CN117541799B - Large-scale point cloud semantic segmentation method based on online random forest model multiplexing - Google Patents
Large-scale point cloud semantic segmentation method based on online random forest model multiplexing Download PDFInfo
- Publication number
- CN117541799B CN117541799B CN202410028448.8A CN202410028448A CN117541799B CN 117541799 B CN117541799 B CN 117541799B CN 202410028448 A CN202410028448 A CN 202410028448A CN 117541799 B CN117541799 B CN 117541799B
- Authority
- CN
- China
- Prior art keywords
- point cloud
- random forest
- forest model
- data
- online random
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000007637 random forest analysis Methods 0.000 title claims abstract description 156
- 230000011218 segmentation Effects 0.000 title claims abstract description 123
- 238000000034 method Methods 0.000 title claims abstract description 71
- 238000009826 distribution Methods 0.000 claims abstract description 137
- 238000007781 pre-processing Methods 0.000 claims abstract description 7
- 238000012549 training Methods 0.000 claims description 99
- 238000012937 correction Methods 0.000 claims description 30
- 230000003993 interaction Effects 0.000 claims description 19
- 238000013136 deep learning model Methods 0.000 claims description 18
- 238000000605 extraction Methods 0.000 claims description 15
- 238000000638 solvent extraction Methods 0.000 claims description 14
- 238000004422 calculation algorithm Methods 0.000 claims description 9
- 238000011156 evaluation Methods 0.000 claims description 7
- 238000004364 calculation method Methods 0.000 claims description 5
- 238000003066 decision tree Methods 0.000 claims description 3
- 239000003550 marker Substances 0.000 claims description 3
- 230000002708 enhancing effect Effects 0.000 claims description 2
- 238000005192 partition Methods 0.000 claims description 2
- 230000002452 interceptive effect Effects 0.000 claims 2
- 238000012545 processing Methods 0.000 abstract description 8
- 238000013461 design Methods 0.000 abstract description 6
- 238000003672 processing method Methods 0.000 abstract description 2
- 230000035800 maturation Effects 0.000 description 12
- 230000008569 process Effects 0.000 description 10
- 230000000694 effects Effects 0.000 description 9
- 238000010801 machine learning Methods 0.000 description 9
- 238000010586 diagram Methods 0.000 description 6
- 238000002372 labelling Methods 0.000 description 6
- 238000013135 deep learning Methods 0.000 description 4
- 238000012986 modification Methods 0.000 description 4
- 230000004048 modification Effects 0.000 description 4
- 230000000903 blocking effect Effects 0.000 description 3
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 230000008447 perception Effects 0.000 description 2
- 230000006978 adaptation Effects 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 239000003086 colorant Substances 0.000 description 1
- 238000004590 computer program Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000001915 proofreading effect Effects 0.000 description 1
- 230000003252 repetitive effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000003860 storage Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/24323—Tree-organised classifiers
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
- G06N20/20—Ensemble learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
- G06V10/765—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects using rules for classification or partitioning the feature space
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Multimedia (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Software Systems (AREA)
- Medical Informatics (AREA)
- Computing Systems (AREA)
- Data Mining & Analysis (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Databases & Information Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Image Analysis (AREA)
Abstract
The application discloses a large-scale point cloud semantic segmentation method based on online random forest model multiplexing, which relates to the field of point cloud data processing and comprises the following steps: preprocessing point cloud data; extracting characteristics of point cloud block data, and acquiring a mature online random forest model based on the point cloud block data; and matching the characteristic distribution of the cloud data of the points to be predicted in a point cloud characteristic space distribution database, and predicting the cloud data of the points to be predicted by using a mature online random forest corresponding to the matching result. The method improves generalization capability and instantaneity, reduces computing resource requirements, provides a framework design with better compatibility, can be compatible and suitable for different kinds of point cloud data processing methods, and integrates results of a single machine with high efficiency.
Description
Technical Field
The application relates to the field of point cloud data processing, in particular to a large-scale point cloud semantic segmentation method based on online random forest model multiplexing.
Background
With the development of remote sensing technology, it becomes easier to obtain high-precision large-scale remote sensing urban point clouds. However, for remote sensing city market scenic spot clouds covering a range of hundreds of meters, including tens of millions of magnitude points, a great deal of manual labeling work is still required to quickly obtain accurate semantic segmentation results in engineering context. To reduce this tedious effort, researchers need to explore the potential of machine learning techniques in dealing with point cloud semantic segmentation.
Classical machine learning models such as support vector machines SVM, random forest RF, adaBoost, etc. can achieve fast and good training results on point cloud scenarios, but require manual design features to guarantee the classification capability of the model.
In recent years, deep learning has made remarkable progress in the field of point cloud semantic segmentation. For example PointNet, pointNet ++, poinconv can handle the entire point cloud, but the segmentation accuracy on large-scale point clouds is limited. RandLaNet can quickly and accurately perform large-scale semantic segmentation, but is easily affected by data acquisition noise. The SQN reduces labeling costs in a weakly supervised manner, but the model is sensitive to the area and scale of the training data. Although deep learning techniques demonstrate great capability in terms of complex feature extraction, a large amount of training data is still required to avoid overfitting problems in training. Moreover, because the urban scene contains complex regional characteristics and topographic information, the current methods cannot well solve the generalization problem, so that the pre-training models of the methods have a large number of segmentation errors when facing new point cloud data. Therefore, to assist in point cloud semantic segmentation, a large amount of time is still required to annotate new datasets to train a massive deep learning model.
Therefore, there is a need for an efficient and powerful method for semantic segmentation of point cloud data on a large scale. The method can be used for rapidly processing data sets with different distributions, so that efficient and accurate semantic segmentation is realized without a large amount of manual labeling. The method can remarkably improve the efficiency and reliability of the point cloud semantic segmentation and promote the development of smart city modeling and intelligent environment perception.
Disclosure of Invention
The application discloses a large-scale point cloud semantic segmentation method based on online random forest model multiplexing, and aims to solve the problems in the prior art.
