An airborne laser radar method for classifying deep-sea land waveforms

Abstract

The application discloses a kind of airborne laser radar bathymetry sea-land waveform classification methods, it is related to marine surveying and mapping and remote sensing information processing technical field, comprising: obtaining the airborne laser radar bathymetry (ALB) full waveform data of target area, the full waveform data at least includes infrared laser waveform data and blue-green laser deep water channel and shallow channel waveform data;Based on the amplitude characteristics of the infrared laser waveform data, the first clustering is carried out to the collected laser foot point, and the initial classification result containing water point set, land point set and pending point set is obtained;Fusion laser foot point three-dimensional space position information, spatial density clustering analysis is carried out to the pending point set;By adopting waveform amplitude preliminary screening, spatial relationship correction and multi-feature fusion fine judgment three-level progressive strategy, different levels of classification uncertainty is solved layer by layer, effectively reduce the misclassification caused by noise, special feature and boundary fuzzy area, and the overall classification precision is significantly improved.

Description

Technical Field

[0001] This invention relates to the field of marine surveying and remote sensing information processing technology, and more specifically, to a method for classifying sea and land waveforms in airborne lidar depth measurement. Background Technology

[0002] Airborne lidar depth sounding technology is a highly efficient technique that rapidly acquires three-dimensional topographic information of coastal zones and shallow sea areas by emitting infrared and blue-green laser pulses and receiving the echo signals returned from the water surface, water body, and seabed. Accurately distinguishing whether laser footprints belong to land or water (i.e., land-sea waveform classification) is the foundation and key prerequisite for subsequent depth extraction, point cloud processing, and the generation of marine geographic information products.

[0003] Current land-sea waveform classification methods suffer from the following limitations: most methods rely solely on the amplitude and intensity of infrared waveforms for thresholding, making them highly susceptible to noise, waves, and low-reflectivity features (such as black mudflats), easily producing "salt and pepper" misclassifications. The classification process typically processes each laser footprint independently, neglecting the spatial continuity and clustering of adjacent footprints, resulting in poor classification of scattered floating objects in water bodies or small patches of stagnant water on land.

[0004] Therefore, there is an urgent need for an airborne lidar waveform classification method that can comprehensively utilize waveform physical characteristics, spatial distribution patterns, and multi-channel information to achieve high precision and robustness. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, this invention provides a sea-land waveform classification method for airborne lidar depth sounding. Through a multi-level progressive and feature fusion classification strategy, the accuracy and reliability of sea-land classification are significantly improved.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A method for classifying sea-land waveforms using airborne lidar (ALB) bathymetry includes: acquiring full waveform data of an airborne lidar bathymetry (ALB) system in a target area, wherein the full waveform data includes at least infrared laser waveform data and deep-water and shallow-water channel waveform data of blue-green lasers; performing initial clustering on the acquired laser footprints based on the amplitude characteristics of the infrared laser waveform data to obtain an initial classification result containing a water point set, a land point set, and a point set to be determined; fusing the three-dimensional spatial location information of the laser footprints to perform spatial density clustering analysis on the point set to be determined, classifying it into either the water point set or the land point set, generating an intermediate classification result; constructing a multi-channel waveform feature fusion model to extract the combined features of the deep-water and shallow-water channel waveform data of the blue-green lasers, re-discriminating the point set in the intermediate classification result that is adjacent to the water-land boundary region, identifying and correcting misclassified points in extremely shallow water areas; and outputting the final high-precision sea-land waveform classification result.

[0008] In a preferred embodiment, the initial clustering based on the amplitude characteristics of the infrared laser waveform data specifically includes: calculating the peak amplitude of the infrared laser echo waveform corresponding to each laser foot point; using an unsupervised clustering algorithm to cluster the peak amplitude, assigning points with amplitudes below a first threshold to the water point set, points with amplitudes above a second threshold to the land point set, and points with amplitudes between the first and second thresholds to the undetermined point set.

[0009] In a preferred embodiment, the spatial density clustering analysis based on the fusion of three-dimensional spatial location information specifically involves: taking the set of points to be determined and the classified water and land points in its neighborhood as input, and calculating the spatial density connectivity between the points to be determined and adjacent classified point clusters based on a density clustering algorithm; if the points to be determined and a certain classified point cluster satisfy the spatial density reachability condition, then the classification of the points to be determined is corrected to the category of that point cluster.

