CN-122009191-A - Road surface type identification method based on PVDF piezoelectric film and wheel cover near field radar multisource fusion

CN122009191ACN 122009191 ACN122009191 ACN 122009191ACN-122009191-A

Abstract

The invention discloses a pavement type identification method based on PVDF piezoelectric film and wheel cover near field radar multisource fusion, which is characterized in that PVDF piezoelectric film arrays are embedded in tires to collect dynamic strain signals of contact spots, near field radars are arranged in the wheel covers to detect front pavement echoes, PVDF time signals are resampled to be uniform space scales based on vehicle speed, radar ground areas are dynamically determined by combining vehicle body gestures to achieve multi-mode data space alignment, PVDF time-frequency characteristics, radar energy distribution, attenuation slope and specular reflection characteristics are respectively extracted, a transducer model is input to conduct trans-modal fusion, and pavement types and confidence are output. The recognition robustness and the instantaneity under the complex road surfaces such as ponding, ice and snow are obviously improved through a joint mechanism of radar pre-aiming and PVDF touch verification, and the method is suitable for high-level automatic driving vehicle stability control.

Inventors

LI BO
YANG LIN
TI YAN
HUA LEI
ZHU YUNHAI
Bei Shaodie
NI ZHANG
WANG JIE
GONG LINHAO
YU JIANGUO
XU QINYI
ZHAO GUANGJIN

Assignees

江苏理工学院

Dates

Publication Date: 20260512
Application Date: 20260323

Claims (7)

1. The pavement type identification method based on PVDF piezoelectric film and wheel cover near field radar multisource fusion is characterized by comprising the following steps: s1, deploying a PVDF piezoelectric film and a near-field radar, and respectively acquiring a dynamic strain signal and an echo signal; s2, preprocessing and extracting features of the signals acquired in the step S1 respectively; S3, based on a unified time reference, combining the speed and the wheel speed, mapping and processing the features extracted in the step S2, so as to construct a combined feature embedding sequence; S4, inputting the output sequence in the step S3 into a transducer network, fusing the sequence with multiple modes through a multi-head self-attention mechanism, outputting a road surface type classification result and comparing the result; s5, inputting the output result of the step S4 to a vehicle control system for parameter self-adaptive adjustment.
2. The method for identifying the road surface type according to claim 1, wherein the PVDF piezoelectric film in the step S1 is arranged on the inner surfaces of the tire crown and the tire shoulder and a plurality of pieces of PVDF piezoelectric film are arranged to form a touch sensing array, and the near-field radar is arranged inside the wheel cover in front of the tire, so that the radar can detect the road surface area between 0.5m and 5m in front of the tire.
3. The method of claim 1, wherein the step S2 of signal preprocessing converts the PVDF dynamic strain signals to time series, removes low frequency drift and high frequency interference by band-pass filtering, and processes the radar echo signals to distance spectrum or distance-Doppler spectrum sequence by distance FFT and optional Doppler FFT.
4. The method for identifying the road surface type according to claim 1, wherein the feature extraction in the step S2 comprises PVDF feature extraction and radar feature extraction, wherein the PVDF feature extraction comprises feature extraction of root mean square, mean value, variance, skewness, kurtosis, pulse factor, margin factor, spectrum centroid, spectrum bandwidth, spectrum flatness, band energy, strain amplitude difference and phase difference, and the radar feature extraction comprises feature extraction of average energy, maximum energy, energy centroid, half-peak width and attenuation slope.
5. The method for identifying road surface type according to claim 1, wherein the step S3 comprises establishing spatial mapping relationship between PVDF features and radar features in the same time step by unifying time base, and obtaining joint feature vector : , Wherein the method comprises the steps of Representing the characteristic vector of the PVDF, Representing radar eigenvectors and linking the eigenvectors Mapping to an embedded vector By adding position coding Combined with the embedded vector by the following formula: , A joint feature embedding sequence { x (1), x (2),. The term, x (T) } of length T is composed.
6. The method for identifying road surface type according to claim 1, wherein the converter network in the step S4 comprises an input end and an output end, the input end is a 3-layer converter, each layer of the converter comprises a multi-head self-attention sub-layer and a feedforward full-connection sub-layer, the sub-layers adopt a residual connection and layer normalization structure, the embedded sequence output in the step S3 is used as input, and the embedded sequence is mapped into a query matrix Q, a key matrix K and a value matrix V through linear transformation, and the expression is: ; Wherein, the 、、 Is a learnable weight matrix, and multiple self-attentions have independent weight matrix 、、 The correlation between time steps is calculated using a scaled dot product self-attention mechanism: , Wherein the method comprises the steps of The high-dimensional semantic feature representation sequence is obtained through three-layer encoder stacking and multi-head self-attention sub-layer and feedforward full-connection sub-layer processing: , where T is the number of sliding time windows, And the high-dimensional semantic feature representation of the T-th pavement segment after the fusion of the full-sequence information is realized.
7. The method for identifying road surface type according to claim 6, wherein the output end of the transducer network comprises a classification head and a regression head, the classification head adopts a layer of fully-connected network, and the high-dimensional semantic feature representation sequence is globally pooled into a road surface representation vector h, which is expressed as the following modes: , The road surface expression vector is used as the input of a classification head, the classification head outputs the road surface type class probability, and the regression head outputs the predicted attachment coefficient.

