Highway construction quality dynamic detection method based on high-precision laser ranging

Abstract

The invention relates to the technical field of laser ranging, and further relates to a highway construction quality dynamic detection method based on high-precision laser ranging, and the method comprises the steps: 1, arranging a plurality of nodes of high-precision laser ranging terminals at the two sides of the center line of a highway to be built at equal intervals, and building a unified reference three-dimensional coordinate system; 2, the receiving end of each node receives an original echo signal returned through pavement scattering; performing multi-scale wavelet transform on the original echo signal to form a construction quality multi-dimensional parameter matrix so as to represent the current construction quality state in a real-time high-dimensional manner; and 3, performing principal component analysis on the construction quality multi-dimensional parameter matrix to obtain a highway construction quality detection result. The method has the advantages of high ranging precision, high temporal-spatial resolution, high automation degree, high real-time performance, outstanding anomaly recognition capability and the like.

Description

Technical Field

[0001] The invention belongs to the technical field of laser ranging, and in particular relates to a dynamic detection method for highway construction quality based on high-precision laser ranging. Background Art

[0002] In modern infrastructure construction, highway construction quality inspection and monitoring have become a core component in ensuring structural safety and longevity. Traditional highway construction quality inspection methods primarily include manual cross-section measurement, leveling, total station deployment, and static load settlement testing. While these methods met the basic needs of construction quality assessment for a period of time, the increasing construction of modern expressways, smart transportation, heavy-duty roads, and special structural sections has gradually exposed technical bottlenecks and accuracy gaps in dynamic response capture, high-frequency disturbance identification, and millimeter-level displacement measurement.

[0003] Traditional cross-section measurement typically relies on manual wires, flat rulers, and target devices, combined with a theodolite to read height differences at discrete measurement points. This method's accuracy is limited by instrument resolution and human error, making it incapable of achieving millimeter-level, high-precision dynamic tracking of continuous spatial distributions. Furthermore, this method lacks real-time performance and can only be performed once during construction breaks, making it difficult to provide immediate feedback on the quality of critical construction processes (such as rolling, compaction, and backfill).

[0004] Total station deployment introduces automatic angle and distance measurement capabilities, enabling simultaneous observation of multiple points. However, the measurement frequency is typically several times per second and requires a complex deployment of artificial reflective markers. For non-rigid continuous pavement materials (such as water-stabilized layers and asphalt layers), the total station's reliance on echo stability significantly reduces its reliability in strong sunlight, high dust levels, or slippery environments. Fluctuations in the reflective surface can cause distance jumps, seriously affecting displacement continuity analysis. Summary of the Invention

[0005] The main purpose of this invention is to provide a dynamic detection method for highway construction quality based on high-precision laser ranging. This method deploys laser ranging terminal nodes on both sides of the highway centerline, uses a multi-frequency modulation coding sequence to transmit ranging beams, and combines wavelet transforms, tensor decomposition, manifold clustering, and deep temporal networks to achieve millimeter-level displacement and settlement monitoring and prediction. Finally, a multidimensional construction quality parameter matrix is ​​constructed and a comprehensive quality index is extracted to assess construction status in real time. This method offers the advantages of high ranging accuracy, strong spatiotemporal resolution, a high degree of automation, strong real-time performance, and outstanding anomaly recognition capabilities, making it widely applicable to smart highway construction quality control scenarios.

[0006] In order to solve the above problems, the technical solution of the present invention is achieved as follows:

[0007] A dynamic detection method for highway construction quality based on high-precision laser ranging, the method comprising:

[0008] Step 1: Several high-precision laser ranging terminal nodes are evenly spaced on both sides of the centerline of the proposed highway. The spacing between the nodes is a pre-set fixed design spacing. The transmitter of each node transmits a ranging beam processed by a multi-frequency modulation coding sequence at a frequency of one thousand times per second, and simultaneously records the absolute three-dimensional coordinates of the node to establish a unified reference three-dimensional coordinate system.

[0009] Step 2: The receiving end of each node receives the original echo signal scattered back by the road surface; a multi-scale wavelet transform is performed on the original echo signal to generate a multi-scale power spectrum tensor; a tensor sparse kernel decomposition method is used to separate the multi-scale power spectrum tensor into a signal tensor and a noise tensor, and a three-dimensional similarity spectrum is constructed; variable-weight Lorentz manifold metric clustering is applied to the three-dimensional similarity spectrum to obtain feature clusters with maximum spectral coherence; the feature clusters are input into a deep temporal gated recurrent network to obtain the predicted joint displacement and settlement deviation for the next time period; the current joint displacement and settlement deviation is aggregated with the predicted joint displacement and settlement deviation according to the time series to form a multi-dimensional construction quality parameter matrix, which represents the current construction quality status in a real-time, high-dimensional manner;

[0010] Step 3: Perform principal component analysis on the multidimensional parameter matrix of construction quality to generate highway construction quality inspection results.

[0011] Furthermore, in step 1, the process of processing the beam multi-frequency modulation coding sequence includes: within a unified frame time slot, dividing the target transmission frequency band into several sub-frequency bands according to a preset multi-layer frequency domain stacking rule; inserting a dedicated start and end synchronization mark into each sub-frequency band to keep the sub-frequency bands aligned on the time-frequency plane to form a basic beam multi-frequency modulation coding sequence frame; implementing inter-layer orthogonal code diffusion on the basic beam multi-frequency modulation coding sequence frame, and mapping each sub-frequency band to a mutually orthogonal pseudo-random diffusion code; realizing self-synchronization interference suppression between different sub-frequency bands through inter-code cyclic shift to ensure that there is no crosstalk during multi-frequency parallel transmission; performing time-frequency interleaving and rearranging on the beam multi-frequency modulation coding sequence frame after orthogonal code diffusion with a preset interleaving matrix; in the process of time-frequency interleaving and rearranging, adjacent time slots and the same Adjacent sub-bands are alternately swapped to obtain an interleaved sequence, which is then subjected to bipolar phase randomization: a pseudo-random phase flip mode is introduced to embed a phase randomization mask into the frame header of the beam multi-frequency modulation coding sequence; a multi-level self-correction error correction embedded segment, including a frame-level block check code and an inter-frame-level cyclic check code, is inserted into the beam multi-frequency modulation coding sequence frame embedded with the phase randomization mask; the embedded segment uses a combination of flipped parity check and redundancy segmentation; the embedded beam multi-frequency modulation coding sequence frame is subjected to spectral robust suppression shaping: a hybrid shaping strategy of symmetric Hanning window and raised cosine window is used to perform bidirectional decremental shaping on the sideband energy of each sub-band to reduce emission spectrum leakage; and an adaptive suppression condensation region is inserted at the center of the spectrum to suppress sidelobe noise caused by multipath echoes.

[0012] Furthermore, the process of inserting an adaptive suppression condensation zone at the center of the spectrum includes: extracting the overall spectral shape envelope of the embedded beam multi-frequency modulation coding sequence frame in real time to determine the center position of the spectrum and the symmetrical energy threshold surface; opening the start and end boundaries of the adaptive suppression condensation zone at the center of the spectrum according to the symmetrical energy threshold surface, and setting progressive connection bands on both sides of the boundary; performing amplitude reduction processing on the inside of the condensation zone according to the adaptive suppression strategy, and performing smooth transition processing on the connection band to make the spectrum shape continuous and consistent; in the subsequent transmission cycle of the beam multi-frequency modulation coding sequence frame, using the echo monitoring results to dynamically correct the condensation zone width and the decreasing depth to keep the sidelobe noise continuously suppressed; whenever a sudden change in the multipath echo characteristics is detected, the adaptive reconstruction of the condensation zone is immediately triggered to ensure that the emission spectrum shape and the suppression effect are updated synchronously.

[0013] Furthermore, the process of implementing inter-layer orthogonal code diffusion on the basic beam multi-frequency modulation coding sequence frame includes: loading a unique and pairwise orthogonal pseudo-random spreading code index field in each sub-band of the beam multi-frequency modulation coding sequence frame; performing inter-code cyclic shift with the frame sequence number as the seed, shifting the pseudo-random spreading codes of adjacent sub-bands clockwise by a fixed code relative to the previous frame; setting an adaptive trigger gate at the transmitting end of the node, and immediately latching the bit sequence when the inter-code cyclic shift is detected to achieve self-synchronization of the sub-band boundary; at the receiving end of the node, reversely recovering the pseudo-random spreading code according to the same cyclic shift sequence to automatically offset cross-sub-band crosstalk; if it is detected that the inter-code alignment deviation exceeds the threshold, immediately triggering the next cyclic shift cycle to dynamically maintain the non-crosstalk of multi-frequency parallel transmission.

[0014] Furthermore, in step 2, the multi-scale power spectrum tensor is separated into a signal tensor and a noise tensor by using a tensor sparse kernel decomposition method, and the process of constructing a three-dimensional similarity spectrum graph includes: performing multi-directional resampling on the multi-scale power spectrum tensor to establish a tensor sparse kernel initial index table; in the tensor sparse kernel initial index table, iteratively removing weakly correlated kernel blocks according to the sparse density threshold, and aggregating the retained kernel blocks to generate a signal tensor; synchronously capturing the removed kernel blocks to form a noise tensor; expanding the signal tensor layer by layer along the frequency domain, scale domain and time domain, calculating the three-way similarity weights, and generating a primary correlation matrix; applying piecewise progressive mapping and cross-stretching processing on the primary correlation matrix to obtain a three-dimensional similarity spectrum graph.

