Near-field MIMO-SAR dangerous target detection method based on improved YOLOv8-seg

Abstract

The invention discloses a near-field MIMO-SAR (Multiple Input Multiple Output-Synthetic Aperture Radar) dangerous target detection method based on improved YOLOv8-seg. The near-field MIMO-SAR dangerous target detection method specifically comprises the following steps: step 1, acquiring a near-field MIMO-SAR dangerous target image; 2, marking a dangerous target in the MIMO-SAR image, constructing a data set, and dividing the data set into a training set, a verification set and a test set according to a certain proportion; 3, expanding the data set by adopting a data enhancement method; step 4, an improved YOLOv8-seg network is constructed; step 5, inputting the training set into an improved YOLOv8-seg network for training, obtaining optimal network parameters, and obtaining a near-field synthetic aperture radar dangerous target detection network; and step 6, inputting a near-field synthetic aperture radar image needing to be detected into the near-field synthetic aperture radar dangerous target detection network, and identifying and positioning a dangerous target in the image.

Description

Technical Field

[0001] The present invention relates to the field of computer vision, and in particular to a near-field MIMO-SAR dangerous target detection method based on improved YOLOv8-seg. Background Art

[0002] At present, target detection technology based on radar imaging is widely used in the fields of ship identification, geological monitoring, and offshore oil spill detection. With the rapid development of the transportation and logistics industries, the rapid and accurate detection of dangerous objects has become increasingly important. Traditional detection methods such as X-rays have good detection effects, but they are large in size, complex in structure, and have high-energy radiation. Millimeter waves have good penetration, are safe, and have no electromagnetic radiation, and are expected to become a new means of detecting dangerous goods. Therefore, the application of millimeter-wave radar near-field imaging technology in the field of security inspection has a high research value. The detection accuracy of millimeter-wave radar is directly affected by the radar working bandwidth and the effective aperture of the antenna. Due to the limitations of antenna size and weight, it is impossible to reserve a large enough position to place a large-aperture radar antenna. Synthetic aperture radar (SAR) uses a mobile antenna to transmit and receive the echo signals of each array unit in steps and perform array synthesis processing to achieve the same detection effect as a large-aperture radar, breaking through the limitation of the antenna aperture on the radar azimuth resolution. However, SAR acquires synthetic aperture data through mobile scanning, which takes a long time. Multiple-input multiple-output (MIMO) radar is a new radar system that uses multiple transmitting antennas to synchronously transmit diversity waves and multiple receiving antennas to receive echo signals. It can collect data in a short time, solving the problem of slow data collection speed of traditional SAR imaging. Therefore, millimeter-wave radar near-field imaging based on MIMO-SAR can reduce hardware complexity and reduce data collection time.

[0003] However, the existing target detection algorithms are mostly used to detect remote sensing SAR targets, and there is a lack of detection algorithms for near-field SAR targets. Remote sensing SAR is usually used for long-distance and large-range target detection, with a large imaging range and a relatively small target object relative to the image. Near-field SAR is mainly used for short-distance and small-range target detection, with a small imaging range and a relatively large target object relative to the image. Due to the existence of multipath effects, the echo signal is more complex. Therefore, traditional remote sensing SAR target detection algorithms are prone to failure or reduced detection accuracy in near-field SAR target detection. Therefore, a millimeter-wave image target detection algorithm for MIMO-SAR based on YOLOv8-seg is proposed for near-field dangerous target detection. Summary of the invention

[0004] In view of the above problems, the present invention provides a near-field MIMO-SAR dangerous target detection method based on improved YOLOv8-seg, which can quickly and accurately detect dangerous targets in complex backgrounds.

[0005] In order to achieve the purpose of the present invention, the following technical scheme is proposed: a near-field MIMO-SAR dangerous target detection method based on improved YOLOv8-seg, comprising the following steps:

[0006] Step 1: Acquire near-field multiple-input multiple-output-synthetic aperture radar (MIMO-SAR) dangerous target images;

[0007] Step 2: Label the dangerous targets in the MIMO-SAR images, construct a data set, and divide the data set into a training set, a validation set, and a test set according to a certain ratio;

[0008] Step 3: Use data augmentation methods to expand the dataset;

[0009] Step 4: Build an improved YOLOv8-seg network;

[0010] Step 5: Input the training set into the improved YOLOv8-seg network for training to obtain the optimal network parameters and obtain the near-field synthetic aperture radar dangerous target detection network;

[0011] Step 6: Input the near-field synthetic aperture radar image to be detected into the near-field synthetic aperture radar dangerous target detection network model to identify and locate the dangerous targets in the image.

