Vehicle-mounted SAR Target Detection and Recognition Method, System and Terminal

Abstract

The present invention discloses a vehicle-mounted SAR target detection and recognition method, system and terminal. The method includes: performing imaging processing on vehicle-mounted SAR echo data to obtain a vehicle-mounted SAR image; performing superpixel segmentation on the vehicle-mounted SAR image; performing threshold detection on the segmented vehicle-mounted SAR image to obtain a target binary image, and performing optimization processing using prior target size information to obtain a more accurate target binary image; cropping the framed target image into a sample image; inputting the sample image into a convolutional neural network model for training; and using the trained network model for prediction, and marking the prediction result in the SAR image. The technical effect of the present invention is that, due to the adoption of prior target size information, parts that are obviously inconsistent with the target size are excluded, thus greatly reducing the workload in subsequent detection; second, due to the adoption of rotating the target after framing it, the redundant information introduced by the inability of the target detection frame to rotate in the traditional method is overcome.

Description

Technical Field

[0001] The present invention relates to the technical field of computer processing, and particularly to a vehicle-mounted SAR target detection and recognition method, system, terminal and computer-readable storage medium. Background Art

[0002] SAR (Synthetic Aperture Radar), that is, synthetic aperture radar, is an active earth observation system that can be installed on flight platforms such as airplanes, satellites, and spacecraft, and can observe the ground all day and all weather, and has a certain ground penetration ability.

[0003] Due to the advantages of all-day and all-weather operation, high resolution and non-decreasing with distance increase of synthetic aperture radar (SAR), it has been widely used in military and civilian fields. However, the image characteristics of SAR images in the prior art will change greatly with different imaging parameters, imaging postures, ground object environments, etc., making the target detection and recognition of SAR images very difficult. The target detection and recognition of SAR images in the prior art have large computational amounts, high hardware requirements, and complex implementation.

[0004] Therefore, the prior art still needs to be improved and developed. Summary of the Invention

[0005] The main purpose of the present invention is to provide a vehicle-mounted SAR target detection and recognition method, system, terminal and computer-readable storage medium, aiming to solve the problems of large computational amounts, high hardware requirements, and complex implementation in the target detection and recognition of SAR images in the prior art.

[0006] To achieve the above purpose, the present invention provides a vehicle-mounted SAR target detection and recognition method, and the vehicle-mounted SAR target detection and recognition method includes the following steps:

[0007] Perform imaging processing on the vehicle-mounted SAR echo data to obtain an original vehicle-mounted SAR image;

[0008] Segment the original vehicle-mounted SAR image by using the SLIC algorithm in the superpixel generation algorithm to obtain a segmented vehicle-mounted SAR image;

[0009] Perform threshold detection on the segmented vehicle-mounted SAR image, and obtain a target binary image after morphological operations;

[0010] Perform optimization processing on the target binary image obtained after morphological operations by using prior target size information to obtain an optimized target binary image;

[0011] Based on the optimized target binary image, frame the target on the original vehicle-borne SAR image, and crop the framed target image into a sample image;

[0012] Input the sample image into a convolutional neural network model for neural network training; and use the trained network model for prediction, and mark the prediction results in the SAR image.

[0013] The vehicle-borne SAR target detection and recognition method, wherein the step of segmenting the original vehicle-borne SAR image by using the SLIC algorithm in the superpixel generation algorithm to obtain the segmented vehicle-borne SAR image includes;

[0014] According to the size of the target to be detected, set an appropriate initial size of the superpixel, segment the original vehicle-borne SAR image, and divide the background and the detected target into different superpixels to obtain the segmented vehicle-borne SAR image.

[0015] The vehicle-borne SAR target detection and recognition method, wherein the step of performing threshold detection on the segmented vehicle-borne SAR image and obtaining a target binary image after morphological operations;

[0016] For the segmented vehicle-borne SAR image, adopt a global threshold method, use the superpixel as the basic unit, and use the overall average power of the image as the threshold for threshold detection to obtain a target binary image;

[0017] Perform morphological processing on the obtained target binary image to obtain a target binary image after morphological processing.

[0018] The vehicle-borne SAR target detection and recognition method, wherein the step of optimizing the target binary image obtained after morphological operations by using prior target size information to obtain an optimized target binary image includes:

[0019] Use prior target size information to perform pruning processing on the target binary image obtained after morphological operations, delete the parts that do not meet the basic target size, and retain the parts that are basically the same as the target size to obtain an optimized target binary image.

[0020] The vehicle-borne SAR target detection and recognition method, wherein the step of framing the target on the original vehicle-borne SAR image based on the optimized target binary image and cropping the framed target image into a sample image includes;

[0021] For the optimized target binary image, extract the target on the original vehicle-borne SAR image and rotate the target so that the target is vertical in the square;

[0022] Crop the framed target image into a sample image.

