CN117496309B - Building scene point cloud segmentation uncertainty evaluation method and system and electronic equipment - Google Patents
Building scene point cloud segmentation uncertainty evaluation method and system and electronic equipment Download PDFInfo
- Publication number
- CN117496309B CN117496309B CN202410006781.9A CN202410006781A CN117496309B CN 117496309 B CN117496309 B CN 117496309B CN 202410006781 A CN202410006781 A CN 202410006781A CN 117496309 B CN117496309 B CN 117496309B
- Authority
- CN
- China
- Prior art keywords
- point cloud
- building
- data set
- cloud segmentation
- entropy
- 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
- 230000011218 segmentation Effects 0.000 title claims abstract description 206
- 238000011156 evaluation Methods 0.000 title claims abstract description 28
- 238000000034 method Methods 0.000 claims abstract description 49
- 238000012549 training Methods 0.000 claims abstract description 39
- 238000012795 verification Methods 0.000 claims abstract description 15
- 238000010276 construction Methods 0.000 claims abstract description 13
- 238000012360 testing method Methods 0.000 claims description 44
- 230000002779 inactivation Effects 0.000 claims description 27
- 238000005457 optimization Methods 0.000 claims description 10
- 238000004590 computer program Methods 0.000 claims description 7
- 238000005192 partition Methods 0.000 claims description 7
- 230000009849 deactivation Effects 0.000 claims description 4
- 238000000638 solvent extraction Methods 0.000 claims description 3
- 238000011002 quantification Methods 0.000 abstract description 5
- 238000005259 measurement Methods 0.000 description 11
- 238000001514 detection method Methods 0.000 description 6
- 238000010586 diagram Methods 0.000 description 6
- 230000000694 effects Effects 0.000 description 6
- 238000004891 communication Methods 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 5
- 230000006870 function Effects 0.000 description 5
- 238000002372 labelling Methods 0.000 description 5
- 230000008569 process Effects 0.000 description 5
- 238000000605 extraction Methods 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 3
- 238000003908 quality control method Methods 0.000 description 3
- 125000000205 L-threonino group Chemical group [H]OC(=O)[C@@]([H])(N([H])[*])[C@](C([H])([H])[H])([H])O[H] 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 230000006399 behavior Effects 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000013210 evaluation model Methods 0.000 description 1
- 230000000873 masking effect Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000013139 quantization Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000012876 topography Methods 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
- 239000002699 waste material Substances 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
- G06V10/70—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
- G06V10/776—Validation; Performance evaluation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/136—Segmentation; Edge detection involving thresholding
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10028—Range image; Depth image; 3D point clouds
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Databases & Information Systems (AREA)
- Computing Systems (AREA)
- Artificial Intelligence (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention belongs to the technical field of intelligent construction and discloses a method, a system and electronic equipment for evaluating uncertainty of cloud segmentation of a building scene point, wherein the method comprises the following steps: constructing a point cloud data set, wherein the point cloud data set comprises a training data set and a verification data set; acquiring a trained point cloud segmentation model by using the training data set and the verification data set; performing point cloud segmentation on a building scene in actual application by using a point cloud segmentation model to obtain a point cloud segmentation result; obtaining entropy values corresponding to any building component category obtained through segmentation according to the point cloud segmentation result; and evaluating the uncertainty of the point cloud segmentation of any building component class according to the entropy value corresponding to the building component class. The invention utilizes the entropy value corresponding to the building component category in the segmentation prediction result to reflect the uncertainty of the segmentation prediction result, and utilizes the entropy value to realize the quantification of the uncertainty of the segmentation prediction result, thereby realizing the evaluation of the performance and the reliability of the point cloud segmentation model without marking data in advance.
Description
Technical Field
The invention belongs to the technical field of intelligent construction, and particularly relates to a method, a system and electronic equipment for evaluating uncertainty of cloud segmentation of a building scene point.
Background
The measurement of the geometric quality of the building mainly aims at actual measurement actual quantity in the acceptance specification of the building engineering. Compliance with building geometry quality, deviation from design values, etc. are among the main working matters. At present, in actual field work, the measurement of building geometric quality mainly depends on manual measurement, and tools including a guiding rule and a feeler gauge are adopted, so that the problems of poor accuracy, complex operation, irreproducible measurement results and the like exist, and the application of a more advanced method is urgently needed.
The point cloud has the characteristic of accurately capturing the surface topography points of the object, and is widely researched and applied in the field of building measurement. The point cloud is a data set of points in space and contains rich information such as three-dimensional coordinates, RGB color data and the like. The point cloud segmentation technology is to divide the original point cloud into blocks according to the information such as space, geometry, texture and the like, and extract objects in the point cloud, so as to improve the understanding of the point cloud scanning scene. The point cloud segmentation technology is widely applied to various fields such as construction, unmanned operation and the like at present, and is started to be applied to various works such as construction scene reconstruction, construction measurement and the like. The point cloud segmentation comprises two types of semantic segmentation and instance segmentation, wherein the semantic segmentation classifies the category to which each point belongs, but objects in the same category are not distinguished. The instance division divides the category to which each point belongs, and objects in the same category are also distinguished. The instance partition outputs a Mask (Mask) of the target object and its belonging category.
The point cloud segmentation can realize the work of extracting building components in a point cloud scene, such as a room roof, a floor, walls, windows and the like, and on one hand, the point cloud segmentation can help people to deepen understanding of a scanning scene; on the other hand, the components extracted by the point cloud segmentation can be further applied to the building geometric measurement analysis, and the point cloud can be directly pushed to be applied to the actual measurement of the building. Such applications include, but are not limited to: calculating flatness and perpendicularity of the wall through the extracted wall; calculating the size of the window hole by extracting the window hole; indoor clearance is calculated from the extracted ceiling and floor, etc.
However, the point cloud segmentation result has a problem that visual evaluation cannot be performed in practical application. Typically, the point cloud segmentation uses a deep learning (deep learning) model, such as a point cloud network (pointe++), a point cloud convolution (pointCNN), and the like. During model training and testing, the performance of the point cloud segmentation model is reflected by quantitative indexes such as average precision (Average Precision, AP), cross ratio (Interactionover Union, ioU) and the like, and the calculation of the indexes requires pre-labeled data. However, in practical application, because there is no pre-labeled data, the performance of the point cloud segmentation model in practical application cannot be evaluated by using indexes such as average precision, cross-correlation ratio and the like. In addition, the building environment has the characteristics of unstructured and high dynamic performance, which makes it necessary to adopt a method for helping people evaluate the performance of the building point cloud segmentation model in the absence of pre-labeled data.