[0012] In step 1 described in this scheme, dangerous target data is collected by millimeter wave radar, MIMO-SAR imaging is completed, and a synthetic aperture radar image containing dangerous targets is obtained.

[0013] Furthermore, in step 2, a polygon annotation tool is used to accurately depict the contour of the target, and an annotation file with an instance segmentation mask is generated, and a training set, a validation set, and a test set are randomly allocated in a ratio of 7:2:1.

[0014] Furthermore, in step 3, data enhancement is performed on the training set and the validation set respectively, and the data enhancement methods include but are not limited to mirroring, brightness transformation, Gaussian blur, and saturation change operations to expand the synthetic aperture radar image containing dangerous targets.

[0015] Furthermore, in the step 4, the improved YOLOv8-seg network consists of a backbone network, a neck network and a head network, the CBS (Convolutional-Batchnormal-SiLu) modules of the 1st, 3rd, 5th, 7th, 16th and 19th layers of the conventional YOLOv8-seg backbone network and the neck network are replaced with GhostConv modules, the C2f modules of the 12th, 15th, 18th and 21st layers in the conventional YOLOv8-seg neck network are replaced with C3-RVB modules, and the loss function CIoU of YOLOv8-seg is replaced with the loss function Inner-EioU.

[0016] Furthermore, in step 5, the training set is input into the improved YOLOv8-seg network for training, the loss value is calculated each iteration, the network weights are updated by back propagation, and the weight file with the lowest loss value is saved as the optimal network parameter.

[0017] Furthermore, in step 6, the near-field synthetic aperture radar image to be detected is input into a near-field synthetic aperture radar dangerous target detection network to identify and locate dangerous targets in the image.

[0018] The beneficial effects of the present invention are as follows: the near-field MIMO-SAR dangerous target detection method based on the improved YOLOv8-seg is used to solve the problems of large volume, complex structure, high-energy radiation, etc., which exist in X-ray imaging used in detection fields such as the transportation and logistics industries. Compared with the existing target detection method, the improved YOLOv8-seg network improves the detection accuracy, effectively solves the problems of missed detection and false detection that are prone to occur in conventional algorithms, reduces network parameters to a certain extent, and realizes lightweight model. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solution of the embodiment of the present invention, the following is a brief introduction to the drawings required for describing the embodiment. Among them:

[0020] Figure 1 It is the overall flow chart of the present invention;

[0021] Figure 2 This is the improved YOLOv8-seg network structure diagram of the present invention;

[0022] Figure 3 The improved C3-RVB structure diagram and RepViT Block structure diagram of the present invention are shown in FIG.

[0023] Figure 4 This is the effect diagram of dangerous target detection. DETAILED DESCRIPTION

[0024] The purpose of the present invention is further described in detail below through specific examples. The examples cannot be repeated here one by one, but the implementation mode of the present invention is not limited to the following examples.

[0025] The overall process of the present invention is as follows Figure 1 As shown, a near-field MIMO-SAR dangerous target detection method based on improved YOLOv8-seg specifically includes the following steps:

[0026] Step 1: Acquire near-field MIMO-SAR dangerous target images;

[0027] Step 2: Label the dangerous targets in the MIMO-SAR images, construct a data set, and divide the data set into a training set, a validation set, and a test set according to a certain ratio;

[0028] Step 3: Use data augmentation methods to expand the dataset;

[0029] Step 4: Build an improved YOLOv8-seg network;

[0030] Step 5: Input the training set into the improved YOLOv8-seg network for training to obtain the optimal network parameters and obtain the near-field synthetic aperture radar dangerous target detection network;

[0031] Step 6: Input the near-field synthetic aperture radar image to be detected into the near-field synthetic aperture radar dangerous target detection network to identify and locate the dangerous targets in the image.

[0032] Through the above six steps of this embodiment, dangerous target detection in synthetic aperture radar images under complex backgrounds can be finally achieved.

