CN-121982592-A - Visual scale self-adaptive unmanned aerial vehicle detection method based on HRRP physical perception

CN121982592ACN 121982592 ACN121982592 ACN 121982592ACN-121982592-A

Abstract

The invention relates to a visual scale self-adaptive unmanned aerial vehicle detection method based on HRRP physical perception, and belongs to the technical field of mobile communication. The method comprises the steps of constructing a collaborative monitoring architecture of an ISAC base station and a vision sensor, building a space-time mapping relation between a radio frequency sensing coordinate system and a vision sensor coordinate system, executing macroscopic parameter estimation based on echo signals received by the ISAC base station, resolving space position coordinates of a target, generating a servo control instruction to drive a vision sensor holder to rotate, locking the target at the center of a visual field of view, extracting frequency domain channel response of the echo signals, generating a time domain HRRP through IDFT, estimating the radial physical span of the target by combining a constant false alarm detection algorithm, constructing a self-adaptive image slice model based on physical structure constraint, resampling slices by utilizing a scale normalization scaling factor, inputting normalized image blocks into a feature fusion detection network, and finishing fine detection of the unmanned aerial vehicle target.

Inventors

TANG LUN
MENG JIAQI
CHEN QIANBIN

Assignees

重庆邮电大学

Dates

Publication Date: 20260505
Application Date: 20260203

Claims (8)