Disclosure of Invention
Aiming at the defects or improvement demands of the prior art, the invention provides a method, a system and electronic equipment for evaluating the uncertainty of the cloud segmentation of the building scene, and solves the problem that the existing point cloud segmentation model does not have pre-marked data in the actual building scene application so as to evaluate the performance and the reliability of the point cloud segmentation model.
To achieve the above object, according to a first aspect of the present invention, there is provided a building scene point cloud segmentation uncertainty evaluation method, including:
model offline training phase:
s1, constructing a point cloud data set, wherein the point cloud data set comprises a training data set and a verification data set, and each piece of data in the point cloud data set comprises point cloud data of a building scene and a corresponding labeled building component type;
s2, acquiring a point cloud segmentation model after training optimization by using the training data set and the verification data set;
model actual application stage:
s3, performing point cloud segmentation on the building scene in actual application by using the point cloud segmentation model to obtain a point cloud segmentation result;
s4, obtaining entropy values corresponding to any one of the partitioned building component categories according to the point cloud partition result;
and S5, evaluating the uncertainty of the point cloud segmentation of any building component class according to the entropy value corresponding to the building component class.
According to the building scene point cloud segmentation uncertainty evaluation method provided by the invention, the point cloud data set in S1 also comprises a test data set; correspondingly, after S2 and before S5, the method further comprises:
performing point cloud segmentation prediction on the building scene in the test data set by using the point cloud segmentation model to obtain a prediction result of the test data set;
calculating entropy values and average precision corresponding to any building component category according to the prediction result of the test data set, and establishing a correlation between the entropy values and the average precision corresponding to any building component category;
determining an entropy threshold value for any building component category according to the correlation between the entropy value corresponding to any building component category and the average precision;
correspondingly, S5 specifically includes: and evaluating the uncertainty of the point cloud segmentation of any building component class according to the entropy value corresponding to the building component class and the entropy threshold value.
According to the building scene point cloud segmentation uncertainty evaluation method provided by the invention, the correlation between the entropy value and the average precision corresponding to any building component category is established, and the method specifically comprises the following steps:
carrying out multiple Monte Carlo inactivation predictions on any building scene in the test data set by utilizing the point cloud segmentation model to obtain multiple prediction results of any building scene in the test data set;
Calculating and obtaining entropy values corresponding to any building component category in any prediction result according to a plurality of prediction results of any building scene in the test data set;
calculating and obtaining the average precision corresponding to any building component category in any predicted result according to a plurality of predicted results of any building scene in the test data set and the corresponding marked building component category;
and establishing a correlation between entropy values and average precision corresponding to any building component category according to all prediction results of all building scenes in the test data set.
According to the building scene point cloud segmentation uncertainty evaluation method provided by the invention, the S5 specifically comprises the following steps:
and when the entropy value corresponding to any building component category is greater than or equal to the entropy threshold value, adopting external interference measures to denoise the point cloud segmentation result.
According to the building scene point cloud segmentation uncertainty evaluation method provided by the invention, the correlation between the entropy value and the average precision corresponding to any building component category is established, and the method further comprises the following steps:
judging the degree of negative correlation between the entropy value and the average precision corresponding to any building component category;
and when the degree of the negative correlation is smaller than a preset degree, adjusting model parameters of the point cloud segmentation model.
According to the building scene point cloud segmentation uncertainty evaluation method provided by the invention, the random inactivation rate parameter in the point cloud segmentation model is set to be 0.2-0.5.
According to the building scene point cloud segmentation uncertainty evaluation method provided by the invention, the entropy value corresponding to any building component category is calculated by the following modes:
；
wherein,representing the point cloud segmentation modeliFor building component category in secondary point cloud segmentation resultcEntropy value of (2);p（y = c | x j ) Representing a given firstjExamples of the inventionx j In this case, the point cloud segmentation results show that the instance belongs to the building element classcConfidence of (1)p；MAttributing building element class to partitioned instances in point cloud partitioning resultscIs a sum of (3).
According to the building scene point cloud segmentation uncertainty evaluation method provided by the invention, the average Precision corresponding to any building component category is obtained according to the area under the Precision-Recall curve, wherein the Precision (Precision) and Recall (Recall) are calculated by the following modes:
；
；
TP is a true positive sample, FP is a false positive sample, FN is a false negative sample, and judgment is carried out through different cross ratio thresholds.
According to a second aspect of the present invention, there is provided a building scene point cloud segmentation uncertainty evaluation system, comprising:
The system comprises a data set construction module, a data set generation module and a data set generation module, wherein the data set construction module is used for constructing a point cloud data set, the point cloud data set comprises a training data set and a verification data set, and each piece of data in the point cloud data set comprises point cloud data of a building scene and a corresponding marked building component category;
the model training module is used for acquiring a point cloud segmentation model after training optimization by utilizing the training data set and the verification data set;
the point cloud segmentation module is used for carrying out point cloud segmentation on the building scene in actual application by utilizing the point cloud segmentation model to obtain a point cloud segmentation result;
the entropy acquisition module is used for acquiring entropy values corresponding to any one of the partitioned building component categories according to the point cloud partition result;
and the evaluation module is used for evaluating the uncertainty of the point cloud segmentation of any building component class according to the entropy value corresponding to the building component class.
According to a third aspect of the present invention there is provided an electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the steps of the building scene point cloud segmentation uncertainty assessment method as described in any of the preceding claims when the program is executed.