Claims

1. A near-field MIMO-SAR dangerous target detection method based on improved YOLOv8-seg, characterized by: The following steps are included: Step 1: Acquire near-field multiple-input multiple-output-synthetic aperture radar (MIMO-SAR) dangerous target images; Step 2: Label the dangerous targets in the MIMO-SAR images, construct a data set, and divide the data set into a training set, a validation set, and a test set according to a certain ratio; Step 3: Use data augmentation methods to expand the dataset; Step 4: Build an improved YOLOv8-seg network; Step 5: Input the training set into the improved YOLOv8-seg network for training to obtain the optimal network parameters and obtain the near-field synthetic aperture radar dangerous target detection network; Step 6: Input the near-field synthetic aperture radar image to be detected into the near-field synthetic aperture radar dangerous target detection network to identify and locate the dangerous targets in the image.

2. A near-field MIMO-SAR dangerous target detection method based on improved YOLOv8-seg according to claim 1, characterized in that: In the step 1, the dangerous target data is collected by millimeter wave radar, MIMO-SAR imaging is completed, and a synthetic aperture radar image containing the dangerous target is obtained.

3. A near-field MIMO-SAR dangerous target detection method based on improved YOLOv8-seg according to claim 1, characterized in that: In the step 2, a polygon annotation tool is used to accurately depict the contour of the target, and an annotation file with an instance segmentation mask is generated. The training set, validation set, and test set are randomly allocated in a ratio of 7:2:

1.

4. A near-field MIMO-SAR dangerous target detection method based on improved YOLOv8-seg according to claim 1, characterized in that: In the step 3, data enhancement is performed on the training set and the validation set respectively, and the data enhancement methods include but are not limited to mirroring, brightness transformation, Gaussian blur, and saturation change operations to expand the synthetic aperture radar image containing dangerous targets.

5. A near-field MIMO-SAR dangerous target detection method based on improved YOLOv8-seg according to claim 1, characterized in that: In the step 4, the improved YOLOv8-seg network consists of a backbone network, a neck network and a head network. The CBS (Convolutional-Batchnormal-SiLu) modules of the 1st, 3rd, 5th, 7th, 16th and 19th layers of the conventional YOLOv8-seg backbone network and the neck network are replaced with GhostConv modules, and the C2f modules of the 12th, 15th, 18th and 21st layers in the conventional YOLOv8-seg neck network are replaced with C3-RVB modules. The loss function CIoU of YOLOv8-seg is replaced with the loss function Inner-EioU.

6. The method for collecting dangerous target data by millimeter wave radar according to claim 2, characterized in that: The millimeter-wave radar used to collect data is a multi-input multi-output millimeter-wave radar. The dangerous targets collected include but are not limited to kitchen knives, wrenches, scissors, hammers, and model pistols. The collected radar data is ADC data, and the synthetic aperture radar image is obtained after MIMO-SAR imaging processing of the ADC data.

7. The MIMO-SAR imaging according to claim 2, characterized in that: During the imaging process, the MIMO radar uses multiple transmitting antennas to synchronously transmit diversity waves, and multiple antennas to receive echo signals. It uses synthetic aperture radar technology to perform high-resolution imaging of near-field dangerous targets. The imaging algorithm uses an image reconstruction algorithm based on the back projection algorithm (BP). By compensating for the phase of the intermediate frequency signal in the propagation path, it completes the target plane projection and generates a clear synthetic aperture radar image.

8. The method for expanding the synthetic aperture radar image containing dangerous targets according to claim 4, characterized in that: When performing data augmentation on the training set and validation set, it is necessary to ensure the integrity and consistency of the dangerous targets, and to simultaneously perform augmentation processing on the instance segmentation mask to generate an annotation file that matches the enhanced synthetic aperture radar image.

9. The GhostConv module according to claim 5, characterized in that: Ordinary convolution is used to extract features from the input synthetic aperture radar image, and then linear operations are used to extract feature map features. Finally, the final feature map is generated through a fusion module.

10. The C3-RVB module according to claim 5, characterized in that: RepViT Block is used to replace Bottleneck in CSPDarknet53 to form the C3-RVB module.

11. The loss function Inner-EioU according to claim 5, characterized in that: The loss function Inner-EIoU calculates the difference between the length and width instead of the aspect ratio, and introduces a scaling factor to control the scale of the auxiliary bounding box, using auxiliary bounding boxes of different scales for different data sets.

12. A near-field MIMO-SAR dangerous target detection method based on improved YOLOv8-seg according to claim 1: In described step 5, the training set is input into the improved YOLOv8-seg network for training, and each iteration completes the calculation of the loss value, and the back propagation updates the weight, and the weight file with the lowest loss value is saved as the optimal network parameter.

Images