[0015] Furthermore, in step 2, the variable-weight Lorentz manifold metric clustering is applied to the three-dimensional similarity spectrum to obtain a feature cluster with maximum spectral coherence, which includes: constructing a Lorentz manifold initial grid based on spectral density in the entire domain of the three-dimensional similarity spectrum, and assigning a variable weight vector to each grid node; performing a double-threshold adaptive update on the variable weight vector according to the local spectral gradient of the node to form a multi-scale weight field; implementing a layer-by-layer equidistant expansion search on the grid nodes under the variable-weight Lorentz manifold metric, and recording the manifold metric distance between nodes in real time; iteratively merging the node set according to the principle of minimum manifold path and maximum weight gain to generate a candidate feature cluster set; performing spectral coherence increasing screening on the candidate feature cluster set, and outputting the feature cluster with the largest spectral coherence.

[0016] Furthermore, the process of constructing a spectral density-based Lorentz manifold initial grid over the entire domain of the three-dimensional similarity spectrum and assigning a variable weight vector to each grid node includes: decomposing the three-dimensional similarity spectrum into continuous density layers according to the spectral density grading rule; performing bidirectional Manhattan sweep on each density layer to generate equipotential surfaces, and mapping them into surface grid pieces using the Lorentz coordinate embedding method; weaving along the spectral gradient tangent to seamlessly splice all surface grid pieces to form a Lorentz manifold initial grid covering the entire domain; synchronously collecting neighborhood spectral density, local curvature and temporal similarity at each grid node and splicing them into the original eigenvector; converting the original eigenvector into a variable weight vector through a normalized reweighting cycle; writing the variable weight vector into the grid node metadata to complete the grid node weight initialization.

[0017] Furthermore, a bidirectional Manhattan sweep is performed on each density layer to generate an equipotential surface, and the process of mapping it into a surface grid using the Lorentz coordinate embedding method includes: selecting the boundaries of the density layer as the starting reference lines, setting the horizontal and vertical bidirectional Manhattan sweep paths; collecting density threshold surface intersections row by row along the horizontal sweep path, and marking the candidate nodes of the equipotential surface in real time; after completing the horizontal sweep, switching to the vertical sweep path, filling the equipotential surface gap nodes column by column to form a bidirectional closed node network; performing neighborhood connectivity detection on the bidirectional closed node network, eliminating isolated nodes, and obtaining a continuous and smooth equipotential surface; establishing a local Lorentz tangent plane on the equipotential surface, and converting the equipotential surface fragments into the surface parameter domain according to the Lorentz coordinate embedding method; splicing the surface fragments in the order of the surface parameter domain to generate a completely conformal surface grid; writing the density layer identifier and spatial index for each surface grid to complete the global registration of the surface grid.

[0018] Furthermore, step 3 specifically includes: performing dimension normalization on the multidimensional parameter matrix of construction quality, generating a standardized matrix and retaining the inverse normalization index; constructing a sliding covariance grid, updating the covariance weight network according to the time series correlation, and inputting the obtained network into the principal component analysis core engine; the principal component analysis core engine extracts the principal axis vector set according to the progressive variance absorption rate criterion, and performs orthogonal rotation to obtain a stable principal axis; completing a cascade coordinate transformation of the standardized matrix under the stable principal axis to obtain a compressed feature score sequence; performing adaptive contribution aggregation on the compressed feature score sequence according to the time scale, and outputting a comprehensive quality index column; performing a hierarchical comparison on the comprehensive quality index column and the design tolerance library, marking the qualified state and the abnormal state, and forming a highway construction quality inspection result.

[0019] The present invention's dynamic highway construction quality monitoring method based on high-precision laser ranging has the following beneficial effects: Building on existing technologies, this method achieves continuous, real-time, and high-resolution quality status capture and assessment of key structural layers during construction. By deploying high-precision laser ranging terminal nodes at equal intervals on both sides of the highway centerline and performing multi-frequency modulation coded sequence laser scanning of the ground at a transmission frequency of 1000 times per second, the present invention can acquire the absolute three-dimensional coordinate changes of surface scattering points in real time with millimeter-level accuracy, effectively identifying construction conditions such as compaction, filling, settlement, and uplift. Compared to traditional measurement methods that primarily rely on low-frequency static sampling, the present invention achieves continuous data acquisition with high temporal and spatial resolution, overcoming the long manual detection cycle and lack of dynamic change information. By constructing a unified reference three-dimensional coordinate system and performing multi-scale wavelet transform and tensor sparse kernel decomposition on the raw echo signals, the present invention can extract physically representative signal components from large amounts of non-stationary echo data, significantly enhancing anti-interference capabilities. At the same time, through the three-dimensional similarity spectrum and the variable-weight Lorentz manifold metric clustering method, spatially coherent feature clusters in the construction process are effectively identified, providing a highly correlated input feature sequence for subsequent predictions based on deep temporal gated recurrent networks. Further, through the construction of a multidimensional parameter matrix of construction quality and principal component analysis processing, the present invention can achieve a multidimensional comprehensive evaluation and qualification judgment of construction quality, forming a digital quality index. The index has the characteristics of being quantifiable, traceable, and partitionable and labeled, and is suitable for full-process quality supervision, abnormal warning, and construction decision feedback. Therefore, the present invention has significant engineering application value and technology promotion prospects in terms of ensuring construction accuracy, improving engineering quality control efficiency, and supporting the construction of smart construction sites. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 A schematic diagram of a method flow for a dynamic detection method of highway construction quality based on high-precision laser ranging provided by an embodiment of the present invention;

[0021] Figure 2 Schematic diagram of the characteristic distribution and processing effect of the optical beam multi-frequency modulation coding sequence in the frequency domain in an embodiment of the present invention;

[0022] Figure 3 Schematic diagram of the process of multi-scale power spectrum tensor sparse kernel decomposition in an embodiment of the present invention; Figure 4 Schematic diagram of the process of applying variable-weight Lorentz manifold metric clustering based on a three-dimensional similarity spectrum in an embodiment of the present invention. DETAILED DESCRIPTION

[0023] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0024] refer to Figure 1 A method for dynamic detection of highway construction quality based on high-precision laser ranging, the method comprising:

[0025] Step 1: Several high-precision laser ranging terminal nodes are evenly spaced on both sides of the centerline of the proposed highway. The spacing between the nodes is a pre-set fixed design spacing. The transmitter of each node transmits a ranging beam processed by a multi-frequency modulation coding sequence at a frequency of one thousand times per second, and simultaneously records the absolute three-dimensional coordinates of the node to establish a unified reference three-dimensional coordinate system.

[0026] Step 2: The receiving end of each node receives the original echo signal scattered back by the road surface; a multi-scale wavelet transform is performed on the original echo signal to generate a multi-scale power spectrum tensor; a tensor sparse kernel decomposition method is used to separate the multi-scale power spectrum tensor into a signal tensor and a noise tensor, and a three-dimensional similarity spectrum is constructed; variable-weight Lorentz manifold metric clustering is applied to the three-dimensional similarity spectrum to obtain feature clusters with maximum spectral coherence; the feature clusters are input into a deep temporal gated recurrent network to obtain the predicted joint displacement and settlement deviation for the next time period; the current joint displacement and settlement deviation is aggregated with the predicted joint displacement and settlement deviation according to the time series to form a multi-dimensional construction quality parameter matrix, which represents the current construction quality status in a real-time, high-dimensional manner;

[0027] Step 3: Perform principal component analysis on the multidimensional parameter matrix of construction quality to generate highway construction quality inspection results.

[0028] Laser ranging relies on a dual mechanism of phase accumulation and time-of-flight. When the transmitting node outputs a ranging beam at a frequency of 1000 times per second, phase accumulation can lock onto subwavelength variations, while time-of-flight provides an absolute distance reference. The two complement each other to resolve integer wavelength ambiguity, maintaining ranging consistency at the centimeter or even millimeter scale. To avoid ranging jitter caused by phase jumps, the beam is processed into a multi-frequency modulation code sequence. This sequence simultaneously occupies multiple channels in the time, frequency, and code domains. Orthogonal allocation prevents interference between channels within the same node, while cyclic shifting prevents interference from neighboring nodes. Synchronization identifiers and error correction codes are embedded at the transmitting end, ensuring that echo identification and error correction can be performed simultaneously at the receiving end. The design of robust spectral shape suppression and the insertion of an adaptive suppression cohesion zone at the center of the spectrum are based on understanding the mechanism of multipath echoes: multipath causes sidelobe noise to concentrate in the frequency band adjacent to the mainlobe. The central cohesion zone dynamically weakens the energy at the edge of the mainlobe, thereby reducing the relative height of the sidelobes and improving the feasibility of subsequent signal separation.