In general, compared with the prior art, the technical scheme adopted by the invention has the advantages that the method, the system and the electronic equipment for evaluating the uncertainty of the cloud segmentation of the building scene point are provided by the invention:
1. the method and the device have the advantages that the uncertainty of the point cloud segmentation prediction result is reflected by the entropy value corresponding to the building component category in the point cloud segmentation prediction result, and the entropy value is utilized to reflect the prediction average precision to a certain extent due to the negative correlation between the entropy value and the average precision, so that the prediction uncertainty can be evaluated, the uncertainty of the point cloud segmentation prediction result is quantized by the entropy value, and the performance and the reliability of the point cloud segmentation model are evaluated without marking data in advance;
2. providing a correlation between the entropy value of any building component category and the prediction average precision, wherein the correlation can be obtained through the prediction result and the labeling data of the test data set, the entropy directly represents the uncertainty of the prediction result, the average precision directly represents the prediction performance of the point cloud segmentation model on the building component category, and the prediction performance of the model on the building component category can be indirectly judged through the uncertainty by establishing the correlation; an acceptable entropy threshold can be set according to the relation between the average precision and the entropy so as to directly use the entropy of the prediction result as the performance of the evaluation model;
3. According to the method, a plurality of different prediction results can be randomly obtained for any building scene through Monte Carlo random inactivation, the entropy value and the prediction average precision of any building component category in the prediction results can be obtained according to each prediction result, the uncertainty of the prediction result of the point cloud segmentation model can be better captured through Monte Carlo random inactivation, the corresponding data of more entropy values and average precision for any building component category can be obtained, so that the establishment of the correlation between the entropy values and the average precision is more accurate and complete, and the accuracy of the prediction performance of the point cloud segmentation model is reflected by utilizing the entropy values is improved;
4. the invention provides a method for evaluating the segmentation effect of the point cloud segmentation model in the actual application process, and provides a powerful support for the technology landing of the point cloud segmentation model; the understanding of the model can be effectively improved, and the trust degree of the artificial intelligence model is improved; the method provides a basis for controlling the quality of the point cloud segmentation result, and guides the introduction of external intervention to optimize the segmentation result when the uncertainty of the segmentation result is excessive.
Drawings
Fig. 1 is a flowchart of a building scene point cloud segmentation uncertainty evaluation method provided by the invention.
Fig. 2 is an overall flowchart of the building scene point cloud segmentation uncertainty evaluation method provided by the invention in a specific embodiment.
Fig. 3 is a flow chart based on monte carlo random inactivation provided by the present invention.
Fig. 4 is a flowchart of the control of the segmentation result of the point cloud segmentation model according to the entropy threshold.
Fig. 5 is a schematic diagram of an electronic device provided by the present invention.
Detailed Description
The present invention will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention. In addition, the technical features of the embodiments of the present invention described below may be combined with each other as long as they do not collide with each other.
Referring to fig. 1, the present invention provides a method for evaluating uncertainty of cloud segmentation of a building scene, which includes:
model offline training phase:
s1, constructing a point cloud data set, wherein the point cloud data set comprises a training data set and a verification data set, and each piece of data in the point cloud data set comprises point cloud data of a building scene and a corresponding labeled building component type; namely, the data in the point cloud data set are point cloud scene data with actual result labels;
S2, acquiring a point cloud segmentation model after training optimization by using the training data set and the verification data set;
model actual application stage:
s3, performing point cloud segmentation on the building scene in actual application by using the point cloud segmentation model to obtain a point cloud segmentation result;
s4, obtaining entropy values corresponding to any one of the partitioned building component categories according to the point cloud partition result;
and S5, evaluating the uncertainty of the point cloud segmentation of any building component class according to the entropy value corresponding to the building component class.
In this embodiment, it is proposed to quantify the uncertainty of the point cloud segmentation model in segmenting the prediction result by the entropy of the point cloud segmentation model prediction result, since the average accuracy of the prediction result decreases with the increase of the entropy in the prediction result. Thus, the average accuracy of the prediction result may be reflected to some extent by entropy, by which uncertainty of the prediction result is quantified. Meanwhile, the calculation of the entropy does not need to be marked with data in advance, but can reflect the performance and reliability of the point cloud segmentation model.
According to the building scene point cloud segmentation uncertainty evaluation method provided by the invention, the marked building scene point cloud data are adopted to train and optimize the point cloud segmentation model, and the point cloud segmentation model has good adaptability in actual building scene application; the method and the device for estimating the uncertainty of the point cloud segmentation model by utilizing the entropy value can be used for estimating the uncertainty of the point cloud segmentation model, and realize the quantification of the uncertainty of the point cloud segmentation model by utilizing the entropy value without marking data in advance.
The invention provides a method for evaluating the segmentation effect of the point cloud segmentation model in the actual application process, and provides a powerful support for the technology landing of the point cloud segmentation model; the understanding of the model can be effectively improved, and the trust degree of the artificial intelligence model is improved; the method provides a basis for controlling the quality of the point cloud segmentation result, and guides the introduction of external intervention to optimize the segmentation result when the uncertainty of the segmentation result is excessive.
Further, the point cloud data set in S1 further includes a test data set; correspondingly, after S2 and before S5, the method further comprises:
performing point cloud segmentation prediction on the building scene in the test data set by using the point cloud segmentation model to obtain a prediction result of the test data set;
calculating entropy values and average precision corresponding to any building component category according to the prediction result of the test data set, and establishing a correlation between the entropy values and the average precision corresponding to any building component category;
determining an entropy threshold value for any building component category according to the correlation between the entropy value corresponding to any building component category and the average precision;
correspondingly, S5 specifically includes: and evaluating the uncertainty of the point cloud segmentation of any building component class according to the entropy value corresponding to the building component class and the entropy threshold value.
In other words, the embodiment proposes that the correlation between the entropy value of any building component category and the prediction average precision can be obtained through the prediction result and the labeling data of the test data set, the entropy directly represents the uncertainty of the prediction result, and the average precision directly represents the quality of the prediction performance of the point cloud segmentation model in the building component category. By establishing the interrelation, the method can indirectly judge the quality of the prediction performance of the model on the building component category through uncertainty, and can reflect the prediction performance of the point cloud segmentation model without the need of pre-data labeling. An acceptable entropy threshold may be set based on the relationship of average accuracy and entropy so that the entropy of the prediction results is directly used as an estimate of model performance. Meanwhile, whether the noise needs to be manually removed can be determined according to the entropy of the prediction result.
In a specific embodiment, the correlation between the entropy value and the average precision corresponding to any building element category is established, and specifically includes:
carrying out multiple Monte Carlo inactivation predictions on any building scene in the test data set by utilizing the point cloud segmentation model to obtain multiple prediction results of any building scene in the test data set;
Calculating and obtaining entropy values corresponding to any building component category in any prediction result according to a plurality of prediction results of any building scene in the test data set;
calculating and obtaining the average precision corresponding to any building component category in any predicted result according to a plurality of predicted results of any building scene in the test data set and the corresponding marked building component category;
and establishing a correlation between entropy values and average precision corresponding to any building component category according to all prediction results of all building scenes in the test data set. Namely, the entropy values and average precision corresponding data belonging to the same building component class are collected in all prediction results of all building scenes, and the correlation between the entropy values and the average precision of the building component class is obtained.
