gave full play to their respective advantages, made up for the problems that the brightness / contrast discrimination method is vulnerable to environmental and noise signal interference and the deep learning target recognition method is insensitive to distance information, and improved the accuracy of target discrimination. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0020] Figure 1 It is a schematic flow chart of coarse scanning and fine scanning;

[0021] Figure 2 It is a flow chart of an underwater target laser range-gated search imaging method based on target recognition provided by this embodiment;

[0022] Figure 3 It is a target recognition result map of an ultra-wide distance slice image;

[0023] Figure 4 It is a change curve graph of the average gray value and standard deviation within the ROI area. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0024] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention.

[0025] Refer to Figure 1 and Figure 2 , in this embodiment, a resolution target board (US military standard) is used as the target to be measured and placed at any distance within 30 m from the detector. The position of the target to be measured is found through the underwater target laser range-gated search imaging method based on target recognition proposed in this application for imaging, including:

[0026] First, divide the search ranges of different target distances into several ultra-wide distance slice images in advance and perform target annotation to obtain a training dataset for the target to be measured. Use the training dataset for the target to be measured to train the YOLOv4-tiny network model based on the DarkNet framework to be trained, and obtain a trained YOLOv4-tiny target recognition model based on the DarkNet framework.

[0027] Then, perform the following steps:

[0028] Step 1: Divide the target distance search range of 30m into 4 ultra-wide distance slice images, each slice image with a width of 7.5m, and the rough search ranges are 0 - 7.5m, 7.5 - 15m, 15 - 22.5m, and 22.5 - 30m. According to the formula and calculate the gating parameters of each ultra-wide distance slice image. The gating delay parameters are 0ns, 66.5ns, 133ns, and 199.5ns respectively, and the gating width parameters are all 66.5ns;

[0029] Step 2: Use the pre-trained target recognition model to recognize the 4 ultra-wide distance slice images obtained in Step 1, and determine whether there is a resolution plate in the ultra-wide distance slice images. If there is a resolution plate, as Figure 3 shown, if the resolution plate is determined to be in the range of 22.5 - 30m, then perform Step 3 to perform a detailed search on the ultra-wide distance slice image where the resolution plate is located. If there is no target to be measured, repeat Steps 1 and 2;

[0030] Step 3: Subdivide the ultra-wide distance slice image corresponding to the range of 22.5 - 30m into 8 narrow distance slice images for narrow distance slice subdivision scanning, and set the gating delay parameters to 195ns, 205ns, 215ns... 265ns respectively, and the gating width parameter to 10ns;

[0031] Step 4: According to the ROI region obtained from the target recognition result in Step 2, calculate the average gray value and the standard deviation of each narrow distance slice image obtained in Step 3, and draw the change curve graphs of I mean and I std , as Figure 4 shown;

[0032] Step 5: According to the change curve of I mean , the average gray value of the 5th narrow distance slice image satisfies I mean (5) > I mean (4), I mean (5) > I mean (6). At the same time, the standard deviation I std(5) is the maximum value. Therefore, it can be determined that the resolution plate in the fifth narrow distance slice image corresponds to a distance range of 26.5-27.6 m.

[0033] Through the above-mentioned underwater target laser range-gated search imaging method based on target recognition, the target to be measured at an unknown distance within the range of 30m can be quickly determined for imaging, which not only ensures the accuracy of target discrimination, but also improves the search efficiency.

[0034] Although the present invention has been disclosed as above by way of embodiments, it is not intended to limit the present invention. Any person skilled in the art may make some changes and modifications without departing from the spirit and scope of the present invention. Therefore, the protection scope of the present invention shall be determined by the claims.

Claims

1. An underwater target laser range-gated search imaging method based on target recognition, characterized in that, Including: Step 1: Divide the target distance search range L into K ultra-wide distance slice images for distance-gated imaging, and sequentially set the gating delay parameters as Set the gating width parameters as where L is the target search range, K is the number of ultra-wide distance slice images, R ref is the refractive index of the medium, and c is the propagation speed of light in the medium; Step 2: Use a pre-trained target recognition model to perform target recognition on the ultra-wide distance sliced image, and determine whether there is a target to be measured in the ultra-wide distance sliced image. If there is a target to be measured, proceed to Step 3 to perform a subdivision search on the ultra-wide distance sliced image where the target to be measured is located. If there is no target to be measured, repeat Steps 1 and 2; Step 3: Subdivide the ultra-wide distance sliced image with a target to be measured into n narrow distance sliced images for narrow distance sliced sub-scanning; Step 4: Calculate the average gray value and standard deviation of the ROI region in each narrow-distance slice image according to the ROI region obtained from the target recognition result in Step 2 and standard deviation where P ROI (x, y) is the gray value of each pixel point in the ROI region, and N is the number of pixels; Step 5. According to the change curve of I mean , use the peak searching method to determine all narrow-distance slice images that satisfy I mean (j) > I mean (j - 1), I mean (j) > I mean (j + 1), where j = 1, 2, 3... n, and further compare the magnitudes of I std (j - 1), I std (j), and I std (j + 1). Determine the narrow-distance slice image with the largest standard deviation as the gating image of the target to be measured.

2. The underwater target laser range-gated search imaging method based on target recognition according to claim 1, wherein The target recognition model is a YOLOv4-tiny network model based on the DarkNet framework.

3. The underwater target laser range-gated search imaging method based on target recognition according to claim 2, wherein Before Step 2, it further includes: Divide different target distance search ranges into several ultra-wide distance sliced images respectively and perform target annotation to obtain a training data set of targets to be measured; Use the training data set of targets to be measured to train the target recognition model to be trained to obtain a trained target recognition model.

Images