[0023] The described vehicle-mounted SAR target detection and recognition method, wherein the steps of extracting the target from the original vehicle-mounted SAR image and rotating the target to make it vertical to the box for the optimized target binary image include:

[0024] Define a two-dimensional matrix with the same size as the box, and fill the elements in the matrix with the vehicle-mounted SAR echo data at the corresponding positions of the box;

[0025] Sum the matrix row by row to obtain a one-dimensional vector;

[0026] Take the maximum value of the one-dimensional vector as the ordinate and the rotation angle as the abscissa for plotting; the angle corresponding to the highest point of the graph is the required target rotation angle.

[0027] The described vehicle-mounted SAR target detection and recognition method, wherein the steps of inputting the sample image into a convolutional neural network model for neural network training; and using the trained network model for prediction and marking the prediction result in the SAR image include:

[0028] Input the sample image into a convolutional neural network model for neural network training; and use the trained network model for prediction and mark the prediction result in the SAR image to display the position information of the target relative to the radar.

[0029] A vehicle-mounted SAR target detection and recognition system, wherein the vehicle-mounted SAR target detection and recognition system includes:

[0030] An echo data processing module for performing imaging processing on the vehicle-mounted SAR echo data to obtain the original vehicle-mounted SAR image;

[0031] An image segmentation module for segmenting the original vehicle-mounted SAR image using the SLIC algorithm in the superpixel generation algorithm to obtain the segmented vehicle-mounted SAR image;

[0032] A threshold detection module for performing threshold detection on the segmented vehicle-mounted SAR image and obtaining the target binary image after morphological operations;

[0033] An optimization processing module for optimizing the target binary image obtained after morphological operations using the prior target size information to obtain the optimized target binary image;

[0034] A target boxing and cropping module for boxing the target on the original vehicle-mounted SAR image based on the optimized target binary image and cropping the boxed target image into a sample image;

[0035] A target prediction module for inputting the sample image into a convolutional neural network model for neural network training, and using the trained network model for prediction and marking the prediction result in the SAR image.

[0036] A terminal, wherein the terminal includes: a memory, a processor, and a vehicle-mounted SAR target detection and recognition program stored on the memory and executable on the processor. When the vehicle-mounted SAR target detection and recognition program is executed by the processor, the steps of any one of the vehicle-mounted SAR target detection and recognition methods are implemented.

[0037] A computer-readable storage medium, wherein the computer-readable storage medium stores a vehicle-mounted SAR target detection and recognition program. When the vehicle-mounted SAR target detection and recognition program is executed by a processor, the steps of any one of the vehicle-mounted SAR target detection and recognition methods are implemented.

[0038] The present invention provides a vehicle-mounted SAR target detection method combining an image processing method and a deep learning method. First, the vehicle-mounted SAR echo data is subjected to imaging processing to obtain an original vehicle-mounted SAR image, and target detection is performed on the basis of this image. The main process is as follows: the SLIC algorithm in the superpixel generation algorithm is used to segment the image. On this basis, threshold detection is performed with the superpixel as the basic unit to obtain a target binary image, and then some morphological processing is carried out; then, using the prior target size information, parts that are obviously inconsistent with the target size can be excluded, thereby reducing the workload in subsequent detection; next, the target is framed and extracted. Since the target has a certain tilt angle in the vehicle-mounted SAR image, the target needs to be rotated to make it vertical in the square; then, a neural network is used to improve the detection accuracy. The small image extracted in the above steps is used as a sample, and MobileNet-V2 is used for training, and then the trained network is used for prediction. The result is marked in the SAR image to display the position information of the target relative to the radar. The technical effects of the present invention are as follows: since the prior target size information is adopted to exclude parts that are obviously inconsistent with the target size, the workload in subsequent detection is greatly reduced; second, since the target is rotated after being framed, the redundant information introduced by the inability to rotate the target detection frame in the traditional method is overcome. Description of the Drawings

[0039] Figure 1 is a flowchart of a preferred embodiment of the vehicle-mounted SAR target detection and recognition method of the present invention.

[0040] Figure 2 is an algorithm flowchart of a preferred embodiment of the vehicle-mounted SAR target detection and recognition method of the present invention

[0041] Figure 3It is a schematic diagram of the result after superpixel segmentation in a preferred embodiment of the vehicle-mounted SAR target detection and recognition method of the present invention.

[0042] Figure 4 It is a schematic diagram of the binary target image after threshold detection and morphological operation in the vehicle-mounted SAR target detection and recognition method of the present invention.