In a first aspect, the present application provides a method for large-scale point cloud semantic segmentation based on online random forest model multiplexing, including: training a plurality of mature online random forest models in a blocking manner;
(1) Initial semantic segmentation is performed using a pre-trained deep learning model. Although the current deep learning method does not solve the generalization problem well, the method can provide complete annotation information rapidly, reduce the cost and lay a foundation for an online random forest model. (2) fitting a plurality of online random forest models in a blocking mode. And partitioning the result rule, and selecting a plurality of point clouds to extract the characteristics. The segmentation capability and the feature statistical accuracy of the model can be enhanced aiming at the features designed in the aspects of geometry, color, elevation and the like. The initial segmentation results are then fitted using an online random forest model. Through the block training, the training time of the online random forest model can be reduced, and the real-time interaction requirement is realized. In addition, the characteristic distribution is difficult to extract by the large-scale point cloud, and the local characteristic value and the characteristic distribution characteristic can be extracted more accurately by the blocking method. And (3) manually correcting and training the mature online random forest model. And obtaining a mature online random forest model and a correct semantic segmentation result by manually correcting error parameters in the rapid optimization model. Because of the characteristic of online learning, the model can learn the logic of manual interaction to improve error parameters and modify other similar error results, so that the number of manual interaction is greatly reduced, and the training maturation process of the model is accelerated. And (4) initializing a feature distribution database and a model multiplexing library. After the segmentation result is corrected manually, the histogram distribution of the multidimensional feature space is recorded, the distribution result is stored in a feature distribution database, and the corresponding mature online random forest model is stored in a model library. And establishing a considerable characteristic distribution database and an online random forest model library, and providing a proper model for the cloud blocks of the subsequent predicted points to realize multiplexing. This avoids repetitive training and reduces the cost of computation.
Semantic segmentation of the incremental multiplexing maturation model on point cloud blocks with consistent distribution;
(1) Multiplexing the initial semantic segmentation of the maturity model on uniformly distributed point cloud blocks: firstly, extracting multidimensional feature space histogram distribution of a predicted point cloud block, and calculating the Pasteur coefficient similarity of each distribution in a relative feature distribution database by using a model multiplexing strategy. Then, the corresponding model with the highest distribution of similarity is selected for initial semantic segmentation. Since the machine learning model follows independent co-distributed assumptions, the model performs better on a prediction set with a distribution close to the training set. The distribution consistency of the training set and the prediction set is ensured through the multiplexing strategy, so that the accurate prediction capability of the model can be improved. The prediction result is better than a pre-training deep learning model with insufficient generalization capability. Therefore, the errors needing manual correction can be reduced, and the training maturing process of the model is quickened.
(2) Updating a feature distribution library and a model library: and on the accurate initial segmentation result of the mature model, training a mature new online random forest model by utilizing manual correction, and updating a feature distribution database and a model library. Therefore, the range of the model multiplexing strategy can be enlarged, the prediction effect is improved, and the incremental learning effect of prediction while training is realized. With the increase of the size of the database, the best matching model can be found by arbitrarily distributed point cloud blocks, so that the aim of improving the universality is fulfilled. In addition, the initial segmentation result of the model multiplexing strategy is continuously improved along with the scale of the model library, the model approaches to the best deep learning model effect, and the required manual labeling and training waiting time is less.
Further, the preprocessing point cloud data includes: and performing regular segmentation on the point cloud data to obtain point cloud segmented data, and performing initial segmentation on the point cloud segmented data by using a pre-training model of a main stream point cloud segmentation method to obtain an initial segmentation result, wherein the initial segmentation result is used for providing labels for training of an on-line random forest with strong supervision.
Further, the feature extraction of the point cloud block data and the acquisition of the mature online random forest model based on the point cloud block data comprise the following steps:
using three dimensional characteristics of geometry, elevation and color to describe semantic differences, extracting the multi-dimensional characteristics of the point cloud block data and recording the multi-dimensional characteristics to a point cloud characteristic space distribution database;
and (3) taking the artificial correction data based on the semantic segmentation result of the online random forest model as a training set, iterating the online random forest model for training, obtaining a mature online random forest model meeting the prediction accuracy threshold of the point cloud segmentation data, and recording the mature online random forest model into a mature online random forest model library.
Further, for the cloud data of the to-be-predicted point, matching the characteristic distribution in the point cloud characteristic space distribution database, and predicting the cloud data of the to-be-predicted point by using the mature online random forest corresponding to the matching result, including:
matching a feature distribution by using a feature distribution similarity evaluation algorithm, wherein the feature distribution is most similar to the feature distribution of the cloud data of the points to be predicted in a point cloud feature space distribution database;
corresponding to the point cloud multidimensional feature space distribution to be predicted and the mature online random forest model;
judging whether the prediction accuracy of the current point cloud block data obtained by the corresponding mature online random forest model meets a threshold value or not;
and directly outputting a predicted result when the use requirement is met, training a new online random forest model on the predicted result when the use requirement is not met, finishing correction of an error area through manual interaction, accelerating training maturation of the online random forest, and acquiring a semantic segmentation result and a mature online random forest model meeting a threshold value. And adding the distribution and maturation model of the cloud blocks of the predicted points into a distribution database and a maturation model library.
Further, the matching of the feature distribution by using the point feature distribution similarity evaluation algorithm is a feature distribution which is most similar to the feature distribution of the cloud data of the points to be predicted in the point cloud feature space distribution database, and the feature distribution comprises the calculation based on the point cloud multidimensional feature space distribution, and specifically comprises the following steps:
analyzing the ratio of the semantics of the point cloud block data to be predicted to the total point cloud block data to be processed, extracting features aiming at different semantics of the point cloud block data to be predicted, acquiring multi-dimensional features of the point cloud block data to be predicted, and analyzing the feature value distribution condition of the point cloud block data to be predicted.
Further, the artificial correction data based on the semantic segmentation result of the online random forest model is used as a training set, the online random forest model is iterated for training, a mature online random forest model meeting the prediction accuracy threshold of the point cloud segmentation data is obtained, and the mature online random forest model is recorded in a mature online random forest model library, and the method comprises the following steps:
and training a segmentation prediction result of the point cloud semantic segmentation model for feature extraction by adopting an online random forest model, acquiring a random forest model semantic segmentation result in training, generating the manual correction data by adopting an error region of the manual marker segmentation result, inputting the manual correction data as a training set for updating branch nodes of different decision trees in the online random forest, and iteratively training to output a mature online random forest model meeting the point cloud segmentation data prediction accuracy.