According to the embodiment, a plurality of different prediction results can be randomly obtained for any building scene through Monte Carlo random inactivation, the entropy value and the prediction average precision of any building component category in the prediction results can be obtained according to each prediction result, the uncertainty of the prediction results of the point cloud segmentation model can be better captured through Monte Carlo random inactivation, the corresponding data of more entropy values and average precision for any building component category can be obtained, and therefore the establishment of the correlation between the entropy values and the average precision is facilitated to be accurate and complete, and the accuracy of the point cloud segmentation model prediction performance is improved by utilizing the entropy values.
In some specific embodiments, in order to achieve the goal of accurately measuring the house size, the building component size and the like by using the point cloud in the field of actually measured real quantities of buildings, the extraction of the building components needs to be achieved by using a point cloud segmentation technology, so that the detection of the measurement target is further completed on the specified building components. However, the actual application of the point cloud segmentation model is different from the training test of the model. In practical applications, the model may not accurately segment the components, but at the same time, there is no intuitive index that can prompt how the model segments, which hinders further accurate measurement of the segmented components. In order to help evaluate the performance of a point cloud segmentation model used in an actual building scene, and facilitate manual denoising of a point cloud segmented building component result when necessary, the embodiment provides a point cloud segmentation uncertainty evaluation and control method for building geometric quality detection.
Therefore, the embodiment is an evaluation and control method for the point cloud segmentation effect of the building component, which is used for effectively realizing the application of the point cloud segmentation in the aspect of building geometric quality detection and is based on the uncertainty quantification of the point cloud segmentation prediction result. The point cloud segmentation uncertainty evaluation and control method for building geometric quality detection is characterized in that: capturing and quantifying uncertainty of a point cloud instance segmentation result; evaluating the point cloud instance segmentation effect by using uncertainty; and performing external intervention on the segmentation result with excessively high uncertainty of the prediction result so as to further improve the accuracy of the segmentation component.
Referring to fig. 2, a point cloud segmentation uncertainty evaluation and control method for building geometric quality detection includes the following steps:
(1) Building element point cloud data collection and pretreatment: collecting building component point cloud data, labeling the collected data according to component types, training a data set, and dividing a verification data set and a test data set;
wherein the data acquisition, more specifically:
and scanning the construction scene site by adopting a three-dimensional laser scanner to obtain point cloud data, splicing the obtained point cloud data and converting necessary file formats to form a PCD file, so that the model in the point cloud segmentation model can be conveniently input for training.
Data preprocessing, more specifically:
labeling point cloud data belonging to different building element categories, wherein the building element comprises 11 categories: pure plane wall, wall with window hole, wall with door hole, wall with window and door hole, top plate, bottom plate, window, door, construction waste, wash platform, noise point. And performing test data set on the obtained data set, and verifying the division of the data set and the training data set. Meanwhile, the data enhancement method is adopted to increase the quantity of point cloud data available for training, namely, the data quantity in a training data set is expanded.
(2) Training a point cloud segmentation model and optimizing parameters: using point cloud segmentation modelsf(.) model training and parameter optimization on the training dataset and the validation dataset to obtain a model with higher average accuracy (mAP) over all building element categories;
further, the affiliated point cloud instance segmentation modelf(. Cndot.) Voxel-based (Voxel-based) or Pixel-level based models can be employed. Wherein the voxel point cloud based instance segmentation network outputs which instance mask (mask) each voxel belongs to, and the probability or confidence score of the semantic category to which the mask belongspThe method comprises the steps of carrying out a first treatment on the surface of the Also, a pixel-level based point cloud instance segmentation model models and analyzes directly on each point of the point cloud, the model would be a mask (mask) of which instance each point belongs to,and assigning a confidence score to the semantic category to which the mask belongsp。
In a specific embodiment, as a further preferred, a voxel-based point cloud instance segmentation network Mask three-dimensional model (Mask 3D) is employed as the model in the embodiment. Point cloud segmentation model based on Mask three-dimensional model (Mask 3D) networkf(.) model training and optimization of the obtained test dataset and training dataset to obtain a model with high average accuracy (mAP) over all building element categories.
Further, the Mask three-dimensional model (Mask 3D) is the first point cloud segmentation model based on a self-attention model (transducer). It consists of a feature extraction module (Feature Extractor) and a Decoder (Decoder) module, which can be used for instance segmentation and semantic segmentation. The parameter optimization refers to optimizing and adjusting the depth of a Decoder (Decoder) and random inactivation parameters (dropout) of a Mask three-dimensional model (Mask 3D) network, so that the uncertainty and the model segmentation performance have stronger correlation on the premise of ensuring good segmentation performance of the point cloud segmentation model.
The Mask three-dimensional model (Mask 3D) is input as point cloud voxels, which are cube units that convert point cloud data into rules in three-dimensional space, facilitating subsequent processing and analysis. Mask three-dimensional model (Mask 3D) output as outputiConfidence score for the split instance in the secondary run. Said firstiThe confidence score of a split instance in a run-time refers to: normalized exponential function (softmax) output based on dominant semantic category queriesc c1 （c c1 ∈[0，1]) Confidence with mask-basedm k Multiplying:
；
wherein,m i ∈ [0，1]is the first queryiExample mask confidence for individual voxels; MIs the number of voxels into which the point cloud data is partitioned.
Further, the random inactivation rate parameter of the Mask three-dimensional model (Mask 3D) is set to be 0.2-0.5, for example, can be 0.2; mask three-dimensional model (Mask 3D) feature extraction of voxels using a sparse convolution U-shaped backbone network (U-net) based on a MinkowskiEngine (MinkowskiEngine); given a set of specific instance queries, the Mask three-dimensional model (Mask 3D) will output a corresponding binary Mask; predicting semantic tags of the single instance with a normalized exponential function (softmax); item one of the single queryiThe instance masked confidence of each voxel is multiplied by the output of the corresponding normalized index function (softmax) to obtain a confidence scorep。
The random inactivation rate in this embodiment is set within (0, 1) and is not 0 or 1. The setting of the inactivation rate parameter firstly can reduce the complexity of the model, secondly is an essential step for realizing the Monte Carlo inactivation method, and thirdly, the random inactivation rate setting influences the strength of the relation between the model uncertainty and the model performance.