[0043] Figure 5 It is a schematic diagram of the binary target image after deleting parts with obviously inconsistent sizes in the vehicle-mounted SAR target detection and recognition method of the present invention.

[0044] Figure 6 It is a schematic diagram of the extracted small image in the vehicle-mounted SAR target detection and recognition method of the present invention.

[0045] Figure 7 It is a schematic diagram of the final marked result in the vehicle-mounted SAR target detection and recognition method of the present invention.

[0046] Figure 8 It is a functional principle block diagram of the vehicle-mounted SAR target detection and recognition system according to an embodiment of the present invention.

[0047] Figure 9 It is a schematic diagram of the terminal implementation principle according to an embodiment of the present invention. Specific embodiments

[0048] The following describes clearly and completely the technical solutions in the embodiments of the present invention with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of protection of the present invention.

[0049] Many specific details are set forth in the following description in order to provide a thorough understanding of the present invention, but the present invention may be practiced in other ways different from those described herein. Those skilled in the art can make similar extensions without departing from the connotation of the present invention, so the present invention is not limited by the specific embodiments disclosed below.

[0050] SAR (Synthetic Aperture Radar), that is, synthetic aperture radar, is an active earth observation system that can be installed on flight platforms such as airplanes, satellites, and spacecraft, and can conduct observations on the ground all day and all weather, and has a certain ground penetration ability.

[0051] Due to the advantages of Synthetic Aperture Radar (SAR) such as all-weather and all-day operation, high resolution and non-decrease with distance increase, it has been widely applied in military and civilian fields. However, the image characteristics of SAR images vary greatly with different imaging parameters, imaging postures, ground object environments, etc., making the target detection and recognition of SAR images very difficult.

[0052] Before 2014, it was called the "traditional target detection period", and the CFAR (Constant False Alarm Rate) detection algorithm and its derivative algorithms were the most widely used in SAR image target detection methods. However, this method requires a high contrast between the target and the background clutter, and assumes that the background clutter satisfies a certain statistical model distribution, which has great limitations. After 2014, with the explosive development of deep learning, many deep learning-based detection technologies emerged, called the "deep learning-based detection period". Deep learning requires a large amount of data, huge computational volume, and high hardware requirements, while high-resolution SAR images often have large picture sizes; and the model design and training of deep learning are very complex. In recent years, deep learning algorithms have achieved great success in the field of optical target detection, and scholars from various countries have promoted these optical detection algorithms to the research of SAR image detection methods, showing excellent detection performance.

[0053] Among them, in some literature "SAR Image Target Detection Algorithm Based on Convolutional Neural Network", the detection performance after expanding the existing complete data set under the condition of insufficient training samples was studied, and the results proved that this method can obtain good detection effects. However, expanding the data will bring a certain number of false alarms, affecting the detection performance.

[0054] In the literature "SAR Vehicle Target Detection Based on CNN", a scheme for SAR vehicle target detection in complex backgrounds based on the Faster-RCNN framework was proposed. However, deep neural networks are more complex, have high hardware requirements, and low detection efficiency.

[0055] To solve the problems of the prior art, the solution of the present invention provides a vehicle-mounted SAR target detection method combining image processing methods and deep learning methods. First, the vehicle-mounted SAR echo data is subjected to imaging processing to obtain the original vehicle-mounted SAR image, and target detection is performed on this image. The main process is as follows: The SLIC algorithm in the superpixel generation algorithm is used to segment the image. On this basis, threshold detection is performed with superpixels as the basic unit to obtain the target binary image, and then some morphological processing is carried out; Then, using the prior target size information, parts that are clearly not in line with the target size can be excluded, thereby reducing the workload during subsequent detection; Next, the target is framed and extracted. Since the target has a certain tilt angle in the vehicle-mounted SAR image, the target needs to be rotated so that it is vertical in the square; Then, a neural network is used to improve the detection accuracy. The small images extracted in the above steps are used as samples, trained using MobileNet-V2, and then the trained network is used for prediction. The results are marked in the SAR image, showing the position information of the target relative to the radar.

[0056] The advantages of the present invention are as follows: First, by using the prior target size information, parts that are clearly not in line with the target size are excluded, thereby greatly reducing the workload during subsequent detection; Second, after framing the target, the target is rotated to overcome the redundant information introduced by the inability to rotate the target detection frame in the traditional method.

[0057] The following further describes the present invention through specific embodiments. The vehicle-mounted SAR target detection and recognition method described in the preferred embodiment of the present invention is as Figure 1 shown, and the vehicle-mounted SAR target detection and recognition method includes the following steps:

[0058] Step S100: Obtain vehicle-mounted SAR echo data, and perform imaging processing on the vehicle-mounted SAR echo data to obtain the original vehicle-mounted SAR image.