Further, the multi-dimensional feature spatial distribution similarity includes a pasteurized distance metric histogram distribution similarity.
Further, the method specifically comprises the following steps: and obtaining a prediction result for feature extraction of the corresponding point cloud partitioning data by using a pre-training deep learning model.
Further, the method specifically comprises the following steps: and obtaining a prediction result for feature extraction of the corresponding point cloud partitioning data by using the pre-training point cloud semantic partitioning model.
In a second aspect, the present invention further provides a readable storage medium, on which a computer program is stored, which when executed by a processor, implements the large-scale point cloud semantic segmentation method based on online random forest model multiplexing described in any one of the above first aspects.
The beneficial effects of this application include:
the method and the device improve instantaneity and reduce the demand of computing resources: by using an online learning thought and a random forest machine learning model, the requirement on the computing resources of a single computer can be remarkably reduced through block parallel processing, feedback can be provided under the real-time condition, and the cooperation of multiple persons is allowed to be completed;
the application improves generalization capability: according to the method, through combining the machine learning model and manual interaction, point cloud data of different acquisition modes and different ground characteristics can be processed better, and an excellent segmentation effect is achieved;
the framework design with better compatibility is provided, the framework design can be compatible and suitable for different kinds of point cloud data processing methods, and the results of a single machine are integrated with high efficiency;
the model multiplexing framework provided by the application can learn the logic contained in the manual interaction, learn and imitate the manual interaction through a machine learning model, and can achieve the effect of the manual interaction after full training, so that the workload of manual correction is greatly reduced, and the labor cost is reduced.
Drawings
The accompanying drawings, which are included to provide a further understanding of embodiments of the present application and are incorporated in and constitute a part of this application, illustrate embodiments of the present application and together with the description serve to explain the principle of the present application. In the drawings:
FIG. 1 is a schematic diagram of a training stage for obtaining a plurality of maturity models and point cloud distribution characteristics on-line based on partitioning in an exemplary embodiment of the present application.
Fig. 2 is a schematic diagram of a prediction stage of semantic segmentation on distributed consistency point cloud blocks based on a multiplexing maturation model according to an exemplary embodiment of the present application.
Fig. 3 is a segmentation flowchart of the large-scale point cloud semantic segmentation method based on online random forest model multiplexing in the present application.
Detailed Description
Reference will now be made in detail to exemplary embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, the same numbers in different drawings refer to the same or similar elements, unless otherwise indicated. The implementations described in the following exemplary examples are not representative of all implementations consistent with the present application. Rather, they are merely examples of apparatus and methods consistent with some aspects of the present application as detailed in the accompanying claims.
The large-scale point cloud semantic segmentation method based on online random forest model multiplexing requires a large amount of marked complete data, and although collecting the large-scale point cloud data is no longer a tedious process, manually generating point-level labels on a large-scale data set requires high cost. In addition, the current large-scale point cloud segmentation method only trains aiming at specific scenes, model generalization is not ideal, segmentation accuracy depends on a training data set, and a prediction result cannot meet the requirement of being applied to city modeling and also depends on manual proofreading. In summary, research on point cloud segmentation has been advanced, but there are still some problems to be solved:
(1) Time overhead for processing large-scale data: the point cloud model of the smart city scene is huge in scale and reaches billions of point clouds. Processing these large-scale data requires a significant amount of computing resources and time, which greatly limits the application of existing methods. For the most advanced deep learning model, the requirement on the training set scale is larger than that of the machine learning model, and the time cost is further increased.
(2) Prediction accuracy of processing large-scale data: even though the most advanced deep learning method at present prepares a training set with huge scale by high cost manually, a good prediction result is obtained in other areas, the complete correctness of the result still cannot be ensured, and other less error data needs to be manually checked and processed, so that the labor cost is further increased.
(3) Generalization ability: while deep learning models may perform well on certain training data, they may perform poorly on new, unseen data. This is particularly important in the context of smart cities because of the complexity and diversity of urban environments, which results in a model that has been trained with significant manpower that does not achieve satisfactory results in other cities, and can only be retrained to a huge model. In addition, the geographic features of different regions of a large city are also obviously different, for example, the park of a people stream dense region and a suburban region in the city center are greatly different in road broad degree, tree dense degree, building height and dense degree, so that the model needs to respectively manufacture training sets of corresponding scenes to ensure the prediction effect when considering the two scenes, and the labor cost required by model training is further increased.
The technical conception of the application:
a set of high-efficiency multi-person collaborative correction, self-adaptive learning artificial correction logic and large-scale point cloud semantic segmentation framework for reducing the labor cost is designed. Essentially, a trained dynamic growth model library and a corresponding dynamic growth distribution database for recording the distribution condition of the point cloud block in the multidimensional characteristic space are maintained. And the model library is in one-to-one correspondence with the records in the database.
The specific application scene of the application is to extract scenes in complex urban areas and surrounding hilly and cultivated areas.
The large-scale point cloud semantic segmentation method based on the online random forest model multiplexing aims at solving the technical problems in the prior art.
The following describes the technical solutions of the present application and how the technical solutions of the present application solve the above technical problems in detail with specific embodiments. The following embodiments may be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments. Embodiments of the present application will be described below with reference to the accompanying drawings.
The application provides a large-scale point cloud semantic segmentation method based on online random forest model multiplexing, which comprises the following steps:
preprocessing point cloud data; and performing initial semantic segmentation on the new data set by using a pre-training deep learning model, and providing labeling information for the subsequent online random forest.
Extracting characteristics of point cloud block data, and acquiring a mature online random forest model based on the point cloud block data; and partitioning the new data set, training an online random forest model on each block, and fitting the result of the pre-training deep learning model. New features are designed to describe geometry, color, height Cheng Tedian of different semantic categories for enhancing learning ability of the online random forest model. And fitting an initial semantic segmentation result of the corresponding point cloud block by using an online random forest model, and completing error correction by assisting with multiple times of manual interaction. Because of the characteristics of online learning, once each interaction, the online learning model learns the characteristics of the interaction, so that a random forest model is optimized, other non-interacted error areas are applied, the total interaction times are reduced, and a complete semantic segmentation result and a mature online random forest model with the current point cloud distribution characteristics are obtained by manually correcting one point cloud.