[0029] Laser ranging relies on a dual mechanism of phase accumulation and time-of-flight. When the transmitting node outputs a ranging beam at a frequency of 1000 times per second, phase accumulation can lock onto subwavelength variations, while time-of-flight provides an absolute distance reference. The two complement each other to resolve integer wavelength ambiguity, maintaining ranging consistency at the centimeter or even millimeter scale. To avoid ranging jitter caused by phase jumps, the beam is processed into a multi-frequency modulation code sequence. This sequence simultaneously occupies multiple channels in the time, frequency, and code domains. Orthogonal allocation prevents interference between channels within the same node, while cyclic shifting prevents interference from neighboring nodes. Synchronization identifiers and error correction codes are embedded at the transmitting end, ensuring that echo identification and error correction can be performed simultaneously at the receiving end. The design of robust spectral shape suppression and the insertion of an adaptive suppression cohesion zone at the center of the spectrum are based on understanding the mechanism of multipath echoes: multipath causes sidelobe noise to concentrate in the frequency band adjacent to the mainlobe. The central cohesion zone dynamically weakens the energy at the edge of the mainlobe, thereby reducing the relative height of the sidelobes and improving the feasibility of subsequent signal separation.

[0030] When the node transmitter outputs the ranging beam at a frequency of 1000 times per second, The distance measurement results for:

[0031] ;

[0032] in, is the number of subcarriers in the beam multi-frequency modulation code sequence; is the speed of light; is the adaptive amplitude weight corresponding to the subcarrier signal-to-noise ratio; For the The echo phase of the subcarrier in the current frame at the receiving end; For the The initial carrier phase of the subcarrier in this frame at the transmitting end; For the The round-trip flight time of each subcarrier in this frame; For the subcarrier center frequencies.

[0033] The signal tensor carries three-dimensional similarity information in the frequency domain, scale domain, and time domain. The traditional Euclidean metric cannot simultaneously describe the curvature differences of these three similarities. This technology introduces the Lorentz manifold metric, whose time axis has a negative weight characteristic. This can shorten the distance between nodes in areas with high similarity and fast evolution, and extend the distance between nodes in areas with low similarity and slow evolution, thus preventing high-density clusters from being diluted by time drag. In actual implementation, it is necessary to first construct an initial manifold grid on the three-dimensional similarity spectrum. To make the grid fit the high-density area, the system performs a bidirectional Manhattan sweep on each density layer, alternately collecting equipotential surface nodes horizontally and vertically, and then uses the Lorentz coordinate embedding method to map the discrete equipotential surfaces into surface grid patches. Finally, the grid patches are stitched together along the tangent of the spectral gradient. Each grid node collects spectral density, local curvature, and temporal similarity, which are normalized to form a variable weight vector. In the clustering stage, an equidistant expansion search is performed in the manifold metric space according to the variable weight vector. The minimum manifold path constraint is used to ensure that the clustering front only expands in the high similarity area. At the same time, the maximum weight gain is used to drive category fusion, thereby obtaining a feature cluster with maximum spectral coherence.

[0034] The feature clusters contain the core patterns of pavement micro-deformations. To predict pavement settlement and displacement deviations over time, a network structure is required that can model long-range dependencies while selectively memorizing local mutations. A deep temporal gated recurrent network uses gated units to determine the degree of information retention or forgetting. Stacking multiple layers can capture trends and oscillations at different time scales. When the network inputs the feature cluster sequence at the current moment, it outputs the predicted joint displacement-settlement deviation for the next time period. The system concatenates the current joint displacement-settlement deviation with the predicted joint displacement-settlement deviation in chronological order to produce a multidimensional parameter matrix for construction quality. The matrix rows and columns correspond to the three dimensions of spatial position, time window, and state indicator, respectively. The matrix values ​​characterize various state quantities, such as displacement, settlement, amplitude stability, and coherent diffusion. These comprehensive dimensions enable the matrix to simultaneously represent the lateral uniformity, longitudinal continuity, and temporal stability of the current construction state.

[0035] However, multidimensional parameter matrices are data-intensive, and their direct use in decision-making carries the risk of redundancy and overfitting. This method uses principal component analysis to reduce the matrix's dimensionality. The analysis first standardizes the dimensions across all dimensions to ensure comparable variance measures. A sliding covariance grid is then established, and time-series weights are used to adjust the influence of near- and far-flung data. Principal component extraction is performed according to the principle of increasing variance absorption rate. The proportion of the matrix variance information consumed by the extracted data is dynamically adjusted. During periods of high volatility, the number of principal components is reduced to highlight significant events, while during periods of stability, the number of principal components is increased to maintain fine-grained monitoring. The product of the stable principal axis vector and the matrix yields a compressed feature score sequence, which the system then adaptively aggregates to generate a comprehensive quality index column. The comprehensive quality index column is then compared against a construction design tolerance library in layers. If the index falls within each layer's threshold, the corresponding construction status is marked as acceptable; otherwise, it is marked as abnormal and a deviation grade is output. Continuous fluctuations in the index column can also indicate gradual deterioration or improvement in quality, providing guidance for on-site decision-making.

[0036] Furthermore, the system divides the complete frequency band into several sub-bands within the same frame time slot. Each sub-band has a clear and constant bandwidth interface. This hard division ensures that even if there is a slight local oscillator drift between different nodes when high-density nodes are deployed in parallel, there will be no crosstalk. Subsequently, dedicated start and end synchronization markers are inserted into each sub-band. The start synchronization marker provides a reference for the start point of transmission, and the end synchronization marker determines the alignment of the frame end. The two markers together constitute an alignment benchmark that can be observed simultaneously in the time domain and the frequency domain, so that all sub-bands remain aligned in the time-frequency plane and thus form a basic beam multi-frequency modulation coding sequence frame. Inter-layer orthogonal code diffusion is implemented within this basic beam multi-frequency modulation coding sequence frame. Each sub-band is mapped to a unique and pairwise orthogonal pseudo-random diffusion code. The cyclic characteristics of the pseudo-random sequence are used to achieve uniform energy spreading, thereby reducing the possibility of single-frequency spike interference; the orthogonal characteristics also ensure that the cross-correlation between different layers in the same frame is approximately zero.

[0037] To eliminate the low-frequency beat effect generated by the pseudo-random spreading code and the modulated subcarriers during long-term operation, the system introduces inter-code cyclic shifting. By circularly shifting the pseudo-random spreading code between consecutive frames, the spreading codes of adjacent sub-bands are offset by a fixed number of bits clockwise or counterclockwise relative to the previous frame. This shifting creates a self-synchronous interference suppression mechanism: when an echo from a sub-band happens to leak into an adjacent frequency, the correlated shift introduced by the cyclic shift allows the adjacent frequency receiver to easily identify the source of the interference and cancel it out during signal reconstruction, thus ensuring crosstalk during multi-frequency parallel transmission. After completing inter-layer orthogonal code spreading, the system performs time-frequency interleaving of the coded frames according to a preset interleaving matrix. The interleaving matrix is ​​designed to ensure coprime rows and columns and diagonal crossing, allowing adjacent time slots and adjacent sub-bands to alternate positions. The resulting interleaved sequence is distributed in a checkerboard pattern in the time-frequency binary domain. This not only disperses any potential periodic phase noise but also averages the impact of transient channel fading on subsequent decoding. Then, bipolar phase randomization is implemented on the interleaved sequence. The specific approach is to introduce a pseudo-random phase flip mode and write the phase randomization mask into the frame header of the beam multi-frequency modulation coding sequence. The mask is updated once per frame. By randomly switching the transmission phase between 0 degrees and 180 degrees, any potential interceptor will find it difficult to reconstruct valid information even if they synchronize to the carrier. The receiving end performs inverse flipping according to the frame header mask to achieve lossless recovery.

[0038] To further improve error resilience, a multi-level self-correction error-correction embedded segment is embedded within the beam multi-frequency modulation code sequence frame embedded with a phase randomization mask. The frame-level block check code is responsible for locating short-duration burst errors, and the cross-frame cyclic check code provides boundary protection for randomly distributed sparse errors. The combination of the two balances local correction speed and global redundancy efficiency. A combination of flipped parity check and redundancy segmentation is used within the embedded segment to maintain stable error correction capabilities despite scattering fluctuations caused by changes in the route environment. After the embedding is completed, the system performs spectral robust suppression shaping on the beam multi-frequency modulation code sequence frame. A hybrid shaping strategy of symmetric Hanning window and raised cosine window is used to simultaneously act on the sideband energy of each sub-band. The bidirectional decremental shaping results in a bell-shaped spectral slope, with rapid energy decay at the edges and high energy retention at the center. This reduces both emission spectrum leakage and adjacent frequency interference.