[0059] In the present invention, referring to Figure 1 and Figure 2 shown, the vehicle-mounted SAR echo data is obtained, and the vehicle-mounted SAR (synthetic aperture radar) echo data is subjected to imaging processing to obtain the vehicle-mounted SAR (synthetic aperture radar) image, that is, the vehicle-mounted radar image.

[0060] Step S200: Use the SLIC algorithm in the superpixel generation algorithm to segment the original vehicle-mounted SAR image to obtain the segmented vehicle-mounted SAR image.

[0061] In the present invention, according to the size of the target to be detected, an appropriate initial size of the superpixel is set, and the original vehicle-mounted SAR image is subjected to superpixel segmentation, dividing the background and the detected target into different superpixels to obtain the segmented vehicle-mounted SAR image;

[0062] This step mainly uses the method of superpixel segmentation for superpixel segmentation. For example, the present invention uses the SLIC algorithm, which is the most widely used superpixel generation algorithm, to segment the vehicle-mounted SAR image. According to the size of the target to be detected, a suitable initial superpixel size is set (in the present invention, taking the bicycle target as an example, the initial superpixel size is set to 20), the background and the target are divided into different superpixels, and the more the boundaries of the superpixels cover the target edges, the better. As Figure 3 shown, it is a schematic diagram of the result after superpixel segmentation.

[0063] The SLIC algorithm, also known as the method of superpixel segmentation, includes various operations such as contour extraction, clustering, and gradient ascent. The SLIC superpixel segmentation algorithm proposed by the present invention (simple linear iterative clustering) is one of them. It is based on the K-means clustering algorithm and performs clustering according to the color and distance features of pixels to achieve good segmentation results. The SLIC algorithm has the advantages of simplicity, flexibility, good effect, and fast processing speed.

[0064] In the superpixel segmentation of the embodiment of the present invention, in the field of computer vision, image segmentation refers to the process of subdividing a digital image into multiple image sub-regions (sets of pixels) (also known as superpixels). A superpixel is a small region composed of a series of adjacent pixels with similar features such as color, brightness, and texture. Most of these small regions retain the effective information for further image segmentation and generally do not damage the boundary information of the objects in the image.

[0065] Step S300: Perform threshold detection on the segmented vehicle-mounted SAR image with superpixels as the basic unit to obtain a target binary image;

[0066] In this step, mainly threshold detection is performed on the segmented vehicle-mounted SAR image. As Figure 4 shown, Figure 4 is a schematic diagram of the target binary image after threshold detection and morphological operation in the vehicle-mounted SAR target detection and recognition method of the present invention. The power at the location of the target is greater than that at the location without the target. Using the threshold method, detection is performed with superpixels as the basic unit to obtain a target binary image. All the points with pixel value 1 in the target binary image are a target region. In the threshold detection method of the embodiment of the present invention, it can be divided into two categories: global threshold method and local threshold method. The present invention adopts a simple global threshold method and uses the overall average power of the image as the threshold for detection.

[0067] Step S400: Perform morphological processing on the obtained target binary image to obtain a target binary image after morphological processing;

[0068] In the embodiment of the present invention, morphological processing is performed on the obtained target binary image. For the target binary image, the dilation operation in morphological operations is used to connect the broken single targets into a complete target; the erosion operation is used to break the multiple targets connected together.

[0069] Step S500: For the target binary image obtained after morphological operations, optimize it using the prior target size information to obtain an optimized target binary image.

[0070] In this step, for the target binary image after morphological processing, according to the prior target size information, the parts that do not meet the size are deleted, and the parts that are basically the same as the size of the detected target are retained to form a target binary image.

[0071] For example, in this step, for the SAR image after morphological processing, the size information is extracted. If the size of the vehicle-mounted SAR image is M×N and the corresponding actual distance is X×Y, that is, the actual physical size represented by each pixel is: Taking the bicycle target as an example, assuming the size of the bicycle is x×y, in the vehicle-mounted SAR image, the corresponding number of pixels is: Considering that the target has a certain tilt angle θ in the image, the size of a bicycle target in the radar image is Therefore, according to the prior target size information, some parts that obviously do not meet the approximate size of the target can be deleted, and only the parts that are approximately the same as the target size are retained.

[0072] In the embodiment of the present invention, the prior target size information is the information known before the experiment. For example, if the target to be detected is a bicycle, knowing its general actual length (prior information) can calculate how many pixels it corresponds to on the radar image.

[0073] In this step, for the target binary image obtained after morphological operations, using the prior target size information can exclude the parts that obviously do not meet the target size, thereby reducing the workload in subsequent detection. For example Figure 5 The figure shows the target binary image after deleting the parts with obviously inconsistent sizes according to the prior target size information. In this step, for the target binary image obtained after morphological operations, optimizing it using the prior target size information can obtain a more accurate target binary image.