For cloud data of the points to be predicted, matching the characteristic distribution of the cloud data in a point cloud characteristic space distribution database, and predicting the cloud data of the points to be predicted by using a mature online random forest corresponding to the matching result: and as the number of the manually corrected point cloud blocks increases, designing the distribution condition of the histogram distribution statistical point cloud blocks in the multidimensional feature space. The new features of the design used as the feature description of the histogram can embody the distribution differences of different blocks. Providing a distribution similarity calculation formula to establish a distribution relation of the point cloud blocks and the prediction point cloud blocks which are manually corrected, and guaranteeing that a training set and a prediction set are consistent to multiplex a mature online random forest model for initial semantic segmentation, so that the semantic segmentation accuracy far higher than that of a pre-training deep learning model can be obtained; and continuing to manually correct on the result predicted by the online random forest model, training a new online random forest model, adding the new online random forest model into the multiplexing model, and expanding the multiplexing range of the model.
Further, the preprocessing point cloud data includes: and performing regular segmentation on the point cloud data to obtain point cloud segmented data, and performing initial segmentation on the point cloud segmented data by using a pre-training model of a main stream point cloud segmentation method to obtain an initial segmentation result, wherein the initial segmentation result is used for providing labels for training of an on-line random forest with strong supervision.
Further, the feature extraction of the point cloud block data and the acquisition of the mature online random forest model based on the point cloud block data comprise the following steps:
extracting multidimensional features of the point cloud partitioning data and recording the multidimensional features to a point cloud feature space distribution database;
and (3) taking the artificial correction data based on the semantic segmentation result of the online random forest model as a training set, iterating the online random forest model for training, obtaining a mature online random forest model meeting the prediction accuracy threshold of the point cloud segmentation data, and recording the mature online random forest model into a mature online random forest model library.
Further, for the cloud data of the to-be-predicted point, matching the characteristic distribution in the point cloud characteristic space distribution database, and predicting the cloud data of the to-be-predicted point by using the mature online random forest corresponding to the matching result, including:
matching a feature distribution by using a point feature distribution similarity evaluation algorithm, wherein the feature distribution is most similar to the feature distribution of the cloud data of the points to be predicted in a point cloud feature space distribution database;
corresponding to the point cloud multidimensional feature space distribution to be predicted and the mature online random forest model;
judging whether the prediction accuracy of the current point cloud block data obtained by the corresponding mature online random forest model meets a threshold value or not;
and outputting a prediction result meeting the use requirement, or taking point cloud block data corresponding to the point cloud multi-dimensional characteristic space distribution to be predicted as a check set input, continuing training the mature online random forest model, obtaining a fitted mature online random forest model meeting the check set input prediction accuracy, and binding the fitted mature online random forest model with the point cloud multi-dimensional characteristic space distribution information to be predicted currently plus the point cloud multi-dimensional characteristic distribution information of the original mature online random forest model.
Further, the matching of the feature distribution by using the point feature distribution similarity evaluation algorithm is a feature distribution which is most similar to the feature distribution of the cloud data of the points to be predicted in the point cloud feature space distribution database, and the feature distribution comprises the calculation based on the point cloud multidimensional feature space distribution, and specifically comprises the following steps:
analyzing the ratio of the semantics of the point cloud block data to be predicted to the total point cloud block data to be processed, extracting features aiming at different semantics of the point cloud block data to be predicted, acquiring multi-dimensional features of the point cloud block data to be predicted, and analyzing the feature value distribution condition of the point cloud block data to be predicted.
Further, the artificial correction data based on the semantic segmentation result of the online random forest model is used as a training set, the online random forest model is iterated for training, a mature online random forest model meeting the prediction accuracy threshold of the point cloud segmentation data is obtained, and the mature online random forest model is recorded in a mature online random forest model library, and the method comprises the following steps:
and training a segmentation prediction result of the point cloud semantic segmentation model for feature extraction by adopting an online random forest model, acquiring a random forest model semantic segmentation result in training, generating the manual correction data by adopting an error region of the manual marker segmentation result, inputting the manual correction data as a training set for updating branch nodes of different decision trees in the online random forest, and iteratively training to output a mature online random forest model meeting the point cloud segmentation data prediction accuracy.
Further, the multi-dimensional feature spatial distribution similarity comprises histogram similarity and SSIM structure similarity coefficients.
Further, the method specifically comprises the following steps: and obtaining a prediction result for feature extraction of the corresponding point cloud partition data by using a weakly supervised deep learning model.
Further, the method specifically comprises the following steps: and obtaining a prediction result for feature extraction of the corresponding point cloud partitioning data by using the pre-training point cloud semantic partitioning model.
Fig. 3 is a sectional flowchart of a large-scale point cloud semantic segmentation method based on online random forest model multiplexing, as shown in fig. 3, the whole framework is divided into two stages according to the number of models contained in a model library: training phase and maturation phase, i.e. the prediction phase. Five modules are used in different stages, the modules are related to each other in flow, and the modules are introduced as follows:
module one: the input is the original large-scale remote sensing city point cloud. The output is a regular point cloud with initial segmentation semantic tags. The specific flow is to use a pre-trained deep learning model to make initial semantic segmentation on the original dataset. Because the distribution of the original data set and the distribution of the pre-training model have significant differences, the generalization capability of the pre-training model is insufficient to cope with the data sets with significantly different distributions, so that the initial semantic segmentation result has a great room for improvement. And then, outputting the initial segmentation result rule block to a second module.