[0039] Taking into account that actual highway construction sites are often accompanied by metal reflective surfaces generated by large mechanical equipment such as concrete mixers and road rollers, which causes multipath echoes to concentrate in the frequency band near the main lobe and form sidelobe noise peaks, the present invention is designed to insert an adaptive suppression condensation zone in the center of the spectrum. The system first extracts the emission spectrum envelope in real time and determines the central energy threshold surface, and then dynamically reduces the main lobe width within the central threshold surface, further converging the energy toward the center and making a smooth transition in the connecting band at the same time. Since the condensation zone width and decreasing depth are adaptively adjusted with the echo monitoring results, the sidelobe noise suppression effect can be maintained stable for a long time. When a sudden change in the multipath echo characteristics is detected, the condensation zone is immediately reconstructed to ensure that the emission spectrum and the suppression strategy are updated synchronously. The above-mentioned continuous multi-link combination enables the beam multi-frequency modulation coding sequence processing to still provide high signal-to-noise ratio, high error correction margin and low crosstalk performance in the harsh highway construction site environment, providing highly resolvable original signals for the subsequent multi-scale wavelet transform and tensor sparse kernel decomposition links. With the help of this signal, the system can compare the distance measurement changes between nodes in real time, and capture road surface settlement and displacement information at the micron to millimeter level, so that the dynamic detection method of highway construction quality is both real-time and accurate, providing project managers with reliable quality trend data support.

[0040] Furthermore, to suppress the sidelobe noise introduced by multipath echoes in the spectrum and ensure a high signal-to-noise ratio (SNR) for the ranging beam, the system performs a specialized process at the final stage of the entire transmission chain, inserting an adaptive suppression zone at the center of the spectrum for the embedded multi-frequency modulation code sequence frame. This process involves capturing the overall spectral envelope of the transmitted signal in the frequency domain in real time. The system then accurately locates the center point with the highest and most symmetrical energy distribution. Symmetrical energy thresholds are then determined on either side of this point. These thresholds are then used to define the start and end boundaries at the center of the spectrum, forming an adaptive suppression zone. Within the adaptive suppression zone, the energy at the mainlobe edges is forcibly suppressed in a decreasing amplitude pattern from the center outward. A gradual transition zone is constructed outside the boundary, creating a smooth transition within the zone. This ensures that the suppression curve remains continuous and consistent across the entire spectrum, without creating new sharp inflection points or abrupt depressions.

[0041] To ensure this suppression process doesn't become ineffective over time, the system dynamically adjusts the width and decrement depth of the adaptive suppression cohesion zone during each transmission cycle using echo monitoring results from the node's receiving end. This correction logic is driven by the real-time ratio of the sidelobe noise peak to the mainlobe peak in the echo. When the ratio increases, indicating sidelobe elevation, the system immediately expands the cohesion zone width or increases the decrement gradient. When the ratio decreases, the system appropriately relaxes the width to avoid excessive suppression and loss of mainlobe effective power. This adaptive process is fully automated, requiring no on-site manual intervention. Therefore, it can maintain a low level of sidelobe noise in construction environments where large equipment such as rollers and sprinklers are operating in a staggered manner and reflective surfaces are changing instantaneously. Furthermore, to address sudden changes in multipath echo characteristics caused by the removal of metal formwork or the exposure of rebar cages during the concrete pouring phase, the system incorporates an event detection unit within the echo monitoring module. This unit uses a rate-of-change threshold to determine whether a sudden change has occurred. Once a sudden change is detected, a reconstruction command is sent to the transmitter, which immediately triggers adaptive reconstruction of the suppression cohesion zone. The reconstruction process first clears the existing cohesion zone parameters, then recalculates the center, threshold, and boundary using the new spectral envelope after the sudden change. The real-time control table of the shaper is then updated according to the latest amplitude reduction strategy, ensuring that the emission spectrum shape and suppression effect remain synchronized. This ensures that regardless of any changes in the reflective environment during highway construction, the laser ranging link can suppress the sidelobe noise introduced by multipath to a constant and controllable low level. This provides a clean and stable source of raw data for subsequent high-level signal processing steps such as multiscale wavelet transform, tensor sparse kernel decomposition, and variable-weight Lorentz manifold metric clustering. This improves the accuracy of the calculation of the predicted joint deviation of displacement and settlement and the multidimensional parameter matrix of construction quality, and enhances the reliability of highway construction quality inspection results derived from principal component analysis.

[0042] Furthermore, implementing inter-layer orthogonal code diffusion on the basic beam multi-frequency modulation code sequence frame is a key process that determines signal purity and the ability of nodes to prevent crosstalk. The system first writes a unique and pairwise orthogonal pseudo-random diffusion code index field within each sub-band divided by the beam multi-frequency modulation code sequence frame. Leveraging the orthogonal nature of the field, the system isolates the different sub-bands transmitted by the same node into independent channels with extremely low cross-correlation, ensuring that the receiver can accurately determine the sub-band attribution by locating the correlation peak even in high-scattering noise environments. The system then performs inter-code cyclic shift using the frame sequence number as the seed, shifting the pseudo-random diffusion codes of adjacent sub-bands by a fixed number of code positions clockwise relative to the previous frame. This shift creates a circular arrangement of diffusion codes on the timeline that rotates continuously with the frame sequence number. This is equivalent to continuously refreshing the relative offset angle across sub-bands at a transmission rhythm of one thousand times per second, making it difficult for any interference originating from a fixed location to maintain stable coupling across consecutive frames. In order to align this dynamic misalignment with the node's local reference, the system sets up a special adaptive trigger gate at the node's transmitter. The adaptive trigger gate monitors the displacement progress in real time. When it detects that the inter-code cyclic shift is completed, it immediately latches the bit sequence, writes the current sub-band boundary into the local register, and broadcasts it to the receiving front end of the same node through the synchronous bus, forming a physical layer self-synchronization closed loop.

[0043] After obtaining the latched bit sequence, the receiver reversely recovers the pseudo-random spreading code according to the same cyclic shift sequence, reconstructing the original orthogonal relationship through the correlator array. This automatically cancels cross-subband crosstalk and leverages orthogonality to maintain maximum channel capacity in the presence of frequency offset or amplitude imbalance. When large-scale machinery movement or steel formwork removal at the construction site causes a sudden increase in multipath reflection paths, the correlation peaks will statistically manifest as a sudden increase in the spreading code alignment deviation. If the inter-code alignment deviation exceeds the threshold, the system determines that the current cyclic shift strategy no longer meets the isolation requirements and immediately triggers the next cyclic shift cycle. The spreading code index field iterates forward clockwise to a fixed code position. Simultaneously, the adaptive trigger gate resets the latch counter, ensuring that the non-interference performance of multi-frequency parallel transmissions is updated in real time with the environment. Because the entire process is driven entirely by frame sequence numbers and relies on hardware clock retention, the high-precision laser ranging signal maintains a stable pseudo-random orthogonal pattern under the complex electromagnetic and optical conditions of highway sites. This provides high signal-to-noise, high orthogonality, and low mutual interference raw input for subsequent computational stages such as multi-scale wavelet transform, tensor sparse kernel decomposition, and variable-weight Lorentz manifold metric clustering. This enables the estimation accuracy of the joint deviation of predicted displacement and settlement and the multi-dimensional parameter matrix of construction quality to remain at the millimeter or even micron level, further improving the reliability and real-time performance of dynamic highway construction quality monitoring results.

[0044] Furthermore, the receiving node first samples the road surface scattered echoes in continuous time and performs a multiscale wavelet transform. Energy bands and phase shifts at different scales are projected onto a two-dimensional grid formed by the intersection of the frequency and scale domains. These are then stacked in the sampling order to form a three-dimensional multiscale power spectrum tensor. This tensor is represented by frequency band partitions and scale hierarchies in the spatial dimension and sequentially arranged as a frame sequence in the temporal dimension. As a result, it exhibits both a sparse, blocky distribution and traces of long-range evolution. To enable subsequent algorithms to capture the core structure that truly reflects road surface microdeformation, the system first performs multidirectional resampling on the multiscale power spectrum tensor. Multidirectional resampling simultaneously adjusts the sampling step size in the frequency, scale, and time domains to compress the original tensor, which is too high in resolution and contains redundant noise, to a range with minimal information loss, thereby reducing the computational burden and eliminating high-frequency artifacts. The system then establishes an initial tensor sparse kernel index table for the compressed tensor. This index table divides the tensor into equal-volume kernel blocks and records statistics such as energy density, neighbor correlation, and temporal persistence for each kernel block.

[0045] The core concept of sparse kernel decomposition is the assumption that useful information within the signal tensor exhibits strong block correlation within a three-dimensional neighborhood, while ambient noise is scattered as discrete points or fine fragments. Therefore, the system sets a sparsity density threshold in the initial index table of the tensor's sparse kernels and employs an iterative elimination strategy to gradually remove weakly correlated kernel blocks. In each iteration, the correlation weights of all kernel blocks with their three-dimensional neighborhood are calculated. Then, weakly correlated kernel blocks below the sparsity density threshold are marked for deletion, while the remaining kernel blocks are marked as retained. The deletion process simultaneously captures the removed weakly correlated kernel blocks and aggregates them in real time to form a noise tensor. The retained kernel blocks are reaggregated to form the signal tensor after each iteration. This iterative process continues until the rate of change in the number of retained kernel blocks falls below a predetermined convergence threshold, ensuring that the final signal tensor contains only highly correlated, highly sparse, and temporally continuous three-dimensional energy blocks, while the noise tensor concentrates on low-energy, low-correlation information fragments. The system then expands the signal tensor layer by layer along the frequency domain, scale domain, and time domain, and calculates the three-way similarity weights between kernel blocks through a sliding window. The three-way similarity weights comprehensively consider frequency proximity, scale similarity, and time synchronization, thereby forming a primary correlation matrix covering the entire tensor.