1. The visual scale self-adaptive unmanned aerial vehicle detection method based on HRRP physical perception is characterized by comprising the following steps of: S1, constructing a collaborative monitoring architecture of an ISAC base station and a vision sensor, configuring an ISAC base station to transmit an integrated OFDM waveform, and building a space-time mapping relation between a radio frequency sensing coordinate system and a vision sensor coordinate system, wherein ISAC represents general sense integration, and OFDM represents orthogonal frequency division multiplexing; S2, performing macroscopic parameter estimation based on echo signals received by an ISAC base station, resolving space position coordinates of a target, generating a servo control instruction to drive a visual sensor holder to rotate, and locking the target in the center of a visual image field of view; s3, extracting the frequency domain channel response of the echo signal, generating a time domain HRRP through inverse discrete Fourier transform, and estimating the radial physical span of the target by combining a constant false alarm detection algorithm, wherein the HRRP represents a high-resolution range profile; S4, constructing a self-adaptive image slice model based on physical structure constraint, fusing real-time distance information perceived by an ISAC base station and radial physical span estimated by HRRP, dynamically calculating optimal slice resolution and window size of a visual image in horizontal and vertical dimensions, and resampling a slice by utilizing a scale normalization scaling factor to generate a scale normalized target image block; And S5, inputting the normalized image blocks into a feature fusion detection network to finish the fine detection of the unmanned aerial vehicle target.
2. The visual scale adaptive unmanned aerial vehicle detection method of claim 1, wherein in step S1, the constructed collaborative monitoring architecture comprises a physical awareness layer and a data processing layer; The physical perception layer comprises an ISAC base station, a visual sensor, a cradle head and a servo controller, wherein the ISAC base station is connected with the visual sensor, and the signals of the ISAC base station and the visual sensor are synchronous in time; The data processing layer comprises a macroscopic parameter resolving module, a microscopic feature analyzing module, a servo guiding control module, a self-adaptive slice computing module and a target recognition reasoning module, wherein the data processing layer acquires target distance and direction from an ISAC base station through the macroscopic parameter resolving module, acquires target physical scale characteristics from the ISAC base station through the microscopic feature analyzing module, acquires images from a vision sensor through the self-adaptive slice computing module, combines the acquired target distance and direction and the target physical scale characteristics to generate a normalized image block, and finally outputs a detection result through the target recognition reasoning module.
3. The method for detecting the visual scale adaptive unmanned aerial vehicle according to claim 1, wherein in the step S1, the space-time mapping relation between the radio frequency sensing coordinate system and the visual coordinate system is established by defining a radar spherical coordinate system ) And visual Cartesian coordinate system ) Acquiring a rotation alignment matrix through a calibration experiment Translation vector with equipment installation deviation Establishing a coordinate mapping relation: Wherein, the For the distance of the target to the ISAC base station, For the azimuth angle, Is the pitch angle.
4. A visual scale adaptive unmanned aerial vehicle detection method according to claim 3, wherein step S2 comprises the steps of: s21, carrying out beam forming two-dimensional fast Fourier transform on multi-channel echo signals received by an ISAC base station, and extracting a distance-angle spectrum of a target; s22, estimating a signal parameter algorithm by adopting a rotation invariant technology, and extracting the distance of the target from the distance-angle spectrum Azimuth angle And pitch angle ; S23, utilizing a coordinate transformation matrix The position of the target under the radar spherical coordinate system ) Mapping to a visual Cartesian coordinate System ) Calculating the yaw angle of the cradle head And pitch angle Generates a servo control command: and generating a servo control instruction to drive the cradle head to rotate, so that the target is positioned at the center of the visual field of the visual image in real time.
5. The visual scale adaptive unmanned aerial vehicle detection method according to claim 1, wherein step S3 specifically comprises the steps of: S31 extracting the first received by the ISAC base station Perceptual subcarrier data in a group of OFDM symbols, constructing a frequency domain echo vector , Is the total number of effective sub-carriers; S32 pair Performing inverse discrete fourier transform to convert the frequency domain channel response into a time domain HRRP, wherein the calculation formula is expressed as follows: Wherein, the Is the first The complex response strengths of the individual distance elements, Is the first Frequency domain response values of the individual subcarriers; S33, calculating an amplitude spectrum of the HRRP And identifying a non-zero support area of the target echo by using a unit average constant false alarm detection algorithm; s34, counting the number of distance units continuously exceeding the detection threshold in the non-zero supporting area And calculate the radial physical span of the target according to the ISAC system parameters : Wherein, the In order to achieve the light velocity, the light beam is, For the effective bandwidth of the OFDM signal, Is the subcarrier spacing.
6. The visual scale adaptive unmanned aerial vehicle detection method according to claim 5, wherein in step S4, an adaptive image slice model is constructed, specifically comprising the steps of: S41, defining reference parameters of the vision detection network, including reference training distance Physical length of reference target Reference slice window size Wherein 、 The width and height of the window are respectively; s42, calculating a physical scale correction factor The method comprises the steps of representing the deviation degree of the current target estimated size and the reference size; Wherein, the In order to correct the function for clipping, The dynamic range of the correction factor is preset; S43, combining real-time distance And correction factor Calculating optimal slice window vector on current visual image : Wherein, the For an optimal slice window width, For an optimal slice window height, Representing a rounding down.
7. The method for detecting the visual scale-adaptive unmanned aerial vehicle according to claim 6, wherein in the step S4, the slice is resampled by using the scale normalization scaling factor, in particular, the slice is resampled by using the scale normalization scaling factor Wherein For the standard input size of the network, a bicubic interpolation algorithm or a pre-trained super-resolution reconstruction network is adopted, and the size is as follows Mapping the original slice images of (a) to a constant size Normalized image tensor of (a) 。
8. The method for detecting the visual scale adaptive unmanned aerial vehicle according to claim 7, wherein in the step S5, the feature fusion detection network specifically comprises the steps of tensor of normalized images Input convolutional neural network containing residual connection, extract multi-scale visual semantic features ; HRRP amplitude spectrum Mapping to a one-dimensional feature vector, and encoding to be embedded into physical features through a full connection layer; And embedding visual semantic features and physical features into a feature fusion layer of the detection head for channel splicing, inputting the confidence coefficient of the target category and the regression value of the bounding box into the detection head, and outputting the classification result and pixel-level positioning coordinates of the unmanned aerial vehicle.

