A semantic segmentation method and system for complex slope point clouds based on optical information

Abstract

This invention discloses a semantic segmentation method and system for point clouds of complex slopes based on optical information, comprising: acquiring multiple representative point cloud data corresponding to vegetation and rock mass areas as sample data; calculating various vegetation indices corresponding to each sample data based on the corresponding band attribute data; performing statistical analysis on each vegetation index to determine the semantic segmentation thresholds for rock mass and vegetation under different vegetation indices; performing correlation analysis on all sample data to determine the correlation coefficients between rock mass and vegetation under different vegetation indices, and configuring the index weights of each vegetation index according to the correlation coefficients; acquiring point cloud data of the area to be segmented; calculating various vegetation indices corresponding to each point cloud in the area to be segmented, comparing the obtained vegetation indices with the corresponding semantic segmentation thresholds to determine the semantic segmentation results of each point cloud under different vegetation indices, and determining the final semantic segmentation result of each point cloud by combining the index weights corresponding to each vegetation index.

Description

Technical Field

[0001] This invention relates to the technical field of airborne LIDAR point cloud data processing, specifically to a semantic segmentation method and system for complex slope point clouds based on optical information. Background Technology

[0002] Existing technologies for ground vegetation segmentation are relatively mature, typically using elevation difference information. However, due to the complexity of elevation information on complex facades or slopes, elevation difference information alone cannot effectively segment these surfaces, thus presenting certain limitations. Using machine learning methods for segmentation requires building models from a large amount of sample data, and the internal structure of these models is complex, difficult to describe directly with formulas or rules, and thus has weak interpretability. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention proposes a semantic segmentation method and system for complex slope point clouds based on optical information. This method can perform effective, reliable, and adaptive semantic segmentation of complex slopes and facades without requiring a large amount of sample data. The specific technical solution is as follows:

[0004] In a first aspect, a semantic segmentation method for complex slope point clouds based on optical information is provided. In a first implementable manner of the first aspect, it includes:

[0005] Multiple representative point cloud data corresponding to vegetation and rock mass areas were obtained as sample data. The point cloud data included optical information of the vegetation and rock mass areas, including R, G, and B band attribute data.

[0006] Calculate the vegetation indices corresponding to each sample data based on the corresponding R, G, and B band attribute data;

[0007] Statistical analysis was performed on each vegetation index to determine the semantic segmentation threshold between rock mass and vegetation under different vegetation indices.

[0008] Correlation analysis was performed on all the sample data to determine the correlation coefficient between the rock mass and vegetation under different vegetation indices, and the index weights corresponding to each vegetation index were configured according to the corresponding correlation coefficients.

[0009] Acquire point cloud data and visible light image data corresponding to the areas to be segmented on complex slopes;

[0010] The vegetation indices corresponding to each point cloud in the region to be segmented are calculated using the R, G, and B band attribute data of visible light image data. The obtained vegetation indices are compared with the corresponding semantic segmentation thresholds to determine the semantic segmentation results of each point cloud under different vegetation indices. The final semantic segmentation results of each point cloud are determined by weighted voting method, taking into account the index weights corresponding to each vegetation index.

[0011] In conjunction with the first possible implementation of the first aspect, in the second possible implementation of the first aspect, the vegetation indices include: EXR, EXG, EXB, EXGR, NGRDI, MGRVI, GLI, RGBVI, EGRBDI, and VARI.

[0012] In conjunction with the first feasible method of the first aspect, in the third feasible method of the first aspect, the semantic segmentation threshold corresponding to the vegetation index is determined by the bimodal method of frequency distribution histogram.

[0013] In conjunction with the first feasible method of the first aspect, in the fourth feasible method of the first aspect, the Pearson correlation coefficient algorithm is used to calculate the correlation coefficient between the rock mass and vegetation under different vegetation indices.

[0014] In conjunction with the first implementable method of the first aspect, in the fifth implementable method of the first aspect, determining the final semantic segmentation result of the point cloud includes:

[0015] Based on the semantic segmentation results of the point cloud under different vegetation indices, and the index weights corresponding to each vegetation index, the vegetation attribution index and rock mass attribution index corresponding to the point cloud are calculated respectively.