[0046] The primary correlation matrix still retains the inherent multi-resolution characteristics of a tensor, but the weight distribution may visually appear as discrete stripes, making it difficult to directly use for manifold clustering. To make the correlation information more coherent and usable for subsequent variable-weight Lorentz manifold metric clustering, the system applies piecewise progressive mapping and cross-stretching to the primary correlation matrix. The piecewise progressive mapping divides the weight interval into multiple levels, dynamically amplifying or compressing each level according to a nonlinear mapping function. This appropriately elevates low-weight regions and softens high-weight regions, preventing extreme values ​​from dominating the overall structure. The cross-stretching process alternates between reference lines in the frequency domain scale plane and the time axis, simultaneously extending local high-weight ridges along two orthogonal paths to fill gaps in the primary correlation matrix caused by sparse sampling. After these two reshaping steps, the primary correlation matrix is ​​converted into a three-dimensional similarity spectrum, which visually presents continuous and smooth energy peaks and valleys. This 3D similarity spectrum fully preserves the frequency domain fingerprint of road deformation while enhancing the temporal evolution context with a stretched structure. At this point, the three-dimensional similarity spectrum has all the necessary conditions to provide high-resolution input for variable-weight Lorentzian manifold metric clustering, enabling subsequent feature cluster extraction to obtain clear, noise-free spectral coherence peaks. Through this complete tensor sparse kernel decomposition process, this method purifies signal tensors representing true pavement settlement and displacement patterns from complex construction site background noise, while simultaneously constructing a structured, clusterable three-dimensional similarity spectrum, laying a solid data foundation for dynamic prediction and quality assessment. This separation strategy, which relies solely on the tensor's internal sparse features and similarity topology, does not require the introduction of a priori environmental models and is adaptable to construction conditions in different seasons, with different materials and different combinations of machinery and equipment. Therefore, it significantly improves the robustness and universality of dynamic highway construction quality detection methods to changing site conditions.

[0047] Spectral coherence evaluation index of three-dimensional similarity spectrum for:

[0048] ;

[0049] in, The signal tensor is indexed in the frequency band , scale index , time index The power spectral density at ; is the power spectrum density of the corresponding position of the noise tensor; 、 and are the discrete partition numbers of frequency domain, scale domain and time domain respectively; is the power spectral density of the noise tensor at the corresponding position.

[0050] Furthermore, the 3D similarity spectrum is responsible for explicitly geometrizing the high-order correlations of the signal tensor, while variable-weight Lorentz manifold metric clustering is responsible for extracting feature clusters from this geometrized image that best represent the evolution of road micro-deformation. The system first constructs an initial Lorentz manifold mesh based on spectral density over the entire 3D similarity spectrum. This meshing process treats the spectral density as a source of spatiotemporal curvature. Regions with high curvature are densely packed with grid nodes, while regions with low curvature are appropriately sparse, allowing the grid to naturally conform to the energy distribution. Each grid node is then assigned a variable weight vector, which is a normalized combination of the spectral density, local curvature, and temporal similarity at the node's location. The vector dimensions maintain a one-to-one correspondence with the three axes of the 3D similarity spectrum, thereby unifying the importance of different scales within a single metric framework. To ensure that weights automatically adjust to local structure, the system performs a dual-threshold adaptive update of the variable weight vector based on the node's local spectral gradient: when the spectral gradient falls below the first threshold, the node is considered to be on a plateau, and the weight decreases slowly to avoid over-emphasizing the flat background; when the spectral gradient exceeds the second threshold, the node is considered to be on a ridge or crack, and the weight increases rapidly to highlight the prominent feature; linear interpolation is used to smoothly transition between the two thresholds. After the update, the variable weight vector presents a multi-scale weight field in three-dimensional space. The continuity of the weight field ensures that weight changes on any section of the surface are derived from the same source as changes in the spectral gradient.

[0051] Using the variable-weight Lorentz manifold metric, the system performs a layer-by-layer equidistant expansion search of grid nodes. The Lorentz manifold metric uses negative weights on the time axis. Its physical meaning is to bring nodes with fast temporal evolution and high similarity closer together, while moving nodes with slow temporal evolution and low similarity farther apart, thereby maintaining the spatial aggregation of rapidly changing features. During this layer-by-layer equidistant expansion search, the system begins at the node with the highest spectral density peak and expands the search layer outward according to the manifold equidistant principle. With each expansion step, the manifold metric distance between nodes is recorded in real time. The manifold metric distance depends on both the geometric path length and the variable weight difference, allowing it to spatially circumvent valleys with steep weight drops while rapidly advancing along ridges with similar weights. When the expansion boundary reaches the manifold metric distance threshold, the current search layer stops, and the system begins iteratively merging node sets within the visited nodes according to the minimum manifold path and maximum weight gain principles to generate candidate feature clusters. The merging adopts a greedy strategy: priority is given to connecting the node pairs with the shortest manifold path and the largest weight gain at both ends, to avoid low-value nodes from occupying the cluster center, and to prevent nodes that are too far away from each other from destroying internal consistency.

[0052] After obtaining the candidate feature clusters, the system calculates the spectral coherence for each cluster. The spectral coherence is an index derived from the consistency of the accumulated energy of all nodes within the cluster on the three-dimensional similarity spectrum. To ensure that the final output feature clusters are both locally dense and globally representative, the system performs an incremental spectral coherence screening of the candidate feature clusters: first, they are sorted in descending order by spectral coherence, and then the redundant coverage between adjacent clusters is evaluated one by one. If two clusters are highly overlapping in space or time and the spectral coherence difference is less than the set step threshold, the system retains the cluster with higher coherence and discards the cluster with lower coherence. If the overlap is less than the threshold, both clusters are retained to avoid over-compression that may lead to the loss of key patterns. After multiple rounds of incremental screening, the feature cluster with the highest spectral coherence is determined as the final output. This feature cluster not only encompasses the most active microscopic settlement patterns of the current construction pavement phase, but also uses a variable-weight Lorentzian manifold metric to ensure high similarity within the cluster in both the temporal and spatial dimensions. This maximizes the activation of gated units in the subsequent deep temporal gated recurrent network to capture long-term and short-term dependencies, thereby improving the accuracy of predicting the joint deviation of displacement and settlement. Leveraging this clustering process, the dynamic highway construction quality detection method avoids the cluster boundary distortion problem associated with traditional Euclidean clustering in high-curvature spectral spaces. Furthermore, through adaptive weighting, it introduces the ability to inherently adjust differentiated features in the frequency, scale, and time domains. This ensures the stable extraction of feature clusters with maximum spectral coherence across multi-scenario, multi-material, and all-weather construction sites, providing a high-confidence, high-resolution foundational input for the dynamic detection framework.

[0053] Furthermore, the system first divides the three-dimensional similarity spectrum into continuous density layers according to energy levels based on the spectral density grading rule. The spectral density grading rule follows the principle of equal energy decrease, making the energy distribution within each density layer as smooth as possible and maintaining a monotonically decreasing energy distribution between layers. This hierarchical structure provides a clear topological order for subsequent surface extraction. After obtaining the density layer, the system performs a bidirectional Manhattan sweep on each density layer to generate an equipotential surface. The bidirectional Manhattan sweep selects to move alternately along the horizontal and vertical directions of the Cartesian coordinate plane, captures equal energy contour points through staggered scanning, and completes the missing nodes in real time, thus drawing a closed and non-self-intersecting equipotential surface for each density layer. In order to transform the discrete equipotential surface into a computable surface entity, the system uses the Lorentz coordinate embedding method to embed the equipotential surface into a surface mesh. The Lorentz coordinate embedding method utilizes the negative weight characteristic of the time axis of the three-dimensional similarity spectrum to ensure that the surface mesh naturally contains time curvature information while being smoothly embedded, ensuring that the subsequent manifold metric can perceive the speed difference of dynamic evolution.

[0054] After the surface meshes are generated, the system performs directional weaving along the spectral gradient tangent, seamlessly splicing all the surface meshes into a Lorentz manifold initial mesh covering the entire domain. The core idea of ​​directional weaving is to spread the mesh edges along the direction of maximum spectral gradient and then cross-stitch in the tangent direction. This avoids creases caused by manual stitching while retaining the continuous direction of the main spectral density pattern. When all the surface meshes are spliced ​​together, an initial Lorentz manifold mesh that is completely isomorphic to the energy topology of the three-dimensional similarity spectrum is formed. The nodes in this mesh are dense in the high-density ridge peak area and sparse in the low-density valley area, which meets the differentiated neighborhood resolution requirements of subsequent clustering.