[0074] Step S600: Based on the optimized target binary image, frame the target on the original vehicle-mounted SAR image and crop the framed target image into a sample image;

[0075] In the present invention, the target binary image after pruning processing is framed, and the framed part is cropped as a new image (small image), which is also called a sample image in the present invention.

[0076] That is, in the embodiment of the present invention, for the optimized target binary image, the target is extracted from the original vehicle-mounted SAR image and rotated so that the target is vertical in the square; the framed target image is cropped into a sample image. The present invention adopts a method for determining the target rotation angle to overcome the redundant information introduced by the inability to rotate the target detection frame in the traditional method.

[0077] In this step, for example, in order to completely frame the target, the size of the frame should be set slightly larger than the target. Taking the bicycle target as an example, the size of the frame can be set as where a is slightly greater than 1. Since the target has a certain inclination angle in the vehicle-mounted SAR image, after the target is framed, the target is rotated by a certain angle so that the target is vertical in the middle of the frame.

[0078] In the present invention, the rotation angle can be determined by the following method: Define a two-dimensional matrix with the same size as the square frame, and fill the elements in the matrix with the vehicle-mounted SAR echo data corresponding to the positions of the square frame. Then sum the matrix by rows to obtain a one-dimensional vector; take the maximum value of the one-dimensional vector as the ordinate and the rotation angle as the abscissa to plot a graph; the angle corresponding to the highest point of the graph is the required target rotation angle. Other methods such as the Hough transform can also be used to determine the rotation angle.

[0079] As Figure 6 shown, Figure 6 is a schematic diagram of the small image extracted by the vehicle-mounted SAR target detection and recognition method of the present invention. Figure 6 Among them, the small image is the extracted one. Some of the small images are targets and some are not. Among the framed small images, some small images are not targets. Next, the method of deep learning is used to improve the accuracy of target detection.

[0080] Step S700: Input the sample image into a convolutional neural network model for neural network training; and use the trained network model for prediction, and mark the prediction result in the SAR image to display the position information of the target relative to the radar.

[0081] In the present invention, the small image extracted by the above steps is used as a sample image to be input into a convolutional neural network for training, that is, the sample image is input into a lightweight convolutional neural network for neural network training. The adopted lightweight convolutional neural network is MobileNet-V2. Testing the trained network, the accuracy rate reaches 83%.

[0082] Among them, the lightweight convolutional neural network MobileNet-V2 was proposed by the Google team in 2018. Compared with MobileNetV1, it has higher accuracy and a smaller model. The lightweight convolutional neural network MobileNetV2 is based on an inverted residual structure, and the skip connections are located between the thinner bottleneck layers. The intermediate expansion layer uses lightweight depth convolution to extract features and introduce non-linearity. Moreover, in order to maintain the representation ability of the network, the author removed the non-linear activation function of the narrower layers.

[0083] In the present invention, the extracted SAR targets are trained by a neural network, and the predicted results are displayed in the vehicle-mounted SAR image, and the positions of the targets are marked.

[0084] The present invention uses the trained network for prediction, and the predicted results are displayed in the vehicle-mounted SAR image, thereby marking the positions of the targets.

[0085] In the embodiment of the present invention, regarding the position information of the target center relative to the radar. For example, if a SAR image is regarded as a two-dimensional plane coordinate system with the radar position as the origin of the coordinate system, then the position where a certain target center is located can be represented by the coordinates (m, n), that is, the target center is m pixels horizontally away from the radar and n pixels vertically away from the radar. Substituting into the formula for the actual physical size represented by each pixel in step 4, it can be known that the position of the target relative to the radar is The straight-line distance is

[0086] As Figure 7 shown, it is a schematic diagram of the final marked result of the embodiment of the present invention. Taking the radar as the origin, the position information of the target relative to the radar is shown. That is, in the embodiment of the present invention, finally, according to the pixel position where the target center is located, its position relative to the radar is calculated, converted into meters, and the radar is used as the coordinate origin.

[0087] The effect of the present invention can be further illustrated by Figures 2 - 7 simulation experiments. When simulating, MATLAB software is used for simulation (MATLAB software is a commercial mathematical software produced by MathWorks Company in a certain country, and is used in fields such as data analysis, wireless communication, deep learning, image processing and computer vision, signal processing, quantitative finance and risk management, robotics, control systems, etc.), and Python is used for the training of the neural network (Python provides efficient high-level data structures).

[0088] However, this method is more suitable for detecting smaller SAR image targets in complex large scenes.