[0016] Compare the vegetation attribution index and rock mass attribution index corresponding to the point cloud;

[0017] The point cloud is considered to be vegetation when the vegetation attribution index is greater than the rock mass attribution index.

[0018] In response to the vegetation attribution index being less than the rock mass attribution index, the point cloud is a rock mass;

[0019] In response to the vegetation attribution index being equal to the rock mass attribution index, the point cloud is the semantic segmentation result under the vegetation index corresponding to the maximum value of the index weight.

[0020] Secondly, a semantic segmentation system for complex slope point clouds based on optical information is provided. In a first implementable method of this second aspect, it includes:

[0021] The sample selection module is configured to acquire multiple representative point cloud data corresponding to vegetation and rock mass areas as sample data. The point cloud data includes optical information of the vegetation and rock mass areas, and the optical information includes R, G, and B band attribute data.

[0022] The index calculation module is configured to calculate the vegetation indices corresponding to each sample data based on the corresponding R, G, and B band attribute data.

[0023] The statistical analysis module is configured to perform statistical analysis on each vegetation index to determine the semantic segmentation threshold between rock mass and vegetation under different vegetation indices.

[0024] The weight allocation module is configured to perform correlation analysis based on all the sample data, determine the correlation coefficient between the rock mass and vegetation under different vegetation indices, and configure the index weights corresponding to each vegetation index according to the corresponding correlation coefficients.

[0025] The data acquisition module is configured to acquire point cloud data and visible light image data corresponding to the area to be segmented on a complex slope.

[0026] The semantic segmentation module is configured to calculate the vegetation indices corresponding to each point cloud in the region to be segmented using the R, G, and B band attribute data of the visible light image data, compare the obtained vegetation indices with the corresponding semantic segmentation thresholds, determine the semantic segmentation results of each point cloud under different vegetation indices, and combine the index weights corresponding to each vegetation index to determine the final semantic segmentation results of each point cloud using a weighted voting method.

[0027] In conjunction with the first possible implementation of the second aspect, in the second possible implementation of the second aspect, the statistical analysis module includes:

[0028] The threshold determination unit is configured to determine the semantic segmentation threshold corresponding to the vegetation index using the bimodal method of frequency distribution histogram.

[0029] In conjunction with the first possible implementation of the second aspect, in the third possible implementation of the second aspect, the weight allocation module includes:

[0030] The correlation coefficient calculation unit is configured to use the Pearson correlation coefficient algorithm to calculate the correlation coefficient between rock mass and vegetation under different vegetation indices.

[0031] In conjunction with the first possible implementation of the second aspect, in the fourth possible implementation of the second aspect, the semantic segmentation module includes:

[0032] The index calculation unit is configured to calculate the vegetation attribution index and rock mass attribution index corresponding to the point cloud based on the semantic segmentation results of the point cloud under different vegetation indices and the index weights corresponding to each vegetation index.

[0033] The semantic segmentation unit is configured to compare the vegetation attribution index and the rock mass attribution index corresponding to the point cloud;

[0034] The point cloud is considered to be vegetation when the vegetation attribution index is greater than the rock mass attribution index.

[0035] In response to the vegetation attribution index being less than the rock mass attribution index, the point cloud is a rock mass;

[0036] In response to the vegetation attribution index being equal to the rock mass attribution index, the point cloud is the semantic segmentation result under the vegetation index corresponding to the maximum value of the index weight.

[0037] Beneficial Effects: The semantic segmentation method and system for complex slope point clouds based on optical information of this invention can determine the accurate segmentation thresholds for rock mass and vegetation under each vegetation index by statistically analyzing multiple sample data. Using the correlation coefficients of different object categories under each vegetation index as a metric, combined with a weighted voting decision-making method, the category of each point cloud is automatically determined, completing adaptive semantic segmentation of the point cloud data. Compared to traditional elevation threshold segmentation, it has higher accuracy and can effectively perform semantic segmentation of point clouds on complex facades and slopes. Compared to machine learning methods, it requires fewer samples and eliminates the need for cumbersome dataset optimization and model parameter tuning. The computation process is highly interpretable, and the results are more reliable. Attached Figure Description

[0038] To more clearly illustrate the specific embodiments of the present invention, the accompanying drawings used in the specific embodiments will be briefly described below. In all the drawings, the elements or parts are not necessarily drawn to scale.