The point cloud voxels are represented as two parts of a coordinate matrix and a feature matrix based on a sparse convolution of a MinkowskiEngine (MinkowskiEngine), the backbone network turns the colored points into voxels, each voxel being assigned the average RGB color of the points within the voxel as its initial feature. In one embodiment of the invention, a U-shaped backbone network (U-net) extracts multi-resolution hierarchical features of input point cloud voxels, the multi-resolution features and full-resolution features are further processed in a self-attention model (transducer) decoder block.
(3) Monte Carlo stochastic inactivation capture model prediction result uncertainty: point cloud segmentation model after training optimizationf(. Cndot.) in a given point cloud scenario in a test datasetNsub-Monte Carlo inactivation (Monte Carlo Dropout, MC-dropout) prediction to obtain random inactivation output under a given point cloud scenarioN-a number of monte carlo samples (Monte Carlo sample, MC-sample), which are prediction results in a given point cloud scenario;
monte Carlo inactivation activates the point cloud segmentation model during the test phasefRandom deactivation of (-)Function. Due to at the firsti（i∈[0，N]) In the process of sub-random inactivation, the modelf(. Cndot.) the different neurons are randomly ignored, i.e. the weight parameters in the model are randomly adjustedΦ i Therefore, the point cloud segmentation model outputs different prediction results for the same given point cloud scene.
The prediction (i.e., monte carlo samples) contains a point cloud segmentation model for each instance segmented (tojRepresent the firstjExamples), which belongs to the confidence of the building element class. Confidence P of the home building element category j i Meaning that in all building element categories, the segmented instances belong to a certain building element category cMaximum probability P of (2) j i =argmax c p j i Whereini∈[0，N]，cIs a building element category.
Specifically, a trained point cloud segmentation model is usedf(. Cndot.) is applied to a test data set, monte Carlo random inactivation is adopted, a random inactivation function is kept in an activated state in the test process, and random inactivation output under a given building point cloud scene is obtainedNA single monte carlo sample, more specifically:
in a monte carlo random inactivation process, the point cloud data X under a certain building scene in a given test set,f(. Cndot.) is a trained masked three-dimensional model, and ω represents model-learnable parameters. Masking a three-dimensional model (Mask 3D) will output the confidence level P that the segmentation instance belongs to a certain semantic class j i ，（i∈[0，N]) Wherein N represents the number of random deactivations of Monte Carlo,jrepresentation model predicted firstjAn instance mask.
P j i =f（X，Φ i (ω)）；
Further, for point cloud data for each given room, i.e., building scene, monte carlo random deactivation may be applied 20 times (n=20) to obtain different instance segmentation results for the same building point cloud scene.
(4) Based on entropy quantization, drawing prediction result uncertainty: and quantifying the uncertainty of the prediction of each building component class in the Monte Carlo sample by utilizing entropy aiming at the Monte Carlo sample outputted by each random inactivation, and obtaining a relation diagram of the uncertainty and the average precision of each building component class.
The quantification of the uncertainty of each building element category prediction in the Monte Carlo sample by using entropy refers to the quantification of the uncertainty of the building element category prediction in the Monte Carlo sample by calculating the entropy of the building element category prediction by using the confidence coefficient P of all the instances belonging to the same building element category. The entropy is calculated using the following formula:
Entropy= ∑-PlogP；
further, the average accuracy is obtained from the area under the accuracy (Precision) -Recall (Recall) curve (i.e., P-R curve). The calculation formulas of the precision and the recall rate are respectively as follows:
Precesion= TP/(TP + FP)；
Recall= TP/(TP + FN)；
TP is a true positive sample, FP is a false positive sample, FN is a false negative sample, and judgment is carried out through different cross ratio thresholds.
Further, in the graph of uncertainty versus average accuracy for each building element category, the abscissa is entropy for each building element category in the monte carlo sample and the ordinate is average accuracy for each building element category. Entropy directly represents uncertainty of a prediction result, and average precision directly represents quality of prediction performance of the model in the component category. By establishing a relation diagram, the method realizes the indirect judgment of the quality of the predicted performance of the model on the building component category through uncertainty.
As shown in fig. 3, the uncertainty of each building element type prediction in each monte carlo sample under the given building point cloud scene X is quantized by entropy, a relation diagram is formed with the average accuracy of the model in each building element type, and the range of the entropy of each building element type prediction result in the monte carlo sample is further drawn. The range may be embodied in the predicted behavior of the model on that building element class for a given building point cloud scenario X.
In order to realize the steps, the confidence scores of all the examples belonging to each building component category in the Monte Carlo sample are calculated to obtain entropy, so that the aim of quantifying the prediction uncertainty is fulfilled. For building element categoriescIts first oneiEntropy of sub-Monte Carlo random inactivation prediction resultEntropy c i Expressed as:
；
wherein,p（y=c|x j ) Representing a given firstjExamples of the inventionx j In the case that the model predictions show that the instance belongs to the class of building elementscConfidence of (1)p；MAttributing building element class to partitioned instances in point cloud partitioning resultscIs a sum of (3).
The predictive representation of the Mask three-dimensional model (Mask 3D) on the test set is represented with average accuracy and a graph of entropy versus average accuracy for each building element class is plotted to determine a quality control threshold for point cloud segmentation.
Specifically, the average accuracy uses an ap@25 index, and ap@25 represents the average accuracy measured using an intersection ratio threshold of 0.25 (25%). This threshold is relatively relaxed and allows for detection results that have a low degree of overlap with the ground truth object.
Further, the relationship diagram formed by the entropy and the average precision of each building component category is obtained by taking the average precision of each building component category as an ordinate and the entropy as an abscissa, so that the point cloud segmentation result is evaluated through an entropy value. Establishing a correlation between entropy values and average precision corresponding to any building component category, and further comprising:
judging the degree of negative correlation between the entropy value and the average precision corresponding to any building component category;
and when the degree of the negative correlation is smaller than a preset degree, adjusting model parameters of the point cloud segmentation model.
Since entropy and average accuracy should theoretically exhibit a negative correlation, i.e. the correlation r e [ -1,0]. When the correlation is not obvious, particularly when r epsilon [ -0.4,0], the random inactivation parameters of the point cloud segmentation model, the model feature extraction network depth, the number of model decoders or the granularity of application can be adjusted, and the prediction is adjusted from voxel-based to pixel-based.