And a second module: the input is a plurality of point clouds with semantic tag information. The output is a plurality of point cloud blocks from which the multidimensional feature values and semantic tag information are extracted. The specific flow is to execute the same feature extraction algorithm for each point cloud block. Because the machine learning method needs to manually design a feature training model, and the more effective features can accurately describe the distribution information of the point cloud block, 16 features in three aspects of geometry, elevation and color are designed. The geometric aspects use normal line differences (DoN) and covariance distribution features. DoN by comparing surface normals calculated over different scales, the roughness or smoothness of the surface can be measured. DoN has a better role in identifying the edges of different objects. For example, at the edges of the vehicle and the ground, normal differences of different dimensions are typically larger, resulting in a larger DoN value. Covariance features can explain the local geometry and structure of the point cloud by combining the three eigenvalues of the covariance matrix. Such as "curvature", "linearity", "flatness", "scattering" and "anisotropy". Because the semantics of the building and the road are regular and smooth, the semantics of the tree and the automobile are distributed irregularly, and the covariance features can play a good role in distinguishing. The elevation features use projection elevation, projection elevation difference, projection elevation variance, flatness based on projection Gao Chengquan weight, projection density. The projection Gao Chenghui projects the point cloud on the XOY plane, clusters the points overlapped in the Z-axis direction, and uniformly uses the maximum elevation value in all the points as the projection elevation characteristic result in each cluster. Since the road surface is flat but lower than the building, the projection elevation can show the differences between the ground and the building, between the ground and the tree, and between the ground and the car. And calculating the difference between the maximum elevation and the minimum elevation in the Z-axis direction projection overlapping point cluster as a characteristic by using the projection elevation difference. The projection elevation variance calculates the offset distance of each point relative to the neighborhood average projection elevation difference on the basis of the projection elevation difference. The flatness of the projection Gao Chengquan is based on the difference in projection elevation as a weight coefficient of the flatness, and the flatness features of different heights are distinguished. It is possible to effectively distinguish between a road which is also very flat and a building roof, and because the building is far above the road, its weight coefficient is greater, and thus the weight flatness is greater. The projection density calculates the number of points contained in the Z-axis projection overlapping point clusters, and as the building is generally higher than the trees, automobiles and roads, a great number of points can be overlapped in the clusters of the outer vertical surfaces of the building after the projection in the Z-axis direction, so that the building is distinguished from other semantics. The color features then use RGB, RGB entropy, LAB. RGB information is self-contained in the dataset and helps to distinguish trees from other semantics. RGB entropy counts the region with severe RGB variation, and can enhance the extraction of different semantic edges. The LAB color space is a color description mode, can accurately represent the perception of the human visual system on the colors, and reduces the influence caused by shadow problems in RGB information. After the multi-dimensional features are extracted by the second module, the feature result and the semantic tag information are combined and output to the third module to serve as fitting input of an online random forest, and the feature result is output to the fifth module to serve as input of feature distribution statistics.
And a third module: the input is an initial segmentation point cloud with feature results and semantic tags. The output is the accurate division point cloud block and the corresponding mature online random forest model which meet the requirements. The specific process is that the unsatisfactory area in the segmentation result is selected through manual interaction, and an online random forest training set is generated. The online learning is characterized in that partial parameters of the model are updated again according to the training set, and the model is optimized towards the direction closer to the latest training set. Therefore, the trained model not only corrects the error area corrected manually, but also modifies other similar error areas together, thereby greatly reducing the number of manual interaction. And repeating the flow of correction, training and prediction until the prediction result meets the threshold value, and completing the task of the current module. In fig. one, the threshold value is set to 95%. And outputting the training mature online random forest model to a model library of the fourth module.
And a fifth module: the input is a point cloud with characteristic results. The output is the histogram distribution of the point cloud block in the multidimensional feature space. The specific flow is that a feature corresponds to a histogram, each histogram is divided into a plurality of intervals, the points of the feature values contained in each interval are counted, and the distribution condition of the points contained in the interval is used as the counting result of the current feature dimension. In addition, the module can also calculate the similarity between the distributions, and the specific flow is to input two multidimensional feature space histogram distributions, measure the histogram distribution similarity of each dimension by using the Pasteur coefficient, and calculate the average similarity of all the dimensions.
And a fourth module: the input is the multidimensional feature space orthometric distribution of the mature online random forest model and the model training point cloud block. The specific flow is to record the corresponding relation between the mature online random forest model and the multidimensional feature space straight distribution, and multiplex the semantic segmentation of the mature model with the closest distribution according to the prediction set of different distributions. Because the machine learning model is built on the assumption that the training set and the prediction set are independently and uniformly distributed, the model needs to predict on the prediction set with consistent distribution to ensure more excellent prediction effect. Along with the increase of the number of the model libraries, consistent model multiplexing can be found for prediction sets of different distribution conditions, the multiplexing segmentation accuracy is relatively better, and the problem that the traditional model has insufficient generalization performance on other distribution data sets can be solved.
Fig. 1 is a schematic diagram of a training stage for obtaining a plurality of mature models and distribution characteristics of point clouds on line based on partitioning in an exemplary embodiment of the present application, where, as shown in fig. 1, the purpose of the training stage is to increase the number of models in a model library and increase the number of feature space distribution records of a database. The method comprises the following steps: preprocessing point cloud data, extracting characteristics of point cloud block data, and acquiring a mature online random forest model based on the point cloud block data; the method comprises the following steps: extracting multidimensional features of the point cloud partitioning data and recording the multidimensional features to a point cloud feature space distribution database;
the artificial correction data based on the semantic segmentation result of the online random forest model is used as a training set, the online random forest model is iterated for training, a mature online random forest model meeting the prediction accuracy threshold of the point cloud segmentation data is obtained, and the mature online random forest model is recorded into a mature online random forest model library
The method comprises the following steps: initial segmentation and regular segmentation are performed on the original data set by means of the pre-trained deep learning model in the first module. And then, extracting a multi-dimensional characteristic result by using a second module. And updating the online random forest model by using the manual interaction in the third module, and training out a mature online random forest model with the prediction accuracy reaching above a threshold value and a complete semantic segmentation result. The threshold is here set to 95% in fig. 1. And a fifth module extracts the histogram distribution condition of the training point cloud block in the multidimensional feature space. And finally, adding the trained online random forest model into a model library in the fourth module, and adding the distribution condition of the training point cloud blocks into a distribution database of the fourth module.