[0089] As can be seen from the above: By combining image processing and deep learning, the present invention detects targets in vehicle-mounted SAR images, marks the target positions and displays them. Since the optimized target binary image is obtained by pruning the target binary image obtained after morphological operations according to the prior target size information, the detection workload is reduced; the method for determining the target rotation angle crops the framed target image into a sample image, overcoming the redundant information introduced by the inability to rotate the target detection box in the traditional method, reducing the target detection and recognition calculation amount of the SAR image, improving the accuracy of target detection, and realizing complexity.

[0090] Further, as Figure 8 shown, based on the above vehicle-mounted SAR target detection and recognition method, the present invention also correspondingly provides a vehicle-mounted SAR target detection and recognition system, wherein the vehicle-mounted SAR target detection and recognition system includes:

[0091] An echo data processing module 81, configured to perform imaging processing on vehicle-mounted SAR echo data to obtain an original vehicle-mounted SAR image;

[0092] An image segmentation module 82, configured to segment the original vehicle-mounted SAR image by using the SLIC algorithm in the superpixel generation algorithm to obtain a segmented vehicle-mounted SAR image;

[0093] A threshold detection module 83, configured to perform threshold detection on the segmented vehicle-mounted SAR image and obtain a target binary image after morphological operations;

[0094] An optimization processing module 84, configured to optimize the target binary image obtained after morphological operations by using prior target size information to obtain an optimized target binary image;

[0095] A target framing and cropping module 85, configured to frame a target on the original vehicle-mounted SAR image based on the optimized target binary image and crop the framed target image into a sample image;

[0096] A target prediction module 86, configured to input the sample image into a convolutional neural network model for neural network training; and use the trained network model for prediction, and mark the prediction result in the SAR image, as specifically described above.

[0097] Further, as Figure 9 shown, based on the above vehicle-mounted SAR target detection and recognition method and system, the present invention also correspondingly provides a terminal, and the terminal includes a processor 10, a memory 20, and a display 30. Figure 9 Only some components of the terminal are shown, but it should be understood that it is not required to implement all the shown components, and more or fewer components can be alternatively implemented.

[0098] In some embodiments, the memory 20 may be an internal storage unit of the terminal, such as the hard disk or memory of the terminal. In some other embodiments, the memory 20 may also be an external storage device of the terminal, such as a plug-in hard disk equipped on the terminal, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, etc. Further, the memory 20 may also include both the internal storage unit of the terminal and the external storage device. The memory 20 is used to store application software installed on the terminal and various types of data, such as program codes for installing the terminal. The memory 20 may also be used to temporarily store data that has been output or will be output. In one embodiment, a vehicle-mounted SAR target detection and recognition program 40 is stored on the memory 20, and the vehicle-mounted SAR target detection and recognition program 40 can be executed by the processor 10, so as to implement the vehicle-mounted SAR target detection and recognition method in the present application.

[0099] In some embodiments, the processor 10 may be a central processing unit (CPU), a microprocessor or other data processing chips, and is used to run the program codes stored in the memory 20 or process data, such as executing the vehicle-mounted SAR target detection and recognition method, etc.

[0100] In some embodiments, the display 30 may be an LED display, a liquid crystal display, a touch liquid crystal display, and an OLED (Organic Light-Emitting Diode) toucher, etc. The display 30 is used to display information on the terminal and to display a visual user interface. The components 10-30 of the terminal communicate with each other through a system bus.

[0101] In one embodiment, when the processor 10 executes the vehicle-mounted SAR target detection and recognition program 40 in the memory 20, the following steps are implemented:

[0102] Perform imaging processing on the vehicle-mounted SAR echo data to obtain an original vehicle-mounted SAR image;

[0103] Use the SLIC algorithm in the superpixel generation algorithm to segment the original vehicle-mounted SAR image to obtain a segmented vehicle-mounted SAR image;

[0104] Perform threshold detection on the segmented vehicle-mounted SAR image, and obtain a target binary image after morphological operations;

[0105] For the obtained target binary image after morphological operations, use the prior target size information for optimization processing to obtain the optimized target binary image;

[0106] Based on the optimized target binary image, frame the target on the original vehicle-mounted SAR image, and crop the framed target image into a sample image;

[0107] Input the sample image into a convolutional neural network model for neural network training; and use the trained network model for prediction, and mark the prediction results in the SAR image.

[0108] Among them, the step of segmenting the original vehicle-mounted SAR image using the SLIC algorithm in the superpixel generation algorithm to obtain the segmented vehicle-mounted SAR image includes;

[0109] According to the size of the target to be detected, set an appropriate initial size of the superpixel, segment the original vehicle-mounted SAR image, and divide the background and the detected target into different superpixels to obtain the segmented vehicle-mounted SAR image.