Further, the further drawing of the range of the entropy of the prediction results of a certain type of building components means that the average value of all the prediction results of the building components is taken as a center point under a given building point cloud scene X, and an ellipse is drawn on a coordinate axis by taking 3 times of the standard deviation of the average precision and 3 times of the standard deviation of the entropy as the height and the width of the ellipse respectively. The ellipse center, width and height can embody the prediction performance of the point cloud segmentation model on the members. The farther the center is from the ordinate (representing average accuracy), the larger the width is, the more unstable the predicted result is, the larger the accuracy fluctuation is, the lower the reliability is, and external intervention is more needed to adjust the segmentation result.
(5) Model segmentation result control: an acceptable entropy threshold is set by the entropy-average accuracy relationship graph. And evaluating whether a prediction result of the point cloud segmentation model in actual application meets the requirements according to the entropy threshold value so as to realize control of the segmentation effect of the point cloud segmentation model in actual application. The undesirable segmentation result is improved by external intervention.
The control of the segmentation effect means that when the entropy value corresponding to any building component category is larger than or equal to the entropy threshold value, external interference measures are adopted to denoise the point cloud segmentation result. And adopting external intervention measures to remove noise points in the segmentation result by using the prediction result with excessive entropy of the point cloud segmentation model.
Further, according to the correlation between entropy and average precision, a relation curve of the entropy and the average precision can be obtained by fitting. Specifically, a quality control threshold (thres) of uncertainty is set for entropy according to a relation, and a corresponding abscissa entropy threshold is determined by average precision ap=0.8. When the entropy of the output result exceeds a quality control threshold (thres), external intervention will be performed, further noise reduction of the segmentation result. The control flow of the point cloud segmentation result is shown in fig. 4.
The invention also provides a system for evaluating the uncertainty of the cloud segmentation of the building scene, which is used for realizing the method for evaluating the uncertainty of the cloud segmentation of the building scene according to any one of the embodiments, and the system can be correspondingly referred to and understood with the method, and comprises the following steps:
the system comprises a data set construction module, a data set generation module and a data set generation module, wherein the data set construction module is used for constructing a point cloud data set, the point cloud data set comprises a training data set and a verification data set, and each piece of data in the point cloud data set comprises point cloud data of a building scene and a corresponding marked building component category;
the model training module is used for acquiring a point cloud segmentation model after training optimization by utilizing the training data set and the verification data set;
The point cloud segmentation module is used for carrying out point cloud segmentation on the building scene in actual application by utilizing the point cloud segmentation model to obtain a point cloud segmentation result;
the entropy acquisition module is used for acquiring entropy values corresponding to any one of the partitioned building component categories according to the point cloud partition result;
and the evaluation module is used for evaluating the uncertainty of the point cloud segmentation of any building component class according to the entropy value corresponding to the building component class.
Further, the invention also provides an electronic device, which comprises a memory, a processor and a computer program stored in the memory and capable of running on the processor, wherein the processor realizes the steps of the building scene point cloud segmentation uncertainty assessment method according to any one of the embodiments when executing the program.
Further, the present invention also provides a non-transitory computer readable storage medium, on which a computer program is stored, which when executed by a processor, implements the steps of the building scene point cloud segmentation uncertainty assessment method according to any one of the embodiments above.
Fig. 5 illustrates a physical schematic diagram of an electronic device, as shown in fig. 5, which may include: a processor (processor), a communication interface (Communications Interface), a memory (memory) and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other via the communication bus. The processor may invoke logic instructions in the memory to perform the building scene point cloud segmentation uncertainty assessment method.
Further, the logic instructions in the memory described above may be implemented in the form of software functional units and stored in a computer-readable storage medium when sold or used as a stand-alone product. Based on this understanding, the technical solution of the present invention may be embodied essentially or in a part contributing to the prior art or in a part of the technical solution, in the form of a software product stored in a storage medium, comprising several instructions for causing a computer device (which may be a personal computer, a server, a network device, etc.) to perform all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a magnetic disk, or an optical disk, or other various media capable of storing program codes.
In another aspect, the present invention also provides a computer program product comprising a computer program stored on a non-transitory computer readable storage medium, the computer program comprising program instructions which, when executed by a computer, enable the computer to perform the building scene point cloud segmentation uncertainty assessment method provided by the methods described above.
The apparatus embodiments described above are merely illustrative, wherein the elements illustrated as separate elements may or may not be physically separate, and the elements shown as elements may or may not be physical elements, may be located in one place, or may be distributed over a plurality of network elements. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art will understand and implement the present invention without undue burden.
From the above description of the embodiments, it will be apparent to those skilled in the art that the embodiments may be implemented by means of software plus necessary general hardware platforms, or of course may be implemented by means of hardware. Based on this understanding, the foregoing technical solution may be embodied essentially or in a part contributing to the prior art in the form of a software product, which may be stored in a computer readable storage medium, such as ROM/RAM, a magnetic disk, an optical disk, etc., including several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the method described in the respective embodiments or some parts of the embodiments.
It will be readily appreciated by those skilled in the art that the foregoing description is merely a preferred embodiment of the invention and is not intended to limit the invention, but any modifications, equivalents, improvements or alternatives falling within the spirit and principles of the invention are intended to be included within the scope of the invention.
Claims (9)
1. The method for evaluating the uncertainty of the cloud segmentation of the building scene points is characterized by comprising the following steps of:
model offline training phase:
s1, constructing a point cloud data set, wherein the point cloud data set comprises a training data set and a verification data set, and each piece of data in the point cloud data set comprises point cloud data of a building scene and a corresponding labeled building component type;
s2, acquiring a point cloud segmentation model after training optimization by using the training data set and the verification data set;
model actual application stage:
s3, performing point cloud segmentation on the building scene in actual application by using the point cloud segmentation model to obtain a point cloud segmentation result;
s4, obtaining entropy values corresponding to any one of the partitioned building component categories according to the point cloud partition result;
s5, evaluating the uncertainty of the point cloud segmentation of any building component class according to the entropy value corresponding to the building component class;
The point cloud data set in S1 further comprises a test data set; correspondingly, after S2 and before S5, the method further comprises:
performing point cloud segmentation prediction on the building scene in the test data set by using the point cloud segmentation model to obtain a prediction result of the test data set;
calculating entropy values and average precision corresponding to any building component category according to the prediction result of the test data set, and establishing a correlation between the entropy values and the average precision corresponding to any building component category;
determining an entropy threshold value for any building component category according to the correlation between the entropy value corresponding to any building component category and the average precision;
correspondingly, S5 specifically includes: and evaluating the uncertainty of the point cloud segmentation of any building component class according to the entropy value corresponding to the building component class and the entropy threshold value.