Fig. 2 is a schematic diagram of a prediction stage of semantic segmentation on distributed consistent point cloud blocks based on a multiplexing maturation model in an exemplary embodiment of the present application, as shown in fig. 2, where the purpose of the maturation stage is to use a maturation random forest model in a model library to assist in segmenting a subsequent prediction point cloud block, and select the most appropriate model as far as possible, so as to improve the prediction accuracy, thereby reducing the number of times of manual correction. The method comprises the following steps: and for cloud data of the points to be predicted, matching training point cloud blocks closest to the characteristic distribution in a multidimensional characteristic space distribution database, multiplexing mature online random forests corresponding to the matching results to predict the cloud data of the points to be predicted, and training a new mature model and statistical characteristic distribution on the prediction results to update a model library and a distribution library. The method comprises the following steps: matching a feature distribution by using a point feature distribution similarity evaluation algorithm, wherein the feature distribution is most similar to the feature distribution of the cloud data of the points to be predicted in a point cloud feature space distribution database; multiplexing mature random forest models trained by point cloud blocks with closest distribution for semantic segmentation; training a new online random forest model on the basis of the segmentation result and counting the characteristic distribution of cloud blocks of the predicted point; judging whether the accuracy rate meets a threshold value; the error area is not satisfied and is optimized by continuing to manually correct; and outputting a prediction result meeting the use requirement. The method comprises the following steps: fig. 2 is a schematic diagram of a prediction stage of semantic segmentation on distributed consistency point cloud blocks based on a multiplexing maturation model according to an exemplary embodiment of the present application. As shown in fig. 2, in the prediction stage schematic diagram, the cloud blocks of the to-be-predicted points are cut and multidimensional feature distribution is extracted, and a random forest model multiplexing algorithm is used for selecting the most proper mature random forest model prediction point cloud blocks, so that semantic segmentation of the model on a prediction set with consistent distribution is ensured. In addition, the effect of multiplexing the mature random forest model is better than that of the pre-training deep learning model, so that the area needing manual correction is obviously reduced, or the result meeting the prediction accuracy threshold is directly output, namely, the use requirement of a user is met. With the continuous addition of random forest models into the database, the number of alternative models is increased, models with higher prediction accuracy are increased, and as long as the number of models reaches a certain scale, the prediction accuracy of the selected random forest model can be finally ensured to directly reach above a threshold, in fig. 2, the threshold is 95%, and the workload of subsequent manual correction is avoided. And for the condition that the prediction accuracy is unsatisfactory, the correction can be carried out through the online random forest training process, and the correction process is as follows: and training a new online random forest model by taking a semantic segmentation result of the multiplexing model as input, training maturation through manual interaction, obtaining a mature online random forest model after fitting meeting the input prediction accuracy of a check set, binding the mature online random forest model with the current point cloud multidimensional characteristic space distribution information to be predicted, adding the model into a model library, and expanding the model multiplexing range. By maintaining a point cloud semantic segmentation model library, an online learning model supporting real-time interaction, namely an online random forest model and a deep learning model with huge training cost and excellent effect can be contained.
In actual operation, in the embodiment, in terms of solving the compatibility problem, the model library supports different kinds of semantic segmentation models, and can select corresponding methods according to different problem needs, for example, an online random forest model can be selected for high timeliness requirements, an OMCBoost model can be selected for high prediction accuracy, a PointNet model can be selected for a complex and numerous data set of scene details, a RandLa-Net model can be selected for a data set with more consistent scene characteristics but large scale and huge volume thereof, a weak supervision model can be selected for a data set with large labeling workload, and the like.
In the several embodiments provided in this application, it should be understood that the disclosed apparatus and method may be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative, and for example, the division of the modules is merely a logical function division, and there may be additional divisions when actually implemented, for example, multiple modules or components may be combined or integrated into another apparatus, or some features may be omitted or not performed.
The modules described as separate components may or may not be physically separate, and components shown as modules may or may not be physical modules, i.e., may be located in one place, or may be distributed over a plurality of network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
In addition, each functional module in each embodiment of the present application may be integrated into one processing module, or each module may exist alone physically, or two or more modules may be integrated into one module. The integrated modules may be implemented in hardware, or in hardware plus software functional modules, such as modules one through five.
It is to be understood that the present application is not limited to the precise arrangements and instrumentalities shown in the drawings, which have been described above, and that various modifications and changes may be effected without departing from the scope thereof. The scope of the application is limited only by the appended claims.
It will be apparent to those skilled in the art that embodiments of the present invention may be provided as methods or apparatus. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects.
It should also be noted that the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article or apparatus that comprises an element.
The foregoing is merely exemplary of the present application and is not intended to limit the present application. Various modifications and changes may be made to the present application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc. which are within the spirit and principles of the present application are intended to be included within the scope of the claims of the present application.
Other embodiments of the present application will be apparent to those skilled in the art from consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the application following, in general, the principles of the application and including such departures from the present disclosure as come within known or customary practice within the art to which the application pertains. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the application being indicated by the following claims.
It is to be understood that the present application is not limited to the precise arrangements and instrumentalities shown in the drawings, which have been described above, and that various modifications and changes may be effected without departing from the scope thereof. The scope of the application is limited only by the appended claims.
Claims (9)
1. The large-scale point cloud semantic segmentation method based on the online random forest model multiplexing is characterized by comprising the following steps of:
preprocessing point cloud data:
performing initial semantic segmentation on the new data set using a pre-trained deep learning model;
extracting characteristics of point cloud block data, and acquiring a mature online random forest model based on the point cloud block data:
dividing a new data set, training an online random forest model on each data set, fitting the result of a pre-training deep learning model, designing new features to describe the characteristics of different semantic categories, fitting the initial semantic segmentation result of corresponding point cloud blocks by using the online random forest model, completing error correction by assisting in multiple times of manual interaction, learning the interactive characteristics by using the online learning model once each interaction, optimizing the random forest model, applying other non-interactive error regions, correcting one point cloud block every time manually to obtain a complete semantic segmentation result, and learning a mature online random forest model with the current point cloud block distribution characteristics;
for cloud data of the points to be predicted, matching the characteristic distribution of the cloud data in a point cloud characteristic space distribution database, and predicting the cloud data of the points to be predicted by using a mature online random forest corresponding to the matching result:
with the increase of the number of the point cloud blocks subjected to manual correction, the distribution situation of the histogram distribution statistical point cloud blocks in the multidimensional feature space is designed, the designed new features are used as feature description of the histogram, a distribution similarity calculation formula is adopted to establish the distribution relation of the point cloud blocks subjected to manual correction and the predicted point cloud blocks, the training set and the predicted set are used for multiplexing the mature online random forest model under the condition of being consistent to perform initial semantic segmentation, manual correction is continued on the result predicted by using the online random forest model, a new online random forest model is trained, and the new online random forest model is added into the multiplexing model, so that the multiplexing range of the model is enlarged.