[0039] Figure 1 A flowchart illustrating a semantic segmentation method for complex slope point clouds based on optical information provided in an embodiment of the present invention;

[0040] Figure 2 This is a system block diagram of a semantic segmentation system for complex slope point clouds based on optical information, provided in an embodiment of the present invention.

[0041] Figure 3 A schematic diagram showing the semantic segmentation thresholds corresponding to 10 vegetation indices;

[0042] Figure 4 Pearson correlation coefficients between vegetation and rock mass under 10 vegetation indices;

[0043] Figure 5 This is a schematic diagram showing the index weights corresponding to the 10 vegetation indices. Detailed Implementation

[0044] The embodiments of the technical solution of the present invention will now be described in detail with reference to the accompanying drawings. These embodiments are merely illustrative of the technical solution of the present invention and are therefore intended to limit the scope of protection of the present invention.

[0045] like Figure 1 The flowchart shown is for a semantic segmentation method of complex slope point clouds based on optical information. This segmentation method includes:

[0046] Step 1: Obtain multiple representative point cloud data corresponding to vegetation and rock mass areas as sample data. The point cloud data includes optical information of the vegetation and rock mass areas, including R, G, and B band attribute data.

[0047] Step 2: Calculate the vegetation indices corresponding to each sample data based on the corresponding R, G, and B band attribute data;

[0048] Step 3: Perform statistical analysis on each vegetation index to determine the semantic segmentation threshold of rock mass and vegetation under different vegetation indices.

[0049] Step 4: Perform correlation analysis based on all the sample data to determine the correlation coefficient between the rock mass and vegetation under different vegetation indices, and configure the index weights corresponding to each vegetation index according to the corresponding correlation coefficients.

[0050] Step 5: Obtain point cloud data and visible light image data corresponding to the complex slope area to be segmented;

[0051] Step 6: Calculate the vegetation indices corresponding to each point cloud in the region to be segmented using the R, G, and B band attribute data of the visible light image data, and compare the obtained vegetation indices with the corresponding semantic segmentation thresholds to determine the semantic segmentation results of each point cloud under different vegetation indices. Combine the index weights corresponding to each vegetation index and use the weighted voting method to determine the final semantic segmentation results of each point cloud.

[0052] Specifically, firstly, a drone equipped with a lidar system can be used to collect multiple point cloud images and visible light images of vegetation and rock formations. These images are then imported into DJI TERRA software for point cloud data processing, yielding multiple point cloud datasets containing R, G, and B band attribute data as sample data. Then, based on the R, G, and B band attribute data in each sample dataset, appropriate algorithms can be used to calculate multiple vegetation indices corresponding to each sample dataset.

[0053] Subsequently, statistical analysis can be performed on each vegetation index to determine the semantic segmentation thresholds for rock mass and vegetation under different vegetation indices. Then, by analyzing the correlation coefficients between vegetation and rock mass under different vegetation indices, the index weights corresponding to each vegetation index can be determined, providing a foundation for subsequent adaptive segmentation of point cloud semantics.

[0054] Next, point cloud data of the area to be segmented can be collected using a drone equipped with a LiDAR system. Finally, based on the R, G, and B band attribute data of each point cloud in the segmented area's point cloud data, different vegetation indices corresponding to the point clouds to be segmented are calculated. These indices are then compared with corresponding semantic segmentation thresholds to determine the semantic segmentation results of the point clouds under different vegetation indices. Finally, based on the index weights corresponding to each vegetation index, the semantic segmentation results of the point clouds under different vegetation indices are fused to obtain the final semantic segmentation result of the point clouds to be segmented.

[0055] Compared to traditional elevation thresholding segmentation, it achieves higher accuracy and can effectively perform point cloud semantic segmentation on complex facades and slopes. Compared to machine learning methods, it requires fewer samples and eliminates the need for cumbersome dataset optimization and model parameter tuning. The computation process is highly interpretable, and the results are more reliable.

[0056] In this embodiment, when acquiring sample data, the point cloud data can be screened manually to ensure the purity of the two types of sample data, namely vegetation and rock mass, and to make the number of the two types of samples relatively balanced.