[0010] In a preferred embodiment, the construction of the multi-channel waveform feature fusion model and the extraction of combined features include: for the deep-water channel and shallow-water channel waveform data of the blue-green laser, calculating at least two features among waveform width, amplitude asymmetry, and backscattering slope respectively; and splicing and normalizing the features from the two channels to form a combined feature vector for describing the interaction between the laser and the water body and the seabed.

[0011] In a preferred embodiment, the re-discrimination of the point set adjacent to the land-water boundary region includes: inputting the combined feature vector into a pre-trained shallow water discriminator classifier, the classifier outputting the probability that each point to be judged belongs to extremely shallow water; if a point is judged as land in the intermediate classification result, but its probability of belonging to extremely shallow water exceeds a preset probability threshold, and its spatial location is within the possible water boundary indicated by hydrological features, then its final category is corrected to water.

[0012] In a preferred embodiment, the shallow water discriminant classifier is trained by: collecting ALB multi-channel waveform data containing accurately labeled extremely shallow water samples; extracting the combined feature vector of the samples as input and the true class label of the samples as output; and training the shallow water discriminant classifier using a machine learning algorithm.

[0013] In a preferred embodiment, after acquiring the full waveform data and before performing the first clustering, the method further includes: performing joint denoising processing on the full waveform data, including machine learning background noise modeling and removal based on non-echo signal segments, and frequency domain low-pass filtering optimized based on the maximum signal-to-noise ratio criterion.

[0014] In a preferred embodiment, after outputting the final classification result, the method further includes: fusing the classification result with the three-dimensional point cloud data generated by the ALB system, assigning each point cloud an accurate land and water attribute label; and generating a geographic information product for coastal zone resource surveys, change monitoring, or ecological protection applications based on the point cloud with attribute labels.

[0015] The technical effects and advantages of the airborne lidar depth measurement waveform classification method for land and sea in this invention are as follows:

[0016] This invention employs a three-tiered progressive strategy—waveform amplitude initial screening, spatial relationship correction, and multi-feature fusion for refined judgment—to progressively address classification uncertainties at different levels. This effectively reduces misclassification caused by noise, special features, and ambiguous boundary areas, significantly improving overall classification accuracy. The introduction of the concept of undetermined point sets and spatial density clustering enables the method to properly handle classification ambiguities, enhancing the algorithm's adaptability to data noise and complex scenarios. The multi-channel feature fusion model is specifically optimized for extremely shallow water areas, addressing the weaknesses of existing technologies and effectively resolving systemic misclassification problems in this region. Machine learning methods are fully utilized for background noise modeling and shallow water area discrimination, reducing manual intervention and reliance on empirical thresholds, thus improving the automation level and intelligent decision-making capabilities of the processing workflow. The final classification results can be directly fused with 3D point clouds to generate high-quality integrated land and sea geographic information products, providing a more reliable data foundation for applications such as coastal zone mapping, resource surveys, and environmental monitoring and protection, and promoting the in-depth application of ALB technology. Attached Figure Description

[0017] Figure 1 The above is a flowchart of an airborne lidar method for classifying sea and land waveforms for depth measurement, provided in an embodiment of the present invention.

[0018] Figure 2 A schematic diagram illustrating the initial clustering principle of an airborne lidar depth measurement sea and land waveform classification method based on infrared waveform amplitude, provided in an embodiment of the present invention.

[0019] Figure 3 This is a schematic diagram illustrating the spatial density clustering correction of a sea-land waveform classification method for airborne lidar depth measurement, provided in an embodiment of the present invention.