[0055] After the initial Lorentzian manifold mesh is established, the system simultaneously collects neighborhood spectral density, local curvature, and temporal similarity for each mesh node. Neighborhood spectral density describes the energy concentration around the node, local curvature characterizes the curvature of the mesh surface at that point, and temporal similarity measures the degree of coherence of the signal evolution over time corresponding to that point. These three metrics are spatially orthogonal and complementary, allowing them to be combined into the original feature vector. To ensure comparability of features of different dimensions within the same metric space, the system performs a normalization and reweighting cycle on the original feature vector. The normalization phase maps each dimension to a uniform range from zero to one. The reweighting phase dynamically adjusts the weight coefficients based on the current construction scenario. For example, the weight of temporal similarity is appropriately increased during the high-frequency vibration phase of the roller, while the weight of spectral density is increased during the concrete pouring phase. This normalization and reweighting cycle is iteratively executed multiple times at each node until the contributions of all dimensions converge to the designed threshold, thus converting the original feature vector into a variable weight vector. The variable weight vector not only encodes the local geometry and dynamic information of the node, but also reflects the adjustment result of the intensity of the external environment on the significance of the intrinsic characteristics during construction, thereby greatly improving the adaptability of subsequent measurement calculations to different construction conditions.

[0056] In the final step, the system writes the variable weight vector into the grid node metadata, completing the initialization of the grid node weights. From this point on, the initial Lorentz manifold mesh becomes a weighted metric mesh that carries dynamic information. It can not only calculate the distance between nodes under the variable weight Lorentz manifold metric, but also respond in real time to the weight changes brought about by the subsequent dual-threshold adaptive update. Because the grid has uniformly encoded the spectral density, curvature, and time information in advance, its topological structure and weight distribution are highly consistent with the microscopic settlement pattern of the highway construction site. This provides dual sensitivity for feature cluster extraction that takes into account both geometric shape and time evolution speed. In the entire dynamic highway construction quality detection process, this approach creates a structured and interpretable data foundation for the input of deep time series models, thereby significantly improving the credibility of the predicted joint deviation of displacement and settlement, as well as the accuracy of the subsequent comprehensive quality index in reflecting the actual construction quality.

[0057] Furthermore, the algorithm first selects the four boundaries of a certain density layer as the starting reference line. The boundary selection follows the principle of gradually decreasing energy, so that the horizontal and vertical sweep paths have clear threshold coordinates at the beginning. The system then moves row by row on the horizontal sweep path. Each time it moves a row, it marks the candidate nodes of the equipotential surface in real time according to the multiple intersection positions on the current row that intersect with the density threshold surface. This can capture the continuous trend of the energy contour line in the horizontal direction, while reducing the horizontal jump points caused by discrete sampling. After the horizontal sweep is completed, the vertical sweep path is immediately switched to supplement the equipotential surface gap nodes that were not touched in the horizontal process in a vertical column-by-column manner. The two sets of staggered sampling in the horizontal and vertical directions compensate each other, and finally form a two-way closed node network in the density layer that retains high-resolution curved details and avoids the omission of isolated islands. To ensure that the generated node network does not have cracks or overhangs in subsequent geometric calculations, the system performs neighborhood connectivity detection on the bidirectional closed node network, eliminating all isolated nodes and small clusters unrelated to the main connected branches, and only retaining the main structure with a connectivity that reaches the set threshold. This ensures that the surface converges normally during the subsequent interpolation fitting, resulting in a continuous and smooth equipotential surface.

[0058] After obtaining the equipotential surfaces, the system constructs local Lorentz tangent planes on each equipotential surface. These tangent planes not only provide a first-order approximation of the surface but also embed information about the negative weights of the time axis, allowing the surface to perceive differences in temporal evolution rates within a small area. Within each Lorentz tangent plane, the algorithm uses Lorentz coordinate embedding to map the surface segments into a two-dimensional surface parameter domain. This mapping process maintains angles, thus maintaining geometric conformality even in areas with ridges or valleys with dramatic changes in curvature. All surface parameter domains are concatenated sequentially in the order in which they were generated. The system then uses common control points at overlapping edges for flexible fusion, avoiding scale misalignment caused by seams, thereby generating a fully conformal surface mesh. Finally, a density layer identifier and spatial index are assigned to each surface mesh, making it globally unique. This registration step enables subsequent meshing, weight assignment, manifold metrics, and even deep temporal analysis to quickly locate target segments within a unified coordinate system. This enables precise mapping of energy terrain to actual road mileage at highway construction sites. The entire process, from sampling and connectivity detection to embedding and registration, is completed with purely geometric means, without relying on any statistical priors. Therefore, it can operate stably on construction surfaces of different materials, thicknesses, and compaction degrees, providing the dynamic detection framework with high-resolution, topologically correct, and time-sensitive surface grid basic data, greatly improving the credibility of subsequent feature clustering and settlement prediction.

[0059] Furthermore, the system writes the dynamic feature streams transmitted in real time by each node and undergoing multi-level signal purification into a multidimensional construction quality parameter matrix. The dimensions of this matrix cover multiple indicators such as spatial position, ranging displacement, settlement trend, amplitude stability, and coherent diffusion. Once the amount of data rapidly increases with the construction progress, dimensional redundancy and scale inconsistency will hinder engineers' rapid perception of the overall quality evolution. Therefore, the multidimensional construction quality parameter matrix must first be subjected to fractal normalization. By applying linear mapping and nonlinear compression to columns of different dimensions, the data in each column is scaled to a comparable range from zero to one. During the mapping, an inverse normalization index is generated to facilitate the subsequent derivation of the dimensionality reduction results back to the original engineering dimensions. The resulting standardized matrix retains all information but eliminates dimensional differences, laying the foundation for calculating the synergy between related features. The system then constructs a sliding covariance grid along the construction timeline, calculates the sample covariance for each time window and writes it into the grid cell, and then dynamically updates the covariance weight network based on the temporal correlation: when the correlation between consecutive windows is high, the weight network thickens the connection in space to emphasize stable patterns; when the correlation drops suddenly, the weight network weakens the connection in the corresponding channel to highlight sudden events.

[0060] The covariance weight network generated in this way is input into the principal component analysis core engine, which performs eigendecomposition on the sliding covariance grid and selects principal axis vector sets from high to low according to the progressive variance absorption rate criterion. Extraction stops when the cumulative variance coverage reaches a preset threshold, ensuring that only the characteristic directions that are most explanatory of the overall variation are retained. The core engine then performs an orthogonal rotation on the principal axis vector set, making each principal axis independent and physically interpretable. The resulting stable principal axis is not affected by even minor perturbations from single outliers during continuous monitoring, and thus can represent the baseline trend of construction quality over the long term. The system then performs a cascaded coordinate transformation on the standardized matrix under the stable principal axis, projecting the high-dimensional state into the principal component space, generating a compressed feature score sequence; this sequence is compressed to tens or even single digits in terms of dimension, but maintains the millisecond refresh rate of the original data on the time axis, thus balancing real-time performance with visualization convenience. To further transform the compressed feature score sequence into a single, decision-enhancing metric, the system performs adaptive contribution aggregation on the compressed feature score sequence, based on the time scales of minutes, hours, shifts, or days defined by the construction schedule. A weighted scheduler dynamically increases or decreases the contribution of each principal component within the same time window, taking into account the quality risk priorities of the current construction phase. This generates a comprehensive quality index column, which is presented on a percentage or thousandth scale, providing project managers with a clear and intuitive health curve.

[0061] Finally, the system compares the comprehensive quality index column with the design tolerance library in layers. The design tolerance library encodes the allowable fluctuation ranges for different processes, materials, and equipment combinations into multi-layered threshold intervals. The dwell time of the comprehensive quality index within each interval becomes the quantitative basis for quality compliance. If the index exceeds the limit, the system not only indicates the abnormal state but also traces the cause of the deviation to the specific original dimension through inverse normalization indexing, indicating the mileage, material, or piece of equipment causing the quality risk. At this point, the highway construction quality inspection results are output in a closed loop. Engineers can observe the index curve and spatial heat map in real time on the visualization panel. They can also export abnormal segments and hand them over to on-site construction personnel for immediate rectification. This achieves data-driven, refined quality control, maintains millimeter-level ranging accuracy and second-level quality update cycles throughout the entire process, and provides reliable protection for the construction safety and lifespan of high-grade highways.

[0062] Comprehensive quality index column for:

[0063] ;

[0064] in, For the moment Comprehensive quality index; is the number of principal components participating in the contribution aggregation; For the The principal component at time Adaptive weights of For the The principal component at time 's feature score; is the reference value of the kth principal component under the design reference state; The kth principal component tolerance given by the design tolerance library.

[0065] The following examples are based on a length of The linear test roadbed is the object, and the monitoring time is . Lay on the left and right of the center line High-precision laser ranging terminal nodes, horizontally from the center line . Vertical coordinate . Left node coordinates , right node coordinates The node has its own three-axis inertial navigation system, which is initially aligned with the unified reference coordinate system. .

[0066] The parameters of the multi-layer frequency domain stack are set to three layers, and the number of subcarriers in each layer is ; The example only writes the lowest four subcarriers Start and end synchronization marker time slot width . Pseudo-random spreading code length Frame Sequence Number from Count, inter-code cyclic shift .

[0067] Single frame ranging solution (based on the left node , No. frame as an example), echo phase difference measurement ; Round trip flight time is locked by correlation peak ; Signal-to-noise ratio ; Mapped to amplitude weight ; Numeric value .

[0068] Fusion ranging formula ; bring in have to ; and initialization Comparison, instantaneous lift .

[0069] In continuous The frame window performs three-layer discrete wavelet packet on each frame of ranging sequence ; Frequency band division , scale division . Get the multi-scale power spectrum tensor Multidirectional resampling step size form Core Block .