[0057] In this embodiment, optionally, the vegetation indices include: EXR, EXG, EXB, EXGR, NGRDI, MGRVI, GLI, RGBVI, EGRBDI, and VARI. The specific calculation formulas for these vegetation indices are shown in the table below:

[0058] Vegetation index Formula EXR 1.4*B-G EXG 2*G-R-B EXB 1.4R-G EXGR (2*G-R-B)-(1.4*B-G) NGRDI (G-R) / (G+R) MGRVI (G2-R2) / (G2+R2) GLI (2*G-R-B) / (2*G+R+B) RGBVI (G2-R*B) / (G2+R*B) EGRBDI ((2G)2-R-B) / ((2G)2+R*B) VARI (G-R) / (G+R-B)

[0059] Where R is the reflectance of the ground feature in the red band, G is the reflectance of the ground feature in the green band, and B is the reflectance of the ground feature in the blue band.

[0060] In this embodiment, optionally, in step 3, the semantic segmentation threshold corresponding to the vegetation index can be determined by the bimodal method of frequency distribution histogram.

[0061] Specifically, such as Figure 3 As shown, when determining the semantic segmentation thresholds for vegetation and rock mass under a vegetation index, statistical analysis of all sample data can be performed to determine the frequency distribution of vegetation samples belonging to vegetation and rock mass samples belonging to rock mass under that vegetation index. Normal distribution curves corresponding to the vegetation samples and rock mass samples are then fitted based on the frequency distributions to quantify the bimodal distribution characteristics of the data. Finally, the intersection of the normal distribution curves corresponding to the vegetation samples and the rock mass samples is used as the semantic segmentation threshold for rock mass and vegetation under that vegetation index. In this way, the accurate segmentation thresholds for rock mass and vegetation under each vegetation index can be determined, thereby ensuring the accuracy of the point cloud semantic segmentation results.

[0062] It should be understood that this embodiment is only used as an example of the frequency distribution histogram bimodal method, but the present invention is not limited to this. Other existing methods can also be used to determine the semantic segmentation threshold corresponding to the vegetation index, such as the Otsu method or the watershed method.

[0063] In this embodiment, optionally, in step 4, as follows: Figure 4 As shown, the correlation coefficient between rock mass and vegetation under a given vegetation index can be calculated using the Pearson correlation coefficient algorithm based on the vegetation indices corresponding to the rock mass and vegetation samples. The specific calculation formula is as follows:

[0064]

[0065] Where r is the Pearson correlation coefficient, x i y represents the vegetation index value of the rock mass sample. i The vegetation index value is the vegetation index value of the vegetation sample. This represents the average vegetation index of the rock mass samples. is the average vegetation index of the vegetation samples, and n is the number of samples.

[0066] The smaller the correlation coefficient, the weaker the linear relationship between vegetation and rock mass under that vegetation index, and the more accurately vegetation and rock mass can be distinguished. Therefore, the index weight corresponding to this vegetation index should also be higher. The obtained index weights for each vegetation index are as follows: Figure 5 As shown.

[0067] After determining the semantic segmentation thresholds and index weights corresponding to each vegetation index, point cloud data of the region to be segmented can be obtained. Based on the R, G, and B band attribute data in the point cloud data, the corresponding vegetation indices corresponding to the point cloud to be segmented are calculated using the aforementioned formulas. By comparing each vegetation index of the point cloud to be segmented with the corresponding semantic segmentation thresholds, and according to the index weights corresponding to each vegetation index, the semantic segmentation results of the point cloud to be segmented under different vegetation indices are fused to obtain the final semantic segmentation result.

[0068] In this embodiment, optionally, in step 6, determining the final semantic segmentation result of the point cloud includes:

[0069] Based on the semantic segmentation results of the point cloud under different vegetation indices, and the index weights corresponding to each vegetation index, the vegetation attribution index and rock mass attribution index corresponding to the point cloud are calculated respectively.

[0070] Compare the vegetation attribution index and rock mass attribution index corresponding to the point cloud;

[0071] The point cloud is considered to be vegetation when the vegetation attribution index is greater than the rock mass attribution index.