[0020] Figure 4 This invention provides a module diagram of an airborne lidar depth measurement sea-land waveform classification system. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0022] Example 1, Figure 1 , Figure 2 and Figure 3 This invention presents a method for classifying sea and land waveforms using airborne lidar (ALB) bathymetry, comprising: acquiring full waveform data of the target area from an airborne lidar bathymetry (ALB) system, wherein the full waveform data includes at least infrared laser waveform data and deep-water and shallow-water channel waveform data of blue-green lasers; performing initial clustering on the acquired laser footprints based on the amplitude characteristics of the infrared laser waveform data to obtain an initial classification result containing a water point set, a land point set, and a point set to be determined; fusing the three-dimensional spatial location information of the laser footprints to perform spatial density clustering analysis on the point set to be determined, classifying it into either the water point set or the land point set, generating an intermediate classification result; constructing a multi-channel waveform feature fusion model to extract the combined features of the deep-water and shallow-water channel waveform data of blue-green lasers, re-discriminating the point sets in the intermediate classification result that are adjacent to the water-land boundary region, identifying and correcting misclassified points in extremely shallow water areas; and outputting the final high-precision sea and land waveform classification result.

[0023] In this embodiment, the first clustering is performed based on the amplitude characteristics of the infrared laser waveform data. Specifically, this includes: calculating the peak amplitude of the infrared laser echo waveform corresponding to each laser foot point; using an unsupervised clustering algorithm to cluster the peak amplitude, assigning points with amplitudes below a first threshold to the water point set, points with amplitudes above a second threshold to the land point set, and points with amplitudes between the first and second thresholds to the undetermined point set.

[0024] It should be noted that, based on the physical differences between infrared lasers and water and land media, unsupervised clustering is used to achieve rapid separation of water bodies, land and fuzzy points, providing a foundation for subsequent accurate correction.

[0025] From a physical perspective, infrared lasers (typically with a wavelength of 1064 nm) have an extremely high attenuation coefficient in water, making them almost unable to penetrate the water surface. They are only weakly reflected or completely absorbed by the water surface, resulting in a very small peak amplitude of the infrared echo waveform corresponding to water bodies. In contrast, land features (such as soil, vegetation, and buildings) strongly reflect infrared light, resulting in a significantly higher peak amplitude of the echo waveform. This inherent physical characteristic makes infrared waveform amplitude a core initial screening indicator for distinguishing between land and sea, and is also the fundamental basis for selecting this feature in claim 2.

[0026] In practical implementation, the first step is to calculate the peak amplitude of the infrared echo at each laser footpoint. Based on the jointly denoised data (to avoid noise interfering with amplitude accuracy), the individual infrared echo waveform is smoothed (e.g., using a 5-point moving average). Then, the maximum peak point in the waveform is located, and the signal intensity corresponding to this point is the peak amplitude (the calculation formula can be expressed as A=Max(S(t))-B, where S(t) is the denoised waveform time series, and B is the background noise baseline value). The second step is to automatically group the peak amplitudes of all footpoints using an unsupervised clustering algorithm (preferably K-means clustering, with K set to 3). The algorithm divides the data into three clusters based on the statistical distribution characteristics of the amplitude, corresponding to the low amplitude cluster (dominated by water), the high amplitude cluster (dominated by land), and the medium amplitude cluster (fuzzy region). The third step determines two thresholds based on the clustering results: the first threshold is set as the cluster center of low-amplitude clusters minus 1 standard deviation (ensuring that the vast majority of real water points are included), and the second threshold is set as the cluster center of high-amplitude clusters plus 1 standard deviation (ensuring that the vast majority of real land points are included). Points with amplitudes lower than the first threshold are directly assigned to the water point set, and those with amplitudes higher than the second threshold are assigned to the land point set. Points with medium amplitudes between the two thresholds (such as low-reflectivity tidal flats, water surfaces with scum, and abnormal points after noise interference) are assigned to the undetermined point set. This design avoids the defect of fixed threshold division in existing technologies that is either black or white, and reserves optimization space for subsequent spatial correction and multi-channel fine judgment by retaining fuzzy points.

[0027] Its core function is to quickly and coarsely classify and focus on key points: by combining physical features with unsupervised learning, it can efficiently separate 80%-90% of the clear land and sea points, greatly reducing the amount of data to be processed in the later stages; at the same time, by setting up a set of undetermined points, the classification difficulties are concentrated on a small number of fuzzy points, laying the foundation for the precise application of subsequent two-level optimization strategies, which not only ensures processing efficiency but also takes into account the flexibility of classification.

[0028] In this embodiment, spatial density clustering analysis is performed by integrating three-dimensional spatial location information. Specifically, the spatial density connectivity between the undetermined point set and the classified water and land points in its neighborhood is calculated based on the density clustering algorithm. If the undetermined point and a certain classified point cluster meet the spatial density reachability condition, the classification of the undetermined point is corrected to the category of that point cluster.