Description

Visual scale self-adaptive unmanned aerial vehicle detection method based on HRRP physical perception Technical Field The invention belongs to the technical field of mobile communication, and relates to a visual scale self-adaptive unmanned aerial vehicle detection method based on HRRP physical perception under a general sense integrated system. Background Along with the rapid development of low-altitude airspace economy, unmanned aerial vehicles are increasingly widely applied in the fields of logistics, inspection, mapping and the like, but meanwhile, the problems of illegal invasion, privacy disclosure, public safety threat and the like are brought, and a high-precision and wide-area coverage unmanned aerial vehicle detection technology is urgently needed. The integrated sense of general (ISAC) system is used as one of 6G core technologies, can realize the cooperation of communication and sensing functions through the same hardware platform, and provides a new technical path for unmanned aerial vehicle detection. The existing unmanned aerial vehicle detection technology mainly relies on single sensor or simple multi-sensor fusion, and has a plurality of limitations that the single sensor detection has obvious defects, the acoustic sensor is easy to be interfered by environmental noise, and the effective detection distance is generally less than 100m; the method comprises the steps of enabling a radio frequency sensor to rely on priori knowledge of communication bandwidth of an unmanned aerial vehicle, enabling the unmanned aerial vehicle to be unable to deal with radio silence mode, enabling a radar sensor to achieve long-distance detection, enabling the radar sensor to be similar to radar scattering cross sections of birds and easily generate false alarms, enabling a vision sensor to achieve pixel-level positioning through deep learning, enabling imaging dimensions to change rapidly along with the distance, seriously affecting scale invariance of a convolutional neural network, enabling the detection distance to be generally limited to be 300m, enabling a majority of through sense fusion methods to focus on sensor cooperative positioning, enabling physical dimension information perceived by the radar to be not fully utilized, enabling part of vision enhancement methods to achieve fixed factor super-resolution reconstruction, enabling physical dimension of an unmanned aerial vehicle to be not adaptively matched with the correlation of imaging distances, enabling scale normalization effect to be poor, enabling target physical structure information contained in HRRP to be lacked in a dynamic mapping mechanism of visual imaging resolution, enabling the imaging dimensions of the unmanned aerial vehicle to be difficult to achieve ' real scale invariance ' detection ', enabling the imaging dimensions of the unmanned aerial vehicle to be severely fluctuant under wide-area conditions, enabling the existing visual detection network to be fixed in the input dimensions, enabling the target physical dimensions to be not to be combined with real physical dimensions to be self-adaptively adjusted, and enabling the target physical dimensions to be further large in distance redundancy to be large in size. The High Resolution Range Profile (HRRP) is used as a core feature of radar perception, can reflect scattering intensity distribution of a target along the radar sight line direction, and contains real physical radial dimension information of the target. However, HRRP is used for radar target recognition in the prior art, is not fused with the scale self-adaptive demand depth of visual detection, and the physical scale which cannot be solved by HRRP provides prior constraint for visual slicing and scale normalization, so that the cooperative gain of sense fusion is not fully exerted. Therefore, improving the detection performance of the unmanned aerial vehicle of the general sense integrated system is a problem to be solved urgently. Disclosure of Invention In view of the above, the invention aims to provide a visual scale self-adaptive unmanned aerial vehicle detection method based on HRRP physical perception under a sense-of-general integrated system, which constructs a sense-of-general collaborative monitoring framework, extracts target physical scale information by using HRRP, establishes self-adaptive mapping of physical dimension and visual imaging, realizes true scale invariance of visual input, improves the unmanned aerial vehicle detection performance of the sense-of-general integrated system, and can realize high-precision and high-robustness detection of unmanned aerial vehicle targets in a wide area. In order to achieve the above purpose, the present invention provides the following technical solutions: the visual scale self-adaptive unmanned aerial vehicle detection method based on HRRP physical perception under the general sense integrated system specifically comprises the following steps: S1, constructin