[0072] In response to the vegetation attribution index being less than the rock mass attribution index, the point cloud is a rock mass;

[0073] In response to the vegetation attribution index being equal to the rock mass attribution index, the point cloud is the semantic segmentation result under the vegetation index corresponding to the maximum value of the index weight.

[0074] Specifically, firstly, based on the index weights corresponding to each vegetation index, and according to the semantic segmentation results of the point cloud to be segmented under different vegetation indices, the vegetation attribution index and rock mass attribution index corresponding to the point cloud to be segmented can be calculated. The specific calculation formula is as follows:

[0075]

[0076] Among them, VI kj Let T be the index value corresponding to the j-th vegetation index of the point cloud k to be segmented. j W is the semantic segmentation threshold corresponding to the j-th vegetation index. j WV represents the index weight corresponding to the j-th vegetation index. k WR represents the vegetation ownership index of the point cloud k to be segmented. k The rock mass attribution index is the point cloud k to be segmented.

[0077] Then, the vegetation ownership index WV of the point cloud k to be segmented is... k Rock mass attribution index WR k In comparison, if WV k >WR k If WV k <WR k If the point cloud k to be segmented is the rock mass, then WV k =WR k Then, from all semantic segmentation results S of the point cloud k to be segmented kj In the process, the semantic segmentation result of the point cloud k to be segmented under the vegetation index corresponding to the maximum exponential weight is selected as the final semantic segmentation result.

[0078] like Figure 2 The diagram shown is a system block diagram of a semantic segmentation system for complex slope point clouds based on optical information. The segmentation system includes:

[0079] The sample selection module is configured to acquire multiple representative point cloud data corresponding to vegetation and rock mass areas as sample data. The point cloud data includes optical information of the vegetation and rock mass areas, and the optical information includes R, G, and B band attribute data.

[0080] The index calculation module is configured to calculate the vegetation indices corresponding to each sample data based on the corresponding R, G, and B band attribute data.

[0081] The statistical analysis module is configured to perform statistical analysis on each vegetation index to determine the semantic segmentation threshold between rock mass and vegetation under different vegetation indices.

[0082] The weight allocation module is configured to perform correlation analysis based on all the sample data, determine the correlation coefficient between the rock mass and vegetation under different vegetation indices, and configure the index weights corresponding to each vegetation index according to the corresponding correlation coefficients.

[0083] The data acquisition module is configured to acquire point cloud data and visible light image data corresponding to the area to be segmented on a complex slope.

[0084] The semantic segmentation module is configured to calculate the vegetation indices corresponding to each point cloud in the region to be segmented using the R, G, and B band attribute data of the visible light image data, compare the obtained vegetation indices with the corresponding semantic segmentation thresholds, determine the semantic segmentation results of each point cloud under different vegetation indices, and combine the index weights corresponding to each vegetation index to determine the final semantic segmentation results of each point cloud using a weighted voting method.

[0085] Specifically, the segmentation system consists of a sample selection module, an index calculation module, a statistical analysis module, a weight allocation module, a data acquisition module, and a semantic segmentation module. The sample selection module utilizes a UAV equipped with a LiDAR system to acquire multiple point cloud images and visible light images of vegetation and rock formations. These images are then imported into DJI TERRA software for point cloud data processing, yielding multiple sets of point cloud data containing R, G, and B band attribute data as sample data. The index calculation module uses appropriate algorithms to calculate multiple vegetation indices corresponding to each sample data set based on the R, G, and B band attribute data.

[0086] The statistical analysis module can perform statistical analysis on each vegetation index to determine the semantic segmentation thresholds for rock mass and vegetation under different vegetation indices. The weight allocation module can determine the index weights corresponding to each vegetation index by analyzing the correlation coefficients between vegetation and rock mass under different vegetation indices, providing a foundation for subsequent adaptive segmentation of point cloud semantics.

[0087] The data acquisition module can collect point cloud data of the area to be segmented using a drone equipped with a LiDAR system. The semantic segmentation module calculates different vegetation indices for each point cloud in the segmentation data based on its R, G, and B band attribute data. These indices are then compared with corresponding semantic segmentation thresholds to determine the semantic segmentation results for each vegetation index. Finally, based on the index weights, the semantic segmentation results for each vegetation index are fused to obtain the final semantic segmentation result for the point cloud.