[0029] It should be noted that by fusing the three-dimensional spatial location information of laser footprints and using density clustering algorithms to solve the fuzzy classification problem of the undetermined point set after the first clustering, the essence is to use the spatial continuity of geographical features to correct the risk of misclassification of isolated points.

[0030] The core logic of this step is to infer the unknown from the known: the input data not only includes the set of undetermined points obtained from the initial clustering, but also simultaneously incorporates water and land points that have been clearly classified in their neighborhoods—because in natural scenes, both land and sea features are continuously clustered (e.g., water bodies do not exist in isolation in the center of land, and land does not float scattered on water), and the class correlation between adjacent points is extremely strong. Using a density clustering algorithm (preferably DBSCAN), the spatial density connectivity between each undetermined point and its adjacent classified point clusters is calculated. The core criterion is density reachability: that is, if within a preset neighborhood of an undetermined point (the neighborhood radius is set based on the laser point density), there exists a sufficient number (minimum point threshold) of classified points of the same class, and these points form a continuous density-connected cluster with the undetermined point, then the undetermined point and the point cluster belong to the same land feature type, and its class is corrected to the corresponding point cluster's class.

[0031] This step precisely addresses the shortcomings of existing technologies that isolate individual nodes: for mid-amplitude ambiguous points caused by noise or special features (such as scum in water or scattered water accumulation on land) during the initial clustering, a single waveform feature alone cannot accurately determine their location. Spatial density clustering, however, effectively distinguishes between truly isolated features and noise-interference points by mining the spatial relationships between adjacent nodes. For example, although scattered scum in water may have an infrared amplitude close to that of land, the vast majority of its neighborhood consists of water points, and its density connectivity points towards water, so it will be corrected to be classified as water. Similarly, small patches of water on land, whose neighborhood is predominantly composed of land points, will be classified as land. Ultimately, through this step, the entire set of points to be determined is clearly classified, generating intermediate classification results with stronger category continuity and a lower misclassification rate, thus clearing the fundamental obstacles for subsequent fine-grained optimization in extremely shallow water areas.

[0032] In this embodiment, a multi-channel waveform feature fusion model is constructed to extract combined features, including: for the deep-water channel and shallow-water channel waveform data of blue-green laser, calculating at least two features among waveform width, amplitude asymmetry, and backscattering slope respectively; and splicing and normalizing the features from the two channels to form a combined feature vector for describing the interaction between the laser and the water body and the seabed.

[0033] It should be noted that by extracting and integrating the key physical features of the blue-green laser dual channels (deep water + shallow water), a feature vector that can accurately distinguish between extremely shallow water and land is constructed, providing a highly discriminative input basis for the subsequent shallow water discrimination classifier, thus solving the problem of insufficient characterization of extremely shallow water features in existing technologies.

[0034] Its core design logic stems from the propagation and reflection characteristics of blue-green lasers at different water depths: blue-green lasers (typically with a wavelength of 532nm) can penetrate water, and their echo waveforms carry interactive information from the water surface, the water body, and the seabed. Deep-water channels are optimized for deeper water areas, with waveforms primarily reflecting from the seabed, resulting in clearer characteristics. Shallow-water channels are optimized for extremely shallow waters of 0.5-3m; although affected by the superposition of reflections from the water surface and seabed, the waveform signal is weaker and more complex, but still contains distinguishable physical characteristics. Therefore, relying solely on the characteristics of a single channel is insufficient to fully characterize the waveform differences in extremely shallow waters; it is necessary to integrate the characteristics of both channels to achieve complementarity.

[0035] In practice, at least two key physical characteristics are extracted from the waveform data of the deep-water and shallow-water channels of blue-green lasers: waveform width (half-width at half maximum) reflects the range of interaction between the laser pulse and the water body and seabed. In extremely shallow waters, the waveform width is usually greater than that on land due to the superposition of reflections from the water surface and seabed; amplitude asymmetry reflects the strength distribution of the echo signal. Echoes in extremely shallow waters contain double reflections from the water surface and seabed, and tend to exhibit asymmetric characteristics such as a steep leading edge and a gentle trailing edge or vice versa; backscattering slope reflects the laser energy decay rate. The difference in scattering characteristics between water and land will lead to significant differences in slope.