Description

Road surface type identification method based on PVDF piezoelectric film and wheel cover near field radar multisource fusion Technical Field The invention relates to the field of intelligent vehicles, in particular to a pavement type identification method based on multi-source fusion of PVDF piezoelectric film and wheel cover near field radar. Background With the development of intelligent driving and automatic driving technologies, the real-time perception of the road surface state by vehicles becomes increasingly important. In particular to the accurate identification of different road types such as dryness, wet skid, ponding, snow accumulation, icing, broken stone and the like, and is an important foundation for realizing active safety control, track planning and whole vehicle energy management. The existing pavement identification method mainly comprises two main types (1) an indirect method based on vehicle body dynamics inversion. The method utilizes parameters such as wheel speed difference, lateral acceleration, yaw rate, braking pressure and the like, estimates the adhesion coefficient of the tire and the road surface through a vehicle dynamics model or an observer, often depends on obvious slip or slip working conditions, has obvious response hysteresis, and is highly sensitive to the parameters of the tire model, load distribution and tire pressure, and (2) is based on a camera, radar and other environmental sensing methods. The method is characterized in that the method only depends on a visible light camera or a far-field radar to analyze the texture, brightness or scattering characteristics of a road in front, is highly sensitive to lighting conditions, rain and fog weather and night scenes, and is difficult to distinguish road types with similar visual textures and different physical characteristics. Haptic sensing schemes based on intelligent tires have been used to measure tire vibration and strain signals using in-tire accelerometers or PVDF piezoelectric films, or to infer road roughness using vehicle body acceleration signals. The method reduces the dependence on the tire model to a certain extent, but only uses single touch mode information, and does not effectively fuse with external sensors such as a forward radar, and the identification accuracy and the robustness are still limited under rainy and snowy weather, a multi-medium mixed road surface and complex attachment conditions. Therefore, how to achieve the consistent fusion of the tactile information and the spatial path of the macroscopic information of the front near-field radar near the tire-road surface contact position, and construct a road surface type identification method with real-time performance and high robustness are the problems to be solved by the technicians in the field. Disclosure of Invention The invention aims to overcome the defects of the prior art and provides a pavement type identification method based on multi-source fusion of a PVDF piezoelectric film and a wheel cover near-field radar. The aim of the invention can be achieved by the following technical scheme: the pavement type identification method based on PVDF piezoelectric film and wheel cover near field radar multisource fusion is characterized by comprising the following steps: s1, deploying a PVDF piezoelectric film and a near-field radar, and respectively acquiring a dynamic strain signal and an echo signal; s2, preprocessing and extracting features of the signals acquired in the step S1 respectively; S3, based on a unified time reference, combining the speed and the wheel speed, mapping and processing the features extracted in the step S2, so as to construct a combined feature embedding sequence; S4, inputting the output sequence in the step S3 into a transducer network, fusing the sequence with multiple modes through a multi-head self-attention mechanism, outputting a road surface type classification result and comparing the result; s5, inputting the output result of the step S4 to a vehicle control system for parameter self-adaptive adjustment. Further, the PVDF piezoelectric film in the step S1 is arranged on the inner surfaces of the tire crown and the tire shoulder, a plurality of pieces of PVDF piezoelectric film are arranged to form a touch sensing array, and the near-field radar is arranged inside the tire front wheel cover, so that the radar can detect the road surface area between 0.5m and 5m in front of the tire. Further, in the signal preprocessing in step S2, the dynamic strain signals of the PVDF are converted to form time sequences, and the low-frequency drift and the high-frequency interference are removed through band-pass filtering, and meanwhile, the echo signals of the radar are processed through a distance FFT and an optional doppler FFT to form a distance spectrum or a distance-doppler spectrum sequence. Further, the feature extraction in the step S2 includes a PVDF feature extraction and a radar feature extraction, where t