[0088] In this embodiment, optionally, the statistical analysis module includes:

[0089] The threshold determination unit is configured to determine the semantic segmentation threshold corresponding to the vegetation index using the bimodal method of frequency distribution histogram.

[0090] Specifically, the statistical analysis module includes a threshold determination unit. This unit performs statistical analysis on all sample data to determine the frequency distribution of vegetation samples (belonging to vegetation) and rock mass samples (belonging to rock mass) under the given vegetation index. It then fits normal distribution curves for both vegetation and rock mass samples based on their frequency distributions to quantify the bimodal distribution characteristics of the data. Finally, the intersection of the normal distribution curves for vegetation and rock mass samples is used as the semantic segmentation threshold for rock mass and vegetation under that vegetation index. In this way, the accurate segmentation threshold for rock mass and vegetation under each vegetation index can be determined, ensuring the accuracy of the point cloud semantic segmentation results.

[0091] In this embodiment, optionally, the weight allocation module includes:

[0092] The correlation coefficient calculation unit is configured to use the Pearson correlation coefficient algorithm to calculate the correlation coefficient between rock mass and vegetation under different vegetation indices.

[0093] Specifically, the weight allocation module is equipped with a correlation coefficient calculation unit. This correlation coefficient calculation unit can calculate the correlation coefficient between the rock mass and vegetation under different vegetation indices based on the vegetation indices corresponding to all sample data using the Pearson correlation coefficient algorithm, and configure the index weight of each vegetation index accordingly.

[0094] In this embodiment, optionally, the semantic segmentation module includes:

[0095] The index calculation unit is configured to calculate the vegetation attribution index and rock mass attribution index corresponding to the point cloud based on the semantic segmentation results of the point cloud under different vegetation indices and the index weights corresponding to each vegetation index.

[0096] The semantic segmentation unit is configured to compare the vegetation attribution index and the rock mass attribution index corresponding to the point cloud;

[0097] The point cloud is considered to be vegetation when the vegetation attribution index is greater than the rock mass attribution index.

[0098] In response to the vegetation attribution index being less than the rock mass attribution index, the point cloud is a rock mass;

[0099] In response to the vegetation attribution index being equal to the rock mass attribution index, the point cloud is the semantic segmentation result under the vegetation index corresponding to the maximum value of the index weight.

[0100] Specifically, the semantic segmentation unit includes an index calculation unit and a semantic segmentation unit. The index calculation unit can calculate the vegetation attribution index and rock mass attribution index of the point cloud to be segmented based on the index weights corresponding to each vegetation index and the semantic segmentation results of the point cloud to be segmented under different vegetation indices.

[0101] The semantic segmentation unit compares the vegetation attribution index and the rock mass attribution index of the point cloud to be segmented. If the vegetation attribution index is greater than the rock mass attribution index, the point cloud to be segmented is vegetation. If the vegetation attribution index is less than the rock mass attribution index, the point cloud to be segmented is rock mass. If the vegetation attribution index is equal to the rock mass attribution index, the semantic segmentation result of the point cloud to be segmented under the vegetation index with the highest index weight is selected from all semantic segmentation results of the point cloud to be segmented as the final semantic segmentation result. In this way, effective semantic segmentation of complex slopes and facades can be performed with a small amount of sample data, resulting in more reliable semantic segmentation results.