[0036] Subsequently, the features extracted from the two channels are concatenated and normalized: concatenation integrates the complementary information of the two channels, avoiding the limitations of single-channel features; normalization (such as Z-score normalization) eliminates dimensional differences between different features (e.g., amplitude asymmetry is dimensionless, waveform width is time / distance dimension), ensuring that each feature has a balanced weight in the subsequent classifier. The resulting combined feature vector can comprehensively and accurately describe the differences in the interaction between laser and water (especially very shallow water) and land, providing core feature support for the subsequent identification of misclassified points in very shallow water, and significantly improving the recognition accuracy of the classifier.

[0037] In this embodiment, the point set adjacent to the land-water boundary area is re-discriminated, including: inputting the combined feature vector into a pre-trained shallow water discriminator classifier, and the classifier outputting the probability that each point to be judged belongs to the extremely shallow water area; if a point is judged as land in the intermediate classification result, but its probability of belonging to the extremely shallow water area exceeds the preset probability threshold, and its spatial location is within the possible water area boundary indicated by the hydrological features, then its final category is corrected to water body.

[0038] It should be noted that by using both classifier probability output and spatial location constraints for dual verification, misclassification points in extremely shallow waters can be accurately identified and corrected, thus overcoming the pain point of existing technologies in dealing with the ambiguity of the boundaries of extremely shallow waters and their tendency to be misclassified as land.

[0039] Its core design logic combines feature discrimination with spatial verification: relying solely on the classification probability of multi-channel combined features may result in a small number of low-reflectivity land features (such as black mudflats) being misclassified as extremely shallow water; relying solely on spatial location may lead to the misclassification of real land within the boundary as water. Therefore, a dual constraint is needed to achieve accurate correction and ensure the lowest possible misclassification rate.

[0040] In practice, the constructed combined feature vector is first input into a pre-trained shallow water discriminator classifier. The classifier will output the probability value of each point to be judged as belonging to extremely shallow water. This probability is a quantitative judgment based on the feature patterns of the samples, which can intuitively reflect the degree of matching between the point to be judged and the extremely shallow water samples.

[0041] The system then enters a dual-verification process: the first step is probability threshold verification. If the probability of a point being classified as being in extremely shallow water exceeds a preset threshold (e.g., 0.75, determined through cross-validation to balance precision and recall), it is initially classified as a potential point in extremely shallow water. The second step is spatial location verification. Combining hydrological features indicating possible water body boundaries (e.g., areas extended from coastlines on nautical charts, or water body ranges extracted from NDWI indices in remote sensing images), it is determined whether the point is located within the boundary. Only when the probability exceeds the threshold and the point is located within the boundary, and the point is classified as land in the intermediate classification results, is its final category corrected to water. This design avoids misclassification by single probability discrimination and excludes real land within the spatial boundary, ensuring the reliability of the correction results.

[0042] Ultimately, this step can accurately correct systematic misclassification points in extremely shallow waters, making the classification of land-water boundaries more consistent with the actual distribution of land features, and laying a core foundation for subsequent fusion of classification results with point clouds to generate high-precision geographic information products.

[0043] In this embodiment, the shallow water discriminant classifier is trained by collecting ALB multi-channel waveform data containing accurately labeled extremely shallow water samples; extracting the combined feature vectors of the samples as input and the true class labels of the samples as output; and training the shallow water discriminant classifier using a machine learning algorithm.

[0044] It should be noted that by using high-quality samples, feature-label mapping, and machine learning training, a highly robust shallow water identification model is constructed, providing a reliable tool for subsequent accurate identification.

[0045] The core logic of training is data-driven feature mapping learning: the classifier's discriminative ability depends on the representativeness of the samples and the accuracy of the annotations. Therefore, the first step is to prioritize collecting ALB multi-channel waveform data containing accurately labeled extremely shallow water samples. The samples need to cover different scenarios (such as extremely shallow waters with different sea areas, different tidal states, and different bottom types), while also including easily confused land samples (such as black mudflats and wet sandy areas) to ensure the diversity and balance of the sample set and avoid model overfitting. The annotations need to be based on measured data (such as depth gauge measurements and field surveys) to ensure the absolute accuracy of the labels for extremely shallow water samples. This is the basic premise for model training.