[0070] Define kernel block sparse density ;Threshold . Iteration: Wheel culling Block, reserved ;No. Wheel culling Block, reserved ;No. Wheel retention change rate , stop. Keep the sum of the core block power Noise power .

[0071] 3D similarity spectrum and coherence, calculate weights ;make , spectral coherence .

[0072] Lorentz manifold initial mesh, five energy levels, nodes obtained by bidirectional Manhattan sweep of equipotential surfaces The original feature vector of each node ;in is the neighborhood spectral density, is the local curvature, is the time similarity. Normalized ; Reweighted . Get the variable weight vector .

[0073] Variable weight Lorentz manifold metric clustering, negative weight time axis parameter Manifold distance ;

[0074] Layer-by-layer equidistant threshold The search terminates and the candidate feature cluster is obtained Shortest manifold path and maximum weight gain After normalization, greedy merging and filtering are performed to retain the cluster with the highest coherence, and the number of nodes .

[0075] Deep temporal gated recurrent network prediction, number of gated units , History Window . Input feature sequence ; Output next frame prediction .

[0076] Construction quality multidimensional parameter matrix, number of matrix columns (node, frequency band, scale, displacement, sedimentation, coherent diffusion, etc.) Write the current and predicted deviations, time index .

[0077] Sliding covariance and principal component analysis, sliding window length . Covariance weights ; Variance Absorption Rate Three principal components were selected .

[0078] Comprehensive quality index, adaptive weighting Tolerance .

[0079] Comprehensive Quality Index .

[0080] Quality judgment, threshold setting: ;current , which is abnormal. Inverse normalization positioning found that the longitudinal displacement residual corresponding to the principal component 1 exceeded the limit, and the position interval Insufficient compaction. The system issues a rework instruction and records the abnormal event.

[0081] Figure 2This article details the frequency domain characteristics and processing effects of the beam multi-frequency modulation code sequence. The horizontal axis represents the frequency range, and the vertical axis represents the power spectral density. Within a unified frame time slot, the target transmit frequency band is divided into four main sub-bands, labeled Sub-band 1, Sub-band 2, Sub-band 3, and Sub-band 4, according to a pre-set multi-layer frequency domain stacking rule. Each sub-band exhibits a unique spectral shape. Sub-band 1, located in the lower frequency region, exhibits a bell-shaped power spectral density curve that rises first and then falls, with a peak at approximately 150 Hz. Sub-band 2, located in the mid-low frequency region, has a relatively wide spectral envelope and a more uniform power distribution. Sub-bands 3 and 4, located in the mid-high and high frequency regions, respectively, exhibit increasing spectral characteristics. Dedicated start and end synchronization markers, represented by black rectangles, are inserted at the beginning and end of each sub-band to ensure precise alignment of the sub-bands in the time-frequency plane. These synchronization markers enable strict timing synchronization of the basic beam multi-frequency modulation code sequence frames. The inter-layer orthogonal code spreading effect shown in the figure is reflected by the mutual independence between different sub-bands. Each sub-band is mapped to a mutually orthogonal pseudo-random spreading code, effectively avoiding inter-code interference. Figure 2 The core feature is the adaptive suppression cohesion zone located at the center of the spectrum, marked with a dotted rectangular box. This cohesion zone is set to suppress the sidelobe noise caused by multipath echoes. The boundary position of the cohesion zone is determined by the symmetrical energy threshold surface, and an amplitude decreasing processing strategy is adopted inside it, while progressive transition bands are set on both sides of the boundary to ensure spectrum continuity. The figure also shows the application effect of the symmetric Hanning window and raised cosine window mixed shaping strategy. At the two ends of the spectrum, the sideband energy of each sub-band shows a bidirectional decreasing shaping feature, which effectively reduces the leakage of the emission spectrum. The entire spectrum analysis diagram clearly reflects the final frequency domain characteristics of the beam multi-frequency modulation coding sequence after the complete processing process, providing an optimized signal foundation for subsequent high-precision laser ranging.

[0082] Figure 3The complete process and results of sparse kernel decomposition of a multiscale power spectrum tensor are presented in a three-dimensional format. The three coordinate axes represent the time domain, frequency domain, and scale domain, respectively, forming the basic coordinate system for tensor analysis. After multidirectional resampling of the original multiscale power spectrum tensor, an initial tensor sparse kernel index table is established, providing the data foundation for subsequent decomposition operations. The cube structure on the left of the figure represents the signal tensor extracted after sparse kernel decomposition. This signal tensor contains multiple black squares, representing strongly correlated kernels retained after iteratively removing weakly correlated kernels based on a sparse density threshold. These kernels exhibit a regular distribution in three-dimensional space, reflecting the spatial clustering of effective signal components. The clear boundaries of the cube of the signal tensor indicate good spatial continuity and integrity of the aggregated signal components. The cube structure on the right of the figure represents the synchronously captured noise tensor. Unlike the signal tensor, the noise tensor exhibits a sparse, dotted distribution, with small dots representing the removed weakly correlated kernels. These noise components are relatively randomly distributed in space and have a low density, fully demonstrating the incoherent nature of the noise signal. The far right side of the figure shows a three-dimensional similarity spectrum constructed based on the signal tensor. This spectrum uses contour lines to represent the similarity distribution in different regions. The density and shape of the contour lines reflect the strength of signal correlation in three-dimensional space. After piecewise progressive mapping and cross-stretching, the primary correlation matrix forms a three-dimensional similarity distribution pattern with a clear hierarchical structure. The central arrow marks the "tensor sparse kernel decomposition" process, clearly indicating the direction of transformation from the original tensor to the separated signal and noise tensors. The entire decomposition process demonstrates the advantages of the multi-scale wavelet transform, which can effectively identify and separate signal features at different scales, providing a reliable data foundation for subsequent feature extraction and quality assessment.

[0083] Figure 4This paper presents the detailed process and results of applying variable-weight Lorentzian manifold metric clustering to a three-dimensional similarity spectrum. The curved mesh structure in the figure represents the initial Lorentzian manifold mesh constructed based on the spectral density. These mesh lines exhibit nonlinear surface features, reflecting the spatial geometric properties of the Lorentzian coordinate system. The mesh construction follows the spectral density grading rule. Each equipotential surface curve corresponds to a specific density layer. These curves are generated via bidirectional Manhattan sweeps and mapped into a surface mesh using the Lorentzian coordinate embedding method. The figure shows three major feature clusters: Feature Cluster A, Feature Cluster B, and Feature Cluster C, represented by elliptical bounding boxes of different sizes. Feature Cluster A is located in a region of relatively low spectral coherence and contains five nodes. Its elliptical boundary is compact, indicating high similarity among the nodes within the cluster. Feature Cluster B is located in a region of medium spectral coherence and contains six nodes. Its distribution is relatively dispersed, reflecting the diversity of the data points in this region. Feature Cluster C is located in a region of high spectral coherence and contains seven nodes. Its elliptical boundary is the narrowest, indicating strong correlation in a specific direction. The black dots within each feature cluster represent grid nodes, each of which is assigned a variable weight vector, represented by a short straight line segment emanating from the node. The direction and length of the variable weight vector reflect the characteristics of the node's local spectral gradient. These vectors are dynamically adjusted according to a dual-threshold adaptive update strategy, forming a multi-scale weight field. The dashed lines in the figure connect representative nodes between different feature clusters, annotated with "minimum manifold path" and "maximum weight gain," reflecting the optimization criteria for node merging during the clustering process. The entire clustering process employs a layer-by-layer equidistant expansion search strategy, recording the manifold metric distance between nodes in real time. By iteratively merging the node set, candidate feature clusters are generated. Finally, the feature cluster with the highest spectral coherence is selected through increasing spectral coherence. The final clustering results shown in the figure fully demonstrate the superior performance of the Lorentz manifold metric clustering method in processing complex, high-dimensional data, providing high-quality feature input for subsequent prediction of deep temporal gated recurrent networks.

[0084] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions described in the above embodiments can still be modified, or some of the technical features thereof can be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A dynamic detection method for highway construction quality based on high-precision laser ranging, characterized in that: The method comprises: Step 1: Several high-precision laser ranging terminal nodes are evenly spaced on both sides of the centerline of the proposed highway. The spacing between the nodes is a pre-set fixed design spacing. The transmitter of each node transmits a ranging beam processed by a multi-frequency modulation coding sequence at a frequency of one thousand times per second, and simultaneously records the absolute three-dimensional coordinates of the node to establish a unified reference three-dimensional coordinate system. Step 2: The receiving end of each node receives the original echo signal scattered back by the road surface; a multi-scale wavelet transform is performed on the original echo signal to generate a multi-scale power spectrum tensor; a tensor sparse kernel decomposition method is used to separate the multi-scale power spectrum tensor into a signal tensor and a noise tensor, and a three-dimensional similarity spectrum is constructed; variable-weight Lorentz manifold metric clustering is applied to the three-dimensional similarity spectrum to obtain feature clusters with maximum spectral coherence; the feature clusters are input into a deep temporal gated recurrent network to obtain the predicted joint displacement and settlement deviation for the next time period; the current joint displacement and settlement deviation is aggregated with the predicted joint displacement and settlement deviation according to the time series to form a multi-dimensional construction quality parameter matrix, which represents the current construction quality status in a real-time, high-dimensional manner; Step 3: Perform principal component analysis on the multidimensional parameter matrix of construction quality to generate highway construction quality inspection results.