[0102] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. A semantic segmentation method for point clouds of complex slopes based on optical information, characterized in that, include: Multiple representative point cloud data corresponding to vegetation and rock mass areas were obtained as sample data. The point cloud data included optical information of the vegetation and rock mass areas, including R, G, and B band attribute data. Calculate the vegetation indices corresponding to each sample data based on the corresponding R, G, and B band attribute data; Statistical analysis was performed on each vegetation index to determine the semantic segmentation threshold between rock mass and vegetation under different vegetation indices. Correlation analysis was performed on all the sample data to determine the correlation coefficient between the rock mass and vegetation under different vegetation indices, and the index weights corresponding to each vegetation index were configured according to the corresponding correlation coefficients. Acquire point cloud data and visible light image data corresponding to the areas to be segmented on complex slopes; The vegetation indices corresponding to each point cloud in the region to be segmented are calculated using R, G, and B band attribute data from visible light imagery. These vegetation indices are then compared with corresponding semantic segmentation thresholds to determine the semantic segmentation results for each point cloud under different vegetation indices. Finally, combining the index weights of each vegetation index, a weighted voting method is used to determine the final semantic segmentation results for each point cloud, including: Based on the semantic segmentation results of the point cloud under different vegetation indices, and the index weights corresponding to each vegetation index, the vegetation attribution index and rock mass attribution index corresponding to the point cloud are calculated respectively. Compare the vegetation attribution index and rock mass attribution index corresponding to the point cloud; In response to the vegetation attribution index being greater than the rock mass attribution index, the point cloud is vegetation; In response to the vegetation attribution index being less than the rock mass attribution index, the point cloud is a rock mass; In response to the vegetation attribution index being equal to the rock mass attribution index, the point cloud is the semantic segmentation result under the vegetation index corresponding to the maximum value of the index weight.

2. The semantic segmentation method for complex slope point clouds based on optical information according to claim 1, characterized in that, The vegetation indices include: EXR, EXG, EXB, EXGR, NGRDI, MGRVI, GLI, RGBVI, EGRBDI, and VARI.

3. The semantic segmentation method for complex slope point clouds based on optical information according to claim 1, characterized in that, The semantic segmentation threshold corresponding to the vegetation index was determined using the bimodal method of frequency distribution histogram.

4. The semantic segmentation method for complex slope point clouds based on optical information according to claim 1, characterized in that, The correlation coefficient between rock mass and vegetation under different vegetation indices was calculated using the Pearson correlation coefficient algorithm.

5. A semantic segmentation system for point clouds of complex slopes based on optical information, characterized in that, include: The sample selection module is configured to acquire multiple representative point cloud data corresponding to vegetation and rock mass areas as sample data. The point cloud data includes optical information of the vegetation and rock mass areas, and the optical information includes R, G, and B band attribute data. The index calculation module is configured to calculate the vegetation indices corresponding to each sample data based on the corresponding R, G, and B band attribute data. The statistical analysis module is configured to perform statistical analysis on each vegetation index to determine the semantic segmentation threshold between rock mass and vegetation under different vegetation indices. The weight allocation module is configured to perform correlation analysis based on all the sample data, determine the correlation coefficient between the rock mass and vegetation under different vegetation indices, and configure the index weights corresponding to each vegetation index according to the corresponding correlation coefficients. The data acquisition module is configured to acquire point cloud data and visible light image data corresponding to the area to be segmented on a complex slope. A semantic segmentation module is configured to calculate vegetation indices corresponding to each point cloud in the region to be segmented using R, G, and B band attribute data of visible light image data, compare the obtained vegetation indices with the corresponding semantic segmentation thresholds to determine the semantic segmentation results of each point cloud under different vegetation indices, and determine the final semantic segmentation results of each point cloud using a weighted voting method based on the index weights corresponding to each vegetation index. The semantic segmentation module includes: The index calculation unit is configured to calculate the vegetation attribution index and rock mass attribution index corresponding to the point cloud based on the semantic segmentation results of the point cloud under different vegetation indices and the index weights corresponding to each vegetation index. The semantic segmentation unit is configured to compare the vegetation attribution index and the rock mass attribution index corresponding to the point cloud; In response to the vegetation attribution index being greater than the rock mass attribution index, the point cloud is vegetation; In response to the vegetation attribution index being less than the rock mass attribution index, the point cloud is a rock mass; In response to the vegetation attribution index being equal to the rock mass attribution index, the point cloud is the semantic segmentation result under the vegetation index corresponding to the maximum value of the index weight.

6. The semantic segmentation system for complex slope point clouds based on optical information according to claim 5, characterized in that, The statistical analysis module includes: The threshold determination unit is configured to determine the semantic segmentation threshold corresponding to the vegetation index using the bimodal method of frequency distribution histogram.

7. The semantic segmentation system for complex slope point clouds based on optical information according to claim 5, characterized in that, The weight allocation module includes: The correlation coefficient calculation unit is configured to use the Pearson correlation coefficient algorithm to calculate the correlation coefficient between rock mass and vegetation under different vegetation indices.

Images