2. The building scene point cloud segmentation uncertainty evaluation method according to claim 1, wherein the establishing of the correlation between the entropy value and the average precision corresponding to any building element category specifically comprises:
carrying out multiple Monte Carlo inactivation predictions on any building scene in the test data set by utilizing the point cloud segmentation model to obtain multiple prediction results of any building scene in the test data set;
Calculating and obtaining entropy values corresponding to any building component category in any prediction result according to a plurality of prediction results of any building scene in the test data set;
calculating and obtaining the average precision corresponding to any building component category in any predicted result according to a plurality of predicted results of any building scene in the test data set and the corresponding marked building component category;
and establishing a correlation between entropy values and average precision corresponding to any building component category according to all prediction results of all building scenes in the test data set.
3. The building scene point cloud segmentation uncertainty evaluation method as set forth in claim 1, wherein S5 specifically includes:
and when the entropy value corresponding to any building component category is greater than or equal to the entropy threshold value, adopting external interference measures to denoise the point cloud segmentation result.
4. The building scene point cloud segmentation uncertainty evaluation method according to claim 1, wherein the correlation between the entropy value and the average precision corresponding to any building element category is established, further comprising:
judging the degree of negative correlation between the entropy value and the average precision corresponding to any building component category;
And when the degree of the negative correlation is smaller than a preset degree, adjusting model parameters of the point cloud segmentation model.
5. The building scene point cloud segmentation uncertainty assessment method according to any of claims 1-4, wherein a random deactivation rate parameter in said point cloud segmentation model is set to 0.2-0.5.
6. The building scene point cloud segmentation uncertainty assessment method according to any of claims 1-4, wherein the entropy value corresponding to any building element class is calculated by:
；
wherein,representing the point cloud segmentation modeliIn the secondary point cloud segmentation resultFor building element categoriescEntropy value of (2);p（y = c | x j ) Representing a given firstjExamples of the inventionx j In this case, the point cloud segmentation results show that the instance belongs to the building element classcConfidence of (1)p；MAttributing building element class to partitioned instances in point cloud partitioning resultscIs a sum of (3).
7. The building scene point cloud segmentation uncertainty assessment method according to any of claims 1-4, wherein the average Precision corresponding to any building element class is derived from the area under the Precision-Recall curve, wherein the Precision (Precision) and Recall (Recall) are calculated by:
；
；
TP is a true positive sample, FP is a false positive sample, FN is a false negative sample, and judgment is carried out through different cross ratio thresholds.
8. A building scene point cloud segmentation uncertainty evaluation system, comprising:
the system comprises a data set construction module, a data set generation module and a data set generation module, wherein the data set construction module is used for constructing a point cloud data set, the point cloud data set comprises a training data set and a verification data set, and each piece of data in the point cloud data set comprises point cloud data of a building scene and a corresponding marked building component category;
the model training module is used for acquiring a point cloud segmentation model after training optimization by utilizing the training data set and the verification data set;
the point cloud segmentation module is used for carrying out point cloud segmentation on the building scene in actual application by utilizing the point cloud segmentation model to obtain a point cloud segmentation result;
the entropy acquisition module is used for acquiring entropy values corresponding to any one of the partitioned building component categories according to the point cloud partition result;
the evaluation module is used for evaluating the point cloud segmentation uncertainty of any building component class according to the entropy value corresponding to the building component class;
the point cloud data set further comprises a test data set; correspondingly, the system further comprises:
Performing point cloud segmentation prediction on the building scene in the test data set by using the point cloud segmentation model to obtain a prediction result of the test data set;
calculating entropy values and average precision corresponding to any building component category according to the prediction result of the test data set, and establishing a correlation between the entropy values and the average precision corresponding to any building component category;
determining an entropy threshold value for any building component category according to the correlation between the entropy value corresponding to any building component category and the average precision;
correspondingly, the evaluation module is specifically configured to: and evaluating the uncertainty of the point cloud segmentation of any building component class according to the entropy value corresponding to the building component class and the entropy threshold value.
9. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the steps of the building scene point cloud segmentation uncertainty assessment method according to any one of claims 1 to 7 when the program is executed.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410006781.9A CN117496309B (en) | 2024-01-03 | 2024-01-03 | Building scene point cloud segmentation uncertainty evaluation method and system and electronic equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410006781.9A CN117496309B (en) | 2024-01-03 | 2024-01-03 | Building scene point cloud segmentation uncertainty evaluation method and system and electronic equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117496309A CN117496309A (en) | 2024-02-02 |
| CN117496309B true CN117496309B (en) | 2024-03-26 |
Family
ID=89669469
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410006781.9A Active CN117496309B (en) | 2024-01-03 | 2024-01-03 | Building scene point cloud segmentation uncertainty evaluation method and system and electronic equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117496309B (en) |
Citations (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106780503A (en) * | 2016-12-30 | 2017-05-31 | 北京师范大学 | Remote sensing images optimum segmentation yardstick based on posterior probability information entropy determines method |
| CN109033588A (en) * | 2018-07-13 | 2018-12-18 | 成都爱为贝思科技有限公司 | A kind of uncertain quantization method based on spatial |
| CN110111343A (en) * | 2019-05-07 | 2019-08-09 | 齐鲁工业大学 | A kind of middle intelligence image partition method and device based on improvement fuzzy C-mean algorithm |
| CN111639682A (en) * | 2020-05-13 | 2020-09-08 | 北京三快在线科技有限公司 | Ground segmentation method and device based on point cloud data |