[0110] Among them, perform threshold detection on the segmented vehicle-mounted SAR image, and obtain the target binary image after morphological operations;

[0111] For the segmented vehicle-mounted SAR image, adopt the global threshold method, use the superpixel as the basic unit, and use the overall average power of the image as the threshold for threshold detection to obtain the target binary image;

[0112] Perform morphological processing on the obtained target binary image to obtain the target binary image after morphological processing.

[0113] Among them, the step of using the prior target size information to optimize the target binary image obtained after morphological operations to obtain the optimized target binary image includes:

[0114] Use the prior target size information to perform pruning processing on the target binary image obtained after morphological operations, delete the parts that do not meet the basic target size, and retain the parts that are basically the same as the target size to obtain the optimized target binary image.

[0115] Among them, the step of framing the target on the original vehicle-mounted SAR image based on the optimized target binary image and cropping the framed target image into a sample image includes;

[0116] For the optimized target binary image, extract the target on the original vehicle-mounted SAR image and rotate the target so that the target is vertical in the square;

[0117] Crop the framed target image into a sample image.

[0118] Among them, the steps of extracting a target from the original vehicle-borne SAR image and rotating the target to make the target vertical to the box for the optimized target binary image include:

[0119] Define a two-dimensional matrix with the same size as the box, and fill the elements in the matrix with the vehicle-borne SAR echo data at the corresponding positions of the box;

[0120] Sum the matrix by rows to obtain a one-dimensional vector;

[0121] Take the maximum value of the one-dimensional vector as the ordinate and the rotation angle as the abscissa for plotting; the angle corresponding to the highest point of the graph is the required target rotation angle.

[0122] Among them, the steps of inputting the sample image into a convolutional neural network model for neural network training; and using the trained network model for prediction and marking the prediction result in the SAR image include:

[0123] Input the sample image into a convolutional neural network model for neural network training; and use the trained network model for prediction, mark the prediction result in the SAR image, and display the position information of the target relative to the radar.

[0124] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a vehicle-borne SAR target detection and recognition program, and when the vehicle-borne SAR target detection and recognition program is executed by a processor, the steps of the vehicle-borne SAR target detection and recognition method described above are implemented.

[0125] In summary, the present invention provides a vehicle-mounted SAR target detection method that combines image processing methods and deep learning methods. First, the vehicle-mounted SAR echo data is subjected to imaging processing to obtain the original vehicle-mounted SAR image, and target detection is performed on this image. The main process is as follows: The SLIC algorithm in the superpixel generation algorithm is used to segment the image. On this basis, threshold detection is performed with superpixels as the basic unit to obtain the target binary image, and then some morphological processing is carried out; then, using the prior target size information, parts that are clearly inconsistent with the target size can be excluded, thereby reducing the workload in subsequent detection; next, the target is framed and extracted. Since the target has a certain tilt angle in the vehicle-mounted SAR image, the target needs to be rotated so that it is vertical in the square; then, a neural network is used to improve the detection accuracy. The small images extracted in the above steps are used as samples, and MobileNet-V2 is used for training, and then the trained network is used for prediction, and the results are marked in the SAR image to display the position information of the target relative to the radar. The technical effects of the present invention are as follows: Since the prior target size information is adopted to exclude parts that are clearly inconsistent with the target size, the workload in subsequent detection is greatly reduced; second, since the target is rotated after being framed, the redundant information introduced by the inability to rotate the target detection frame in the traditional method is overcome.

[0126] It should be noted that in this article, the terms "include", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or terminal including a series of elements not only includes those elements but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or terminal. Without further limitation, an element defined by the phrase "including a..." does not exclude the existence of additional identical elements in the process, method, article or terminal including the element.

[0127] Of course, those of ordinary skill in the art can understand that all or part of the processes of implementing the methods in the above embodiments can be completed by instructing relevant hardware (such as a processor, a controller, etc.) through a computer program. The program can be stored in a computer-readable storage medium readable by a computer, and when the program is executed, it can include the processes of the above method embodiments. The computer-readable storage medium can be a memory, a magnetic disk, an optical disk, etc.

[0128] It should be understood that the application of the present invention is not limited to the above examples. For those of ordinary skill in the art, improvements or transformations can be made according to the above description, and all such improvements and transformations should fall within the protection scope of the appended claims of the present invention.