2. The method for dynamic detection of highway construction quality based on high-precision laser ranging as claimed in claim 1, characterized in that: In step 1, the process of processing the beam multi-frequency modulation coding sequence includes: within the unified frame time slot, according to the preset multi-layer frequency domain stack layer rule, dividing the target transmission frequency band into several sub-bands; inserting a dedicated start and end synchronization mark into each sub-band to keep the sub-bands aligned on the time-frequency plane to form a basic beam multi-frequency modulation coding sequence frame; implementing inter-layer orthogonal code diffusion on the basic beam multi-frequency modulation coding sequence frame, and mapping each sub-band to a mutually orthogonal pseudo-random diffusion code; through inter-code cyclic shift, realizing self-synchronization interference suppression between different sub-bands, ensuring that there is no crosstalk during multi-frequency parallel transmission; using a preset interleaving matrix to perform time-frequency interleaving rearrangement on the beam multi-frequency modulation coding sequence frame after orthogonal code diffusion; in the process of time-frequency interleaving rearrangement, adjacent time slots and adjacent sub-bands are aligned. The frequency bands are alternately swapped to obtain an interleaved sequence, which is then subjected to bipolar phase randomization: a pseudo-random phase flip mode is introduced to embed the phase randomization mask into the frame header of the beam multi-frequency modulation coding sequence; a multi-level self-correction error correction embedded segment is inserted into the beam multi-frequency modulation coding sequence frame embedded with the phase randomization mask, including a frame-level block check code and an inter-frame-level cyclic check code; the embedded segment uses a combination of flipped parity check and redundancy segmentation; the embedded beam multi-frequency modulation coding sequence frame is subjected to spectral robust suppression shaping: a hybrid shaping strategy of symmetric Hanning window and raised cosine window is used to perform bidirectional decreasing shaping on the sideband energy of each sub-band to reduce emission spectrum leakage; and an adaptive suppression condensation region is inserted at the center of the spectrum to suppress sidelobe noise caused by multipath echoes.

3. The method for dynamic detection of highway construction quality based on high-precision laser ranging as claimed in claim 2, characterized in that: The process of inserting an adaptive suppression cohesion zone at the center of the spectrum includes: extracting the overall spectral shape envelope of the embedded multi-frequency modulation coding sequence frame in real time to determine the center position of the spectrum and the symmetric energy threshold surface; opening the start and end boundaries of the adaptive suppression cohesion zone at the center of the spectrum based on the symmetric energy threshold surface, and setting progressive transition bands on both sides of the boundary; performing amplitude reduction processing on the interior of the cohesion zone according to the adaptive suppression strategy, and performing smooth transition processing on the transition band to make the spectrum shape continuous and consistent; in the subsequent transmission cycle of the multi-frequency modulation coding sequence frame of the beam, the echo monitoring results are used to dynamically correct the width and decrement depth of the cohesion zone to keep the sidelobe noise continuously suppressed; whenever a sudden change in the multipath echo characteristics is detected, the adaptive reconstruction of the cohesion zone is immediately triggered to ensure that the emission spectrum shape and the suppression effect are updated synchronously.

4. The method for dynamic detection of highway construction quality based on high-precision laser ranging as claimed in claim 3, characterized in that: The process of implementing inter-layer orthogonal code diffusion on the basic beam multi-frequency modulation coding sequence frame includes: loading a unique and pairwise orthogonal pseudo-random spreading code index field in each sub-band of the beam multi-frequency modulation coding sequence frame; performing inter-code cyclic shift with the frame sequence number as the seed, shifting the pseudo-random spreading codes of adjacent sub-bands clockwise by a fixed code relative to the previous frame; setting an adaptive trigger gate at the transmitting end of the node, and immediately latching the bit sequence when the inter-code cyclic shift is detected to achieve self-synchronization of the sub-band boundary; at the receiving end of the node, reversely recovering the pseudo-random spreading code according to the same cyclic shift sequence to automatically offset cross-sub-band crosstalk; if it is detected that the inter-code alignment deviation exceeds the threshold, immediately triggering the next cyclic shift cycle to dynamically maintain the mutual non-interference of multi-frequency parallel transmission.

5. The method for dynamic detection of highway construction quality based on high-precision laser ranging as claimed in claim 4, characterized in that: In step 2, the multi-scale power spectrum tensor is separated into a signal tensor and a noise tensor by using the tensor sparse kernel decomposition method, and the process of constructing a three-dimensional similarity spectrum graph includes: performing multi-directional resampling on the multi-scale power spectrum tensor and establishing a tensor sparse kernel initial index table; in the tensor sparse kernel initial index table, iteratively removing weakly correlated kernel blocks according to the sparse density threshold, and aggregating the retained kernel blocks to generate a signal tensor; synchronously capturing and removing kernel blocks to form a noise tensor; expanding the signal tensor layer by layer along the frequency domain, scale domain and time domain, calculating the three-way similarity weights, and generating a primary correlation matrix; applying piecewise progressive mapping and cross-stretching processing on the primary correlation matrix to obtain a three-dimensional similarity spectrum graph.

6. The method for dynamic detection of highway construction quality based on high-precision laser ranging as claimed in claim 5, characterized in that: In step 2, the variable-weight Lorentz manifold metric clustering is applied to the three-dimensional similarity spectrum to obtain the feature cluster with the maximum spectral coherence, which includes: constructing a Lorentz manifold initial grid based on spectral density in the entire domain of the three-dimensional similarity spectrum, and assigning a variable weight vector to each grid node; performing a double-threshold adaptive update on the variable weight vector according to the local spectral gradient of the node to form a multi-scale weight field; implementing a layer-by-layer equidistant expansion search on the grid nodes under the variable-weight Lorentz manifold metric, and recording the manifold metric distance between nodes in real time; iteratively merging the node set according to the principle of minimum manifold path and maximum weight gain to generate a candidate feature cluster set; performing spectral coherence increasing screening on the candidate feature cluster set, and outputting the feature cluster with the largest spectral coherence.

7. The method for dynamic detection of highway construction quality based on high-precision laser ranging as claimed in claim 6, characterized in that: The process of constructing a spectral density-based Lorentz manifold initial grid over the entire domain of the three-dimensional similarity spectrum and assigning a variable weight vector to each grid node includes: decomposing the three-dimensional similarity spectrum into continuous density layers according to the spectral density grading rule; performing a bidirectional Manhattan sweep on each density layer to generate equipotential surfaces, and mapping them into surface grid pieces using the Lorentz coordinate embedding method; weaving along the spectral gradient tangent to seamlessly splice all surface grid pieces to form a Lorentz manifold initial grid covering the entire domain; synchronously collecting the neighborhood spectral density, local curvature and temporal similarity at each grid node and splicing them into the original feature vector; converting the original feature vector into a variable weight vector through a normalized reweighting cycle; and writing the variable weight vector into the grid node metadata to complete the grid node weight initialization.

8. The method for dynamic detection of highway construction quality based on high-precision laser ranging as claimed in claim 7, characterized in that: The process of performing bidirectional Manhattan sweep on each density layer to generate equipotential surfaces and mapping them into surface meshes using the Lorentz coordinate embedding method includes: selecting the boundaries of the density layer as the starting reference lines and setting the horizontal and vertical bidirectional Manhattan sweep paths; collecting density threshold surface intersections row by row along the horizontal sweep path and marking candidate equipotential surface nodes in real time; switching to the vertical sweep path after completing the horizontal sweep, filling the equipotential surface gap nodes column by column to form a bidirectional closed node network; performing neighborhood connectivity detection on the bidirectional closed node network, eliminating isolated nodes, and obtaining a continuous and smooth equipotential surface; establishing a local Lorentz tangent plane on the equipotential surface, and converting the equipotential surface fragments into the surface parameter domain according to the Lorentz coordinate embedding method; splicing the surface fragments in the order of the surface parameter domain to generate a fully conformal surface mesh; writing the density layer identifier and spatial index for each surface mesh to complete the global registration of the surface mesh.

9. The method for dynamic detection of highway construction quality based on high-precision laser ranging as claimed in claim 8, characterized in that: Step 3 specifically includes: performing dimension normalization on the multidimensional parameter matrix of construction quality, generating a standardized matrix and retaining the inverse normalization index; constructing a sliding covariance grid, updating the covariance weight network based on the time series correlation, and inputting the obtained network into the principal component analysis core engine; the principal component analysis core engine extracts the principal axis vector set according to the progressive variance absorption rate criterion, and performs orthogonal rotation to obtain a stable principal axis; completing a cascade coordinate transformation of the standardized matrix under the stable principal axis to obtain a compressed feature score sequence; performing adaptive contribution aggregation on the compressed feature score sequence according to the time scale, and outputting a comprehensive quality index column; performing a hierarchical comparison on the comprehensive quality index column and the design tolerance library, marking the qualified state and the abnormal state, and forming the highway construction quality inspection result.

Images