[0046] The subsequent steps revolve around constructing the input-output mapping: First, extract the combined feature vectors corresponding to the samples (i.e., the concatenated and normalized dual-channel features) as the model input, and use the true class labels of the samples (e.g., very shallow water = 1, non-very shallow water = 0) as the output to establish the correspondence between features and classes; then, select an appropriate machine learning algorithm (e.g., random forest, support vector machine, lightweight neural network, etc.) for training. During the training process, it is necessary to improve the generalization ability and discrimination accuracy of the model through data partitioning (training set, validation set) and hyperparameter optimization (e.g., grid search, cross-validation).

[0047] The core value of the entire training process lies in enabling the model to learn the unique feature patterns of extremely shallow waters: through learning from a large number of labeled samples, the model can automatically capture the subtle differences between extremely shallow waters and land and deep water in multi-channel waveform features, thereby having the ability to quantify the output discrimination probability, providing accurate and reliable probabilistic basis for re-verification, and ensuring the accuracy and robustness of the correction of misclassification points in extremely shallow waters.

[0048] In this embodiment, after acquiring the full waveform data and before performing the first clustering, the method further includes: performing joint denoising processing on the full waveform data, including machine learning background noise modeling and removal based on non-echo signal segments, and frequency domain low-pass filtering optimized based on the maximum signal-to-noise ratio criterion.

[0049] It should be noted that by using a joint denoising strategy of machine learning background noise modeling and frequency domain low-pass filtering, various noise interferences in the data are removed, providing high-quality and highly reliable clean waveform data for subsequent feature extraction (such as infrared waveform peak amplitude and blue-green dual-channel combined features) and classification steps, thus ensuring classification accuracy from the source.

[0050] Its core design logic is to address different types of noise in a targeted manner: ALB full waveform data is subject to multiple noise interferences during the acquisition process, including system electronic noise, atmospheric scattering noise, background light interference, etc. These noises will distort the true characteristics of the waveform (such as reducing the effective signal amplitude and falsely widening the waveform width), directly leading to deviations in subsequent clustering and discrimination. Single denoising methods are insufficient to cover all noise types. Therefore, a combined denoising approach is adopted to achieve complementarity. The first denoising method focuses on background noise. Pure noise samples are constructed through non-echo signal segments (i.e., the time period before laser emission and after reception when there is no effective echo). Machine learning algorithms (such as Gaussian mixture models and support vector machines) are used to accurately fit the distribution pattern of noise, thereby achieving adaptive modeling and accurate removal of background noise and avoiding the loss of effective signal caused by traditional fixed threshold denoising. The second denoising method focuses on high-frequency impulse noise. Through frequency domain analysis, the signal is decomposed into low-frequency effective components and high-frequency noise components. The cutoff frequency of the low-pass filter is optimized based on the maximum signal-to-noise ratio criterion (rather than relying on empirical values). This ensures that while filtering out high-frequency noise, the key physical characteristics of the waveform (such as peak shape and slope changes) are preserved to the maximum extent, achieving a balance between denoising and signal fidelity.

[0051] The core value of joint denoising lies in improving data quality from the source: clean full-waveform data is the foundation for all subsequent classification steps. If noise is not effectively removed, the calculation of the peak amplitude of the infrared waveform will be biased, leading to inaccurate threshold division in the initial clustering. Features such as waveform width and amplitude asymmetry in the blue and green channels will also be distorted by noise, affecting the discrimination effect of the multi-channel feature fusion model. Through this step, the signal-to-noise ratio of the waveform data can be significantly improved, ensuring that the various features extracted subsequently can truly reflect the interaction law between the laser and the water and land media, laying a solid data foundation for the high-precision implementation of the entire three-level progressive classification strategy.

[0052] In this embodiment, after outputting the final classification result, the method further includes: fusing the classification result with the three-dimensional point cloud data generated by the ALB system, and assigning each point cloud an accurate land and water attribute label; and generating a geographic information product for coastal zone resource surveys, change monitoring, or ecological protection applications based on the point cloud with attribute labels.

[0053] It should be noted that the high-precision land and sea classification results obtained in the previous steps are deeply integrated with the core output of the ALB system, the 3D point cloud data, and transformed into geographic information products with practical application value, completing the closed loop from data classification to industry empowerment, which reflects the practical value of the invention.

[0054] Its core logic lies in the binding of attribute tags with spatial data and the precise adaptation to application scenarios: the 3D point cloud data collected by the ALB system only contains spatial location information (X, Y, Z), lacking clear land and sea attribute identifiers, and cannot be directly used for professional applications related to coastal zones; while this invention, through progressive classification, has assigned precise water or land attribute tags to each laser footprint. Therefore, the first step of the fusion process is essentially to establish a one-to-one correspondence between laser footprint ID, 3D point cloud coordinates, and land and sea attribute tags, embedding attribute information as an extended field into the point cloud data (such as mainstream point cloud formats like LAS and LAZ), so that the point cloud not only has spatial location characteristics but also has clear land cover type attributes, becoming semantically meaningful 3D spatial data.

[0055] Subsequent geographic information product generation is based on attribute-labeled point clouds, with targeted processing tailored to the specific needs of each application scenario. For example, for coastal resource surveys, an integrated land-sea digital elevation model (DEM) can be generated, clearly presenting the continuous distribution of land topography and shallow water depth, providing basic data for port planning and aquaculture area layout. For change monitoring, by comparing labeled point clouds from different periods, changes in shoreline advance and retreat, water area increase and decrease, and land development can be accurately identified. For ecological protection, boundary and topographic data of key ecological areas such as extremely shallow waters and intertidal zones can be extracted, providing support for wetland protection and marine habitat monitoring. The core advantage of these products lies in high-precision attributes and high-reliability spatial data. Because the preceding classification process has effectively solved problems such as misclassification of extremely shallow waters and blurred boundaries, the generated geographic information products have higher accuracy in key indicators such as land-sea boundary delineation and water depth inversion (e.g., shoreline extraction accuracy ±1m, extremely shallow water depth accuracy ±0.2m), far superior to products generated by traditional methods.

[0056] The core value lies in unleashing the application potential of the classification results: the previous steps solved the technical problem of how to classify accurately, while this step answers the practical need of how to use the classification results, transforming technological innovation into practical results that can directly serve fields such as marine surveying, resource management, and ecological protection, so that the technological value of the entire invention can be realized as industry application value, while also expanding the application boundaries of ALB technology in complex coastal scenarios.

[0057] Example 2, Figure 4 This invention presents an airborne lidar depth sounding waveform classification system for land and sea, comprising a data acquisition and preprocessing module, an initial classification module, a spatial correction module, a multi-channel optimization module, and a result output and application module.

[0058] The data acquisition and preprocessing module is used to acquire and preprocess ALB full waveform data;

[0059] The initial clustering module is used to perform the first clustering based on the infrared waveform amplitude features and generate the initial classification results.

[0060] The spatial correction module is used to fuse spatial location information, correct the undetermined points through spatial density clustering, and generate intermediate classification results;

[0061] The multi-channel optimization module is used to build and apply a multi-channel waveform feature fusion model to identify and correct misclassification points in extremely shallow water areas and generate the final classification result.

[0062] The results output and application module is used to output and apply the final classification results.

[0063] In this example, the system is integrated into the ALB data post-processing platform as a software module, or exists as an independent software service; the data acquisition and preprocessing module also includes a waveform denoising submodule, and the multi-channel optimization module includes a shallow water discrimination model loading and inference submodule.

[0064] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0065] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0066] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0067] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0068] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of protection of the claims. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for classifying sea and land waveforms in airborne lidar depth measurement, characterized in that, include: Acquire full waveform data of airborne lidar depth sounding (ALB) for the target area, wherein the full waveform data includes at least infrared laser waveform data and deep-water and shallow-water channel waveform data of blue-green laser. Based on the amplitude characteristics of the infrared laser waveform data, the collected laser footprints are first clustered to obtain an initial classification result that includes water body point sets, land point sets, and undetermined point sets. By integrating the three-dimensional spatial location information of the laser footprints, spatial density clustering analysis is performed on the set of points to be determined, and they are classified into water point set or land point set to generate intermediate classification results. A multi-channel waveform feature fusion model is constructed to extract the combined features of the blue-green laser deep-water and shallow-water channel waveform data. The point set of the intermediate classification result adjacent to the water-land boundary region is re-discriminated to identify and correct the misclassified points of extremely shallow water. Output the final high-precision sea-land waveform classification results; The re-discrimination of point sets in adjacent land-water boundary areas includes: The combined feature vector is input into a pre-trained shallow water discriminator classifier, which outputs the probability that each point to be judged belongs to extremely shallow water. If a point is classified as land in the intermediate classification results, but its probability of belonging to extremely shallow water exceeds a preset probability threshold, and its spatial location is within the possible water boundary indicated by hydrological features, then its final category is corrected to water.

2. The method for classifying sea and land waveforms using airborne lidar depth measurement according to claim 1, characterized in that, The initial clustering based on the amplitude features of infrared laser waveform data specifically includes: Calculate the peak amplitude of the infrared laser echo waveform corresponding to each laser foot point; An unsupervised clustering algorithm is used to cluster the peak amplitude. Points with amplitudes below a first threshold are assigned to the water point set, points with amplitudes above a second threshold are assigned to the land point set, and points with amplitudes between the first and second thresholds are assigned to the undetermined point set.

3. The method for classifying sea and land waveforms using airborne lidar depth measurement according to claim 2, characterized in that, The spatial density clustering analysis performed by fusing three-dimensional spatial location information is specifically as follows: Using the set of points to be determined and the classified water and land points in its neighborhood as input, the spatial density connectivity between the points to be determined and the adjacent classified point clusters is calculated based on the density clustering algorithm. If the undetermined point and a certain already classified point cluster satisfy the spatial density reachability condition, then the classification of the undetermined point is corrected to the category of that point cluster.

4. The method for classifying sea and land waveforms using airborne lidar depth sounding according to claim 3, characterized in that, The construction of the multi-channel waveform feature fusion model and the extraction of combined features include: For the deep-water and shallow-water channel waveform data of the blue-green laser, calculate at least two of the following characteristics: waveform width, amplitude asymmetry, and backscattering slope. The features from the two channels are spliced ​​and normalized to form a combined feature vector that describes the interaction between the laser and the water body and the seabed.

5. The method for classifying sea and land waveforms using airborne lidar depth sounding according to claim 4, characterized in that, The shallow water discriminant classifier is trained in the following way: Collect ALB multichannel waveform data containing accurately labeled samples from extremely shallow water areas; Extract the combined feature vector of the sample as input and the true class label of the sample as output; The shallow water discriminant classifier was obtained by training a machine learning algorithm.

6. The method for classifying sea and land waveforms using airborne lidar depth sounding according to claim 5, characterized in that, After acquiring the full waveform data and before performing the first clustering, the process also includes: The full waveform data is subjected to joint denoising processing, including machine learning background noise modeling and removal based on non-echo signal segments, and frequency domain low-pass filtering optimized based on the maximum signal-to-noise ratio criterion.

7. The method for classifying sea and land waveforms using airborne lidar depth sounding according to claim 6, characterized in that, After outputting the final classification result, it also includes: The classification results are fused with the 3D point cloud data generated by the ALB system to assign precise water and land attribute labels to each point cloud. Geographic information products are generated based on point clouds with attribute labels for applications such as coastal zone resource surveys, change monitoring, or ecological protection.

8. An airborne lidar depth sounding waveform classification system for land and sea, used to implement the method described in any one of claims 1-7, characterized in that, include: The data acquisition and preprocessing module is used to acquire and preprocess ALB full waveform data; The initial clustering module is used to perform the first clustering based on the infrared waveform amplitude features and generate the initial classification results. The spatial correction module is used to fuse spatial location information, correct the undetermined points through spatial density clustering, and generate intermediate classification results; The multi-channel optimization module is used to build and apply a multi-channel waveform feature fusion model to identify and correct misclassification points in extremely shallow waters and generate the final classification result. The results output and application module is used to output and apply the final classification results.

9. A sea-land waveform classification system for airborne lidar depth sounding according to claim 8, characterized in that, The data acquisition and preprocessing module also includes a waveform denoising submodule, and the multi-channel optimization module includes a shallow water discrimination model loading and inference submodule.

Images