| EP3767521A1 (en) * | 2019-07-15 | 2021-01-20 | Promaton Holding B.V. | Object detection and instance segmentation of 3d point clouds based on deep learning |
| CN112733945A (en) * | 2021-01-13 | 2021-04-30 | 中国科学技术大学 | Gaussian dropout-based deep learning model uncertainty calculation method |
| CN113326759A (en) * | 2021-05-26 | 2021-08-31 | 中国地质大学（武汉） | Uncertainty estimation method for remote sensing image building identification model |
| CN113516662A (en) * | 2021-01-08 | 2021-10-19 | 清华大学 | Point cloud segmentation method and device based on multi-resolution fusion |
| CN114241110A (en) * | 2022-02-23 | 2022-03-25 | 北京邮电大学 | Point cloud semantic uncertainty sensing method based on neighborhood aggregation Monte Carlo inactivation |
| 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 |
| CN115273081A (en) * | 2022-08-16 | 2022-11-01 | 西安交通大学 | Point cloud semantic segmentation method and system based on self-adaptive feature sampling |
| CN115496792A (en) * | 2022-09-22 | 2022-12-20 | 北京航空航天大学杭州创新研究院 | Point cloud semantic SLAM method, system and device based on human-in-the-loop optimization |
| CN115540752A (en) * | 2022-09-21 | 2022-12-30 | 中建三局第一建设工程有限责任公司 | Engineering quality allowable deviation detection system based on laser point cloud and construction method |
| 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 |
| CN116524960A (en) * | 2023-05-08 | 2023-08-01 | 华东理工大学 | Speech emotion recognition system based on mixed entropy downsampling and integrated classifier |
| CN116630623A (en) * | 2023-05-19 | 2023-08-22 | 上海交通大学宁波人工智能研究院 | Industrial scene-oriented workpiece point cloud instance segmentation method |
| CN116682021A (en) * | 2023-05-22 | 2023-09-01 | 中国地质大学（武汉） | High-resolution remote sensing image building vector outline data extraction method |
-
2024
- 2024-01-03 CN CN202410006781.9A patent/CN117496309B/en active Active
Patent Citations (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106780503A (en) * | 2016-12-30 | 2017-05-31 | 北京师范大学 | Remote sensing images optimum segmentation yardstick based on posterior probability information entropy determines method |
| CN109033588A (en) * | 2018-07-13 | 2018-12-18 | 成都爱为贝思科技有限公司 | A kind of uncertain quantization method based on spatial |
| CN110111343A (en) * | 2019-05-07 | 2019-08-09 | 齐鲁工业大学 | A kind of middle intelligence image partition method and device based on improvement fuzzy C-mean algorithm |
| EP3767521A1 (en) * | 2019-07-15 | 2021-01-20 | Promaton Holding B.V. | Object detection and instance segmentation of 3d point clouds based on deep learning |
| CN111639682A (en) * | 2020-05-13 | 2020-09-08 | 北京三快在线科技有限公司 | Ground segmentation method and device based on point cloud 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 |
| CN113516662A (en) * | 2021-01-08 | 2021-10-19 | 清华大学 | Point cloud segmentation method and device based on multi-resolution fusion |
| CN112733945A (en) * | 2021-01-13 | 2021-04-30 | 中国科学技术大学 | Gaussian dropout-based deep learning model uncertainty calculation method |
| CN113326759A (en) * | 2021-05-26 | 2021-08-31 | 中国地质大学（武汉） | Uncertainty estimation method for remote sensing image building identification model |
| CN114241110A (en) * | 2022-02-23 | 2022-03-25 | 北京邮电大学 | Point cloud semantic uncertainty sensing method based on neighborhood aggregation Monte Carlo inactivation |
| CN115273081A (en) * | 2022-08-16 | 2022-11-01 | 西安交通大学 | Point cloud semantic segmentation method and system based on self-adaptive feature sampling |
| CN115540752A (en) * | 2022-09-21 | 2022-12-30 | 中建三局第一建设工程有限责任公司 | Engineering quality allowable deviation detection system based on laser point cloud and construction method |
| CN115496792A (en) * | 2022-09-22 | 2022-12-20 | 北京航空航天大学杭州创新研究院 | Point cloud semantic SLAM method, system and device based on human-in-the-loop optimization |
| 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 |
| CN116524960A (en) * | 2023-05-08 | 2023-08-01 | 华东理工大学 | Speech emotion recognition system based on mixed entropy downsampling and integrated classifier |
| CN116630623A (en) * | 2023-05-19 | 2023-08-22 | 上海交通大学宁波人工智能研究院 | Industrial scene-oriented workpiece point cloud instance segmentation method |
| CN116682021A (en) * | 2023-05-22 | 2023-09-01 | 中国地质大学（武汉） | High-resolution remote sensing image building vector outline data extraction method |
Non-Patent Citations (2)
| Title |
|---|
| Quantifying the unknown impact of segmentation uncertainty on image-based simulations;Michael C. Krygier et al.;《Nature Communications 》;20210930;全文 * |
| 基于熵权法-CIM 模型的高速公路施工临近房屋安全风险评价;周炜 等;《中国安全生产科学技术》;20170331;第13卷(第3期);175-179 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117496309A (en) | 2024-02-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105809672A (en) | Super pixels and structure constraint based image's multiple targets synchronous segmentation method | |
| CN112149962B (en) | Risk quantitative assessment method and system for construction accident cause behaviors | |
| CN110415260B (en) | Smoke image segmentation and identification method based on dictionary and BP neural network | |
| CN111401246A (en) | Smoke concentration detection method, device, equipment and storage medium | |
| Liu et al. | Estimating image segmentation difficulty | |
| CN114449343A (en) | Video processing method, device, equipment and storage medium | |
| CN114972759A (en) | Remote sensing image semantic segmentation method based on hierarchical contour cost function | |
| CN113378905B (en) | Small target detection method based on distribution distance | |
| CN117131348B (en) | Data quality analysis method and system based on differential convolution characteristics | |
| CN117078670B (en) | Production control system of cloud photo frame | |
| CN117496309B (en) | Building scene point cloud segmentation uncertainty evaluation method and system and electronic equipment | |
| CN107871315B (en) | Video image motion detection method and device | |
| CN114219936A (en) | Object detection method, electronic device, storage medium, and computer program product | |
| CN117152094A (en) | Method, device and system for analyzing surface defects of steel plate based on computer vision | |
| WO2024021350A1 (en) | Image recognition model training method and apparatus, computer device, and storage medium | |
| CN114492540B (en) | Training method and device of target detection model, computer equipment and storage medium | |
| CN113836617B (en) | Method and system for analogically predicting front fissure of tunneling surface based on time sequence model | |
| CN112133100B (en) | Vehicle detection method based on R-CNN | |
| CN113496159B (en) | Multi-scale convolution and dynamic weight cost function smoke target segmentation method | |
| CN114510873A (en) | Petroleum logging prediction method and device based on big data | |
| CN111914751A (en) | Image crowd density identification and detection method and system | |
| CN112200831A (en) | Dense connection twin neural network target tracking method based on dynamic template | |
| CN114842330B (en) | Multi-scale background perception pooling weak supervision building extraction method | |
| CN116630814B (en) | Quick positioning and evaluating method for building disasters based on machine learning | |
| CN117237339B (en) | Ground screw punching point position selection method and system based on image processing |
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 |