2. The online random forest model multiplexing-based large-scale point cloud semantic segmentation method according to claim 1, wherein the preprocessing of the point cloud data comprises:
and performing regular segmentation on the point cloud data to obtain point cloud segmented data, and performing initial segmentation on the point cloud segmented data by using a pre-training model of a main stream point cloud segmentation method to obtain an initial segmentation result, wherein the initial segmentation result is used for providing labels for training of an on-line random forest with strong supervision.
3. The method for large-scale point cloud semantic segmentation based on online random forest model multiplexing according to claim 2, wherein the feature extraction of the point cloud segmentation data and the acquisition of the mature online random forest model based on the point cloud segmentation data comprise the following steps:
wherein the designed features newly describe different semantic differences, including geometry, color, height Cheng Tedian of different semantic categories for enhancing the prediction capability of the online random forest model;
extracting multidimensional features of the point cloud partitioning data and recording the multidimensional features to a point cloud feature space distribution database;
and (3) taking the artificial correction data based on the semantic segmentation result of the online random forest model as a training set, iterating the online random forest model for training, obtaining a mature online random forest model meeting the prediction accuracy threshold of the point cloud segmentation data, and recording the mature online random forest model into a mature online random forest model library.
4. The method for large-scale point cloud semantic segmentation based on online random forest model reuse according to claim 3, wherein for the point cloud data to be predicted, matching the characteristic distribution in a point cloud characteristic space distribution database, predicting the point cloud data to be predicted by using a mature online random forest corresponding to the matching result, comprising:
matching a feature distribution by using a point feature distribution similarity evaluation algorithm, wherein the feature distribution is most similar to the feature distribution of the cloud data of the points to be predicted in a point cloud feature space distribution database;
corresponding to the point cloud multidimensional feature space distribution to be predicted and the mature online random forest model;
judging whether the prediction accuracy of the current point cloud block data obtained by the corresponding mature online random forest model meets a threshold value or not;
and outputting a prediction result meeting the use requirement, or taking point cloud block data corresponding to the point cloud multi-dimensional characteristic space distribution to be predicted as a check set input, continuing training the mature online random forest model, obtaining a fitted mature online random forest model meeting the check set input prediction accuracy, and binding the fitted mature online random forest model with the point cloud multi-dimensional characteristic space distribution information to be predicted currently plus the point cloud multi-dimensional characteristic distribution information of the original mature online random forest model.
5. The method for large-scale point cloud semantic segmentation based on online random forest model multiplexing according to claim 4, wherein the matching of the feature distribution with the point feature distribution similarity evaluation algorithm is performed by a feature distribution which is most similar to the feature distribution of the cloud data of the points to be predicted in a point cloud feature space distribution database, and comprises the steps of multidimensional feature space distribution calculation based on the point cloud, and specifically comprises the following steps:
the method comprises the steps of analyzing the ratio of semantics of point cloud block data to be predicted to total point cloud block data to be processed, extracting features according to different semantics of the point cloud block data to be predicted, obtaining multi-dimensional features of the point cloud block data to be predicted, and analyzing the feature value distribution condition of the point cloud block data to be predicted.
6. The online random forest model multiplexing-based large-scale point cloud semantic segmentation method according to claim 3, wherein the artificial correction data based on the online random forest model semantic segmentation result is used as a training set, the online random forest model is iterated for training, a mature online random forest model meeting the point cloud segmentation data prediction accuracy threshold is obtained, and the mature online random forest model is recorded in a mature online random forest model library, and the method comprises the following steps:
and training a segmentation prediction result of the point cloud semantic segmentation model for feature extraction by adopting an online random forest model, acquiring a random forest model semantic segmentation result in training, generating the manual correction data by adopting an error region of the manual marker segmentation result, inputting the manual correction data as a training set for updating branch nodes of different decision trees in the online random forest, and iteratively training to output a mature online random forest model meeting the point cloud segmentation data prediction accuracy.
7. The online random forest model multiplexing-based large-scale point cloud semantic segmentation method according to claim 1, wherein the multi-dimensional feature spatial distribution similarity comprises histogram similarity and SSIM structure similarity coefficients.
8. The large-scale point cloud semantic segmentation method based on online random forest model multiplexing as claimed in claim 2, which is characterized by comprising the following steps: and obtaining a prediction result for feature extraction of the corresponding point cloud partition data by using a weakly supervised deep learning model.