Claims

1. A vehicle-mounted SAR target detection and recognition method, characterized in that, the vehicle-mounted SAR target detection and recognition method includes: Performing imaging processing on the vehicle-mounted SAR echo data to obtain an original vehicle-mounted SAR image; Using the SLIC algorithm in the superpixel generation algorithm to segment the original vehicle-mounted SAR image to obtain a segmented vehicle-mounted SAR image; Performing threshold detection on the segmented vehicle-mounted SAR image and obtaining a target binary image after morphological operations; Performing optimization processing on the target binary image obtained after morphological operations by using prior target size information to obtain an optimized target binary image; Based on the optimized target binary image, framing the target on the original vehicle-mounted SAR image and cropping the framed target image into a sample image; Inputting the sample image into a convolutional neural network model for training; and using the trained network model for prediction and marking the predicted results in the SAR image.

2. The vehicle-mounted SAR target detection and recognition method according to claim 1, characterized in that, the step of using the SLIC algorithm in the superpixel generation algorithm to segment the original vehicle-mounted SAR image to obtain a segmented vehicle-mounted SAR image includes; According to the size of the target to be detected, setting an appropriate initial size of the superpixel, segmenting the original vehicle-mounted SAR image, and dividing the background and the detected target into different superpixels to obtain a segmented vehicle-mounted SAR image.

3. The vehicle-mounted SAR target detection and recognition method according to claim 1, characterized in that, performing threshold detection on the segmented vehicle-mounted SAR image and obtaining a target binary image after morphological operations; For the segmented vehicle-mounted SAR image, adopting a global threshold method, using the superpixel as the basic unit and the overall average power of the image as the threshold for threshold detection to obtain a target binary image; Performing morphological processing on the obtained target binary image to obtain a target binary image after morphological processing.

4. The vehicle-mounted SAR target detection and recognition method according to claim 1, characterized in that, the step of performing optimization processing on the target binary image obtained after morphological operations by using prior target size information to obtain an optimized target binary image includes: Using prior target size information to perform optimization processing on the target binary image obtained after morphological operations, deleting the parts that do not meet the basic target size, and retaining the parts that are basically the same as the target size to obtain an optimized target binary image.

5. The vehicle-mounted SAR target detection and recognition method according to claim 1, characterized in that, the step of framing the target on the original vehicle-mounted SAR image based on the optimized target binary image and cropping the framed target image into a sample image includes; For the optimized target binary image, extracting the target on the original vehicle-mounted SAR image and rotating the target to make the target vertical in the square; Cropping the framed target image into a sample image.

6. The vehicle-mounted SAR target detection and recognition method according to claim 5, characterized in that, The steps of extracting the target from the original vehicle-mounted SAR image and rotating the target to make it vertical in the square for the optimized target binary image include: Define a two-dimensional matrix with the same size as the square, and fill the elements in the matrix with the vehicle-mounted SAR echo data at the corresponding positions of the square; Sum the matrix row by row to obtain a one-dimensional vector; Take the maximum value of the one-dimensional vector as the ordinate and the rotation angle as the abscissa for plotting; the angle corresponding to the highest point of the graph is the required target rotation angle.

7. The vehicle-mounted SAR target detection and recognition method according to claim 1, wherein, The steps of inputting the sample image into a convolutional neural network model for neural network training; and using the trained network model for prediction and marking the prediction result in the SAR image include: Input the sample image into a convolutional neural network model for training; and use the trained network for prediction and mark the prediction result in the SAR image to display the position information of the target relative to the radar.

8. A vehicle-mounted SAR target detection and recognition system, wherein, The vehicle-mounted SAR target detection and recognition system includes: An echo data processing module for performing imaging processing on the vehicle-mounted SAR echo data to obtain the original vehicle-mounted SAR image; An image segmentation module for segmenting the original vehicle-mounted SAR image using the SLIC algorithm in the superpixel generation algorithm to obtain the segmented vehicle-mounted SAR image; A threshold detection module for performing threshold detection on the segmented vehicle-mounted SAR image and obtaining a target binary image after morphological operations; An optimization processing module for optimizing the target binary image obtained after morphological operations using prior target size information to obtain an optimized target binary image; A target bounding and cropping module for bounding the target on the original vehicle-mounted SAR image based on the optimized target binary image and cropping the bounded target image into a sample image; A target prediction module for inputting the sample image into a convolutional neural network model for neural network training; and using the trained network model for prediction and marking the prediction result in the SAR image.

9. A terminal, wherein, The terminal includes: a memory, a processor, and a vehicle-mounted SAR target detection and recognition program stored on the memory and executable on the processor. When the vehicle-mounted SAR target detection and recognition program is executed by the processor, the steps of the vehicle-mounted SAR target detection and recognition method according to any one of claims 1-7 are implemented.

10. A computer-readable storage medium, wherein, The computer-readable storage medium stores a vehicle-mounted SAR target detection and recognition program. When the vehicle-mounted SAR target detection and recognition program is executed by a processor, the steps of the vehicle-mounted SAR target detection and recognition method according to any one of claims 1-7 are implemented.

Images