9. The large-scale point cloud semantic segmentation method based on online random forest model multiplexing as claimed in claim 2, which is characterized by comprising the following steps: and obtaining a prediction result for feature extraction of the corresponding point cloud partitioning data by using the pre-training point cloud semantic partitioning model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410028448.8A CN117541799B (en) | 2024-01-09 | 2024-01-09 | Large-scale point cloud semantic segmentation method based on online random forest model multiplexing |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410028448.8A CN117541799B (en) | 2024-01-09 | 2024-01-09 | Large-scale point cloud semantic segmentation method based on online random forest model multiplexing |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117541799A CN117541799A (en) | 2024-02-09 |
| CN117541799B true CN117541799B (en) | 2024-03-08 |
Family
ID=89794178
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410028448.8A Active CN117541799B (en) | 2024-01-09 | 2024-01-09 | Large-scale point cloud semantic segmentation method based on online random forest model multiplexing |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117541799B (en) |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102722725A (en) * | 2012-06-04 | 2012-10-10 | 西南交通大学 | Object tracing method based on active scene learning |
| CN105719279A (en) * | 2016-01-15 | 2016-06-29 | 上海交通大学 | Elliptic cylinder-based human trunk modeling, arm area segmentation and arm skeleton extraction method |
| CN108681525A (en) * | 2018-05-16 | 2018-10-19 | 福州大学 | A kind of road surface point cloud intensity enhancing method based on Vehicle-borne Laser Scanning data |
| CN113989535A (en) * | 2021-10-25 | 2022-01-28 | 辽宁工程技术大学 | Point cloud classification method combining region growing and random forest |
| WO2022088676A1 (en) * | 2020-10-29 | 2022-05-05 | 平安科技（深圳）有限公司 | Three-dimensional point cloud semantic segmentation method and apparatus, and device and medium |
| CN114842262A (en) * | 2022-05-10 | 2022-08-02 | 中国南方电网有限责任公司超高压输电公司检修试验中心 | Laser point cloud ground object automatic identification method fusing line channel orthographic images |
| CN115482380A (en) * | 2022-09-15 | 2022-12-16 | 电子科技大学 | Multi-level highway three-dimensional point cloud target segmentation method based on deep learning |
| CN115761172A (en) * | 2022-10-10 | 2023-03-07 | 哈尔滨工程大学 | Single building three-dimensional reconstruction method based on point cloud semantic segmentation and structure fitting |
| CN115861619A (en) * | 2022-12-20 | 2023-03-28 | 重庆大学 | Airborne LiDAR (light detection and ranging) urban point cloud semantic segmentation method and system of recursive residual double-attention kernel point convolution network |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11710239B2 (en) * | 2020-11-10 | 2023-07-25 | Here Global B.V. | Method, apparatus, and system using a machine learning model to segment planar regions |
-
2024
- 2024-01-09 CN CN202410028448.8A patent/CN117541799B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102722725A (en) * | 2012-06-04 | 2012-10-10 | 西南交通大学 | Object tracing method based on active scene learning |
| CN105719279A (en) * | 2016-01-15 | 2016-06-29 | 上海交通大学 | Elliptic cylinder-based human trunk modeling, arm area segmentation and arm skeleton extraction method |
| CN108681525A (en) * | 2018-05-16 | 2018-10-19 | 福州大学 | A kind of road surface point cloud intensity enhancing method based on Vehicle-borne Laser Scanning data |
| WO2022088676A1 (en) * | 2020-10-29 | 2022-05-05 | 平安科技（深圳）有限公司 | Three-dimensional point cloud semantic segmentation method and apparatus, and device and medium |
| CN113989535A (en) * | 2021-10-25 | 2022-01-28 | 辽宁工程技术大学 | Point cloud classification method combining region growing and random forest |
| CN114842262A (en) * | 2022-05-10 | 2022-08-02 | 中国南方电网有限责任公司超高压输电公司检修试验中心 | Laser point cloud ground object automatic identification method fusing line channel orthographic images |
| CN115482380A (en) * | 2022-09-15 | 2022-12-16 | 电子科技大学 | Multi-level highway three-dimensional point cloud target segmentation method based on deep learning |
| CN115761172A (en) * | 2022-10-10 | 2023-03-07 | 哈尔滨工程大学 | Single building three-dimensional reconstruction method based on point cloud semantic segmentation and structure fitting |
| CN115861619A (en) * | 2022-12-20 | 2023-03-28 | 重庆大学 | Airborne LiDAR (light detection and ranging) urban point cloud semantic segmentation method and system of recursive residual double-attention kernel point convolution network |
Non-Patent Citations (4)
| Title |
|---|
| Martin Weinmann 等.Semantic point cloud interpretation based on optimal neighborhoods, relevant features and efficient classifiers.《ELSEVIER》.2015,第105卷286-304. * |
| 杨超.一种用于遥感影像地表覆盖的语义分割算法.《中国优秀硕士学位论文全文数据库基础科学辑》.2023,(第03期),A008-144. * |
| 陈睿星 等.顾及长尾分布的机载LiDAR点云CNN语义分割.《仪器仪表学报》.2023,第44卷(第7期),282-295. * |
| 韩姗姗.基于深度学习的倾斜摄影测量点云建筑物语义与实例联合分割.《中国优秀硕士学位论文全文数据库基础科学辑》.2021,(第001期),A008-290. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117541799A (en) | 2024-02-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US8238660B2 (en) | Hybrid graph model for unsupervised object segmentation | |
| CN114092832B (en) | High-resolution remote sensing image classification method based on parallel hybrid convolutional network | |
| CN111444821A (en) | Automatic identification method for urban road signs | |
| CN102592268B (en) | Method for segmenting foreground image | |
| CN106503035A (en) | A kind of data processing method of knowledge mapping and device | |
| CN110675421B (en) | Depth image collaborative segmentation method based on few labeling frames | |
| CN105046714A (en) | Unsupervised image segmentation method based on super pixels and target discovering mechanism | |
| CN115049841A (en) | Depth unsupervised multistep anti-domain self-adaptive high-resolution SAR image surface feature extraction method | |
| CN113034511A (en) | Rural building identification algorithm based on high-resolution remote sensing image and deep learning | |
| CN114283285A (en) | Cross consistency self-training remote sensing image semantic segmentation network training method and device | |
| CN113989747A (en) | Terminal area meteorological scene recognition system | |
| CN116187835A (en) | Data-driven-based method and system for estimating theoretical line loss interval of transformer area | |
| CN112116612A (en) | Pavement tree image example segmentation method based on Mask R-CNN | |
| CN117152427A (en) | Remote sensing image semantic segmentation method and system based on diffusion model and knowledge distillation | |
| CN118015369B (en) | Obstacle detection system and method and electric lifting table | |
| CN117274750B (en) | Knowledge distillation semi-automatic visual labeling method and system | |
| CN102609721B (en) | Remote sensing image clustering method | |
| CN113988198A (en) | Multi-scale city function classification method based on landmark constraint | |
| CN117541799B (en) | Large-scale point cloud semantic segmentation method based on online random forest model multiplexing | |
| CN116958476A (en) | Building visual modeling method and system based on BIM data | |
| CN103336781B (en) | A kind of medical image clustering method | |
| CN113008226B (en) | Geomagnetic indoor positioning method based on gated cyclic neural network and particle filtering | |
| Geng et al. | A novel color image segmentation algorithm based on JSEG and Normalized Cuts | |
| Wang et al. | Application of 3D Image Reconstruction on Landscape Architecture in Environmental Design System | |
| CN114022520A (en) | Robot target tracking method based on Kalman filtering and twin network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |