A forest landscape visual quality assessment method and system based on point cloud data and machine learning

Abstract

The present application is suitable for the technical field of forest landscape evaluation, and provides a forest landscape visual quality evaluation method and system based on point cloud data and machine learning, which comprises the following steps: obtaining three-dimensional point cloud data of a forest land by using a laser radar, and performing data preprocessing based on the obtained three-dimensional point cloud data; extracting landscape 3D measurement features based on the preprocessed data; constructing a landscape visual quality evaluation system based on the extracted landscape 3D measurement features, determining the visual quality evaluation score or visual quality evaluation grade corresponding to each landscape 3D measurement feature, and determining a sample data set of machine learning based on the visual quality evaluation score or visual quality evaluation grade; training a preset landscape visual quality evaluation model by using the sample data set, and evaluating the visual quality of a forest landscape by using the trained landscape visual quality evaluation model on any new three-dimensional point cloud data, thereby solving the demand for high efficiency and high precision in the evaluation of the visual quality of a forest landscape.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of forest landscape evaluation, in particular to a forest landscape visual quality evaluation method and system based on point cloud data and machine learning. BACKGROUND

[0002] With the rise of ecotourism and the importance of forest management, the visual quality evaluation of forest landscape has been increasingly concerned. Forest landscape refers to the visual space created by forests and the surrounding environment. Landscape aesthetics is reflected in the perception of landscape by people in terms of vision, hearing, taste, touch and smell, among which vision plays a dominant role. Visual quality is the most important factor affecting landscape aesthetics, and is also the ultimate goal of the improvement and utilization of forest ecosystems. Landscape visual quality evaluation helps to guide forest landscape configuration, which is of great significance to improve the structural stability and ecological diversity of forest ecosystems.

[0003] Landscape visual quality is the state of landscape space, function and visual structure within a given time, which contains two meanings: landscape visual features and human visual perception. The evaluation of landscape visual features focuses more on the intrinsic properties of the landscape. Forest landscape visual quality is closely related to forest feature properties, and the preference for viewing is affected by forest quality or structure, such as tree height, diameter at breast height, stand density, configuration mode, etc. However, forest landscape physical properties usually use 2D, 2.5D indicators based on photos and videos, such as 2D landscape evaluation models based on GIS and 2.5D terrain properties extracted from digital elevation models. These indicators usually evaluate from the top-down perspective, which cannot reflect the real scene seen or experienced by users with human eyes horizontally in a given landscape, and cannot meet the high demand for high efficiency and high precision in forest landscape visual quality evaluation.

[0004] Therefore, there is a need for a forest landscape visual quality evaluation method and system based on point cloud data and machine learning. SUMMARY

[0005] The embodiments of the present application provide a forest landscape visual quality evaluation method and system based on point cloud data and machine learning, which solves the problem that the prior art cannot meet the high demand for high efficiency and high precision in forest landscape visual quality evaluation.

[0006] The first aspect of the embodiments of the present application provides a forest landscape visual quality evaluation method based on point cloud data and machine learning, comprising:

[0007] Obtaining three-dimensional point cloud data of the forest land by using a laser radar, and performing data preprocessing based on the obtained three-dimensional point cloud data;

[0008] Extracting landscape 3D measurement features based on the preprocessed data;

[0009] construct a landscape visual quality evaluation system based on the extracted landscape 3D metric features, determine the visual quality evaluation scores or visual quality evaluation levels corresponding to each landscape 3D metric feature according to the landscape visual quality evaluation system, and determine a sample data set of machine learning based on the visual quality evaluation scores or visual quality evaluation levels;

[0010] train a preset landscape visual quality evaluation model using the sample data set, and perform forest landscape visual quality evaluation on any new three-dimensional point cloud data using the trained landscape visual quality evaluation model.

[0011] Further, the landscape 3D metric features are extracted based on the preprocessed data, including:

[0012] normalize and voxelize the preprocessed data using a PointNet model to obtain global feature information of the point cloud;

[0013] input the voxelized data into a convolutional neural network to extract local feature information of the voxel data;

[0014] fuse the global feature information of the point cloud and the local feature information of the voxel data to obtain a fusion feature vector;

[0015] calculate the landscape 3D metric features based on the fusion feature vector.

[0016] Further, the global feature information of the point cloud and the local feature information of the voxel data are fused to obtain a fusion feature vector, and the expressions are respectively:

[0017]

[0018] G f =σ(G s W f +b f )

[0019]

[0020] wherein F fusion is a fusion feature vector, G f is a transformed PointNet feature, is a transformed feature of a convolutional neural network, σ is an activation function, G s is a key global feature, W f is a weight matrix of a full connection layer, and b f is a bias vector, is a feature vector converted by global average pooling, and h and w are the height and width of the feature map, is the feature value of the jth feature map at the (x, y) position.

[0021] Further, the landscape 3D metric feature is calculated based on the fusion feature vector, comprising:

[0022] The landscape 3D metric feature comprises landscape diversity, landscape shape, landscape connectivity, color uniformity, openness and terrain.

[0023] Landscape diversity:

[0024]

[0025] wherein: HVDD is a diversity index, N is the number of groups, V i is the volume of the ith group, is the average volume of all groups;

[0026] Landscape shape:

[0027]

[0028] wherein: MSI is a shape index, SVR is a surface-volume ratio, S is the actual surface area of the landscape, S min is the minimum surface area of the landscape of the same size, V is the volume of the landscape;

[0029] Landscape connectivity:

[0030] NC = M

[0031]

[0032] wherein: NC is the number of clusters, CD is the cluster density, CS is the cluster size, CDist is the cluster distance, M is the number of clusters obtained after statistical clustering, H a is the horizontal area of the landscape, V h is the maximum height of the landscape, |C i is the number of voxels contained in each cluster C i , d(C i , C j ) is the distance between the ith cluster and the jth cluster centroid voxel.

[0033] Color uniformity:

[0034]

[0035] wherein: CD v is the color distance, Dave is the average color distance, C v is the color variation index, v r , v g and vb respectively, are the values of the red, green, and blue color channels of a voxel, N v is the total number of voxels;

[0036] Openness:

[0037]

[0038] where: VD is the visual distance, K is the number of groups, d min (g k ) is the minimum distance between the viewpoint of the kth group and any voxel within the group;

[0039] Terrain:

[0040]

[0041] where: AS is the average slope, GV is the gradient variation, N g is the number of ground grids, s i is the slope of the ith ground grid, g i is the gradient of the ith ground grid, g ave is the average value of all grid gradients.

[0042] Further, the landscape visual quality evaluation system based on the extracted landscape 3D metric features is constructed, and the visual quality evaluation score or visual quality evaluation level corresponding to each landscape 3D metric feature is determined according to the landscape visual quality evaluation system, and the sample data set of machine learning is determined based on the visual quality evaluation score or visual quality evaluation level, including:

[0043] The extracted landscape 3D metric features are analyzed and screened to determine a key feature subset related to landscape visual quality in each landscape 3D metric feature;

[0044] Expert scores of the screened key feature subset are collected, and the visual quality level or visual quality score corresponding to each landscape 3D metric feature is determined;

[0045] The landscape 3D metric feature vector corresponding to the key feature subset screened from each sample is taken as the feature data of the data set, and the score or rating of the landscape visual quality of each landscape 3D metric feature is taken as the label data of the data set, to obtain the sample data set of machine learning.

[0046] Further, the expression of the collected expert scores of the screened key feature subset includes:

[0047]

[0048] wherein: The normalized value of the landscape visual quality score of the ith key feature subset in the landscape 3D metric feature X for the pth subject, The landscape visual quality score of the ith key feature subset in the landscape 3D metric feature X for the pth subject, The average value of the landscape visual quality scores of all subsets in the landscape 3D metric feature X for the pth subject, The standard deviation of the landscape visual quality scores of all subsets in the landscape 3D metric feature X for the pth subject, N p The total number of subjects, The final normalized landscape visual quality value of the ith key feature subset in the landscape 3D metric feature X.

[0049] Further, the expert scores of the screened key feature subsets are collected, and the visual quality grades or visual quality scores corresponding to each landscape 3D metric feature are determined, comprising:

[0050] According to the equidistance method, the landscape visual quality is divided into 5 grades, namely very good, good, general, poor, and very poor:

[0051] Very good: [(LVD max ―LVD min ) × 0.8 + LVD min ,LVD max ]

[0052] Good: [(LVD max ―LVD min ) × 0.6, (LVD max ―LVD min ) × 0.8 + LVD min ]

[0053] General:

[0054] [(LVD max ―LVD min ) × 0.4 + LVD min ,(LVD max ―LVD min ) × 0.6 + LVD min ]

[0055] Poor:

[0056] [(LVD max ―LVD min ) × 0.2 + LVD min ,(LVD max ―LVD min ) × 0.4 + LVD min ]

[0057] Very poor: [LVD min , LVD max -LVD min ) x 0.2 + LVD min ]

[0058] wherein, LVD max represents the maximum landscape visual quality value, and LVD min represents the minimum landscape visual quality value.

[0059] Further, the expert scores of the collected key feature subsets are screened, and the visual quality grades or visual quality scores corresponding to each landscape 3D measurement feature are determined, including:

[0060]

[0061] wherein, S represents the visual quality score, k represents the number of key feature subsets, w i represents the weight of the i-th key feature subset.

[0062] Further, the preset landscape visual quality evaluation model is trained by using the sample data set, and the trained landscape visual quality evaluation model is used to evaluate the forest landscape visual quality of any new three-dimensional point cloud data, including:

[0063] The model learns the mapping relationship between the key feature subsets in each landscape 3D measurement feature and the corresponding landscape 3D measurement feature, and learns the mapping relationship between each landscape 3D measurement feature and the landscape visual quality, and determines the trained landscape visual quality evaluation model according to the learned mapping relationship.

[0064] The second aspect of the embodiment of the application provides a forest landscape visual quality evaluation system based on point cloud data and machine learning, including:

[0065] A data acquisition and preprocessing unit is configured to acquire three-dimensional point cloud data of a forest land by using a laser radar, and perform data preprocessing based on the acquired three-dimensional point cloud data;

[0066] A landscape 3D measurement feature extraction unit is configured to extract landscape 3D measurement features based on the preprocessed data;

[0067] A sample data set determination unit is configured to construct a landscape visual quality evaluation system based on the extracted landscape 3D measurement features, determine visual quality evaluation scores or visual quality evaluation grades corresponding to each landscape 3D measurement feature according to the landscape visual quality evaluation system, and determine a sample data set for machine learning based on the visual quality evaluation scores or visual quality evaluation grades;

[0068] The landscape visual quality evaluation model training unit is configured to train a preset landscape visual quality evaluation model using the sample data set, and to evaluate the forest landscape visual quality of any new three-dimensional point cloud data using the trained landscape visual quality evaluation model.

[0069] From the above technical solutions, the embodiments of the present application have the following advantages:

[0070] The present application combines point cloud data and machine learning, and the evaluation of the extracted 3D landscape metrics can more objectively and truly reflect the landscape visual quality, and the automated evaluation process greatly improves the evaluation efficiency, saving manpower, material resources and time cost; the three-dimensional point cloud data of the forest land is obtained by using the laser radar, and the data preprocessing is performed based on the obtained three-dimensional point cloud data; the landscape 3D metric features are extracted based on the preprocessed data; the landscape visual quality evaluation system is constructed based on the extracted landscape 3D metric features, and the sample data set of machine learning is determined according to the landscape visual quality evaluation system; finally, the preset landscape visual quality evaluation model is trained using the sample data set, and the forest landscape visual quality of any new three-dimensional point cloud data is evaluated using the trained landscape visual quality evaluation model. The model can be continuously optimized according to new data and requirements, adapt to different scene and requirement evaluation, effectively quantify the visual quality of forest landscape, improve the objectivity and efficiency of landscape visual quality evaluation, and provide a new technical means for landscape visual quality evaluation.

[0071] Other advantages, objects, and features of the present application will be set forth in part in the following specification, and in part will become apparent to those skilled in the art from the examination of the following text, or can be learned from the practice of the present application. BRIEF DESCRIPTION OF DRAWINGS

[0072] Figure 1 An embodiment flowchart of a forest landscape visual quality evaluation method based on point cloud data and machine learning in the present application;

[0073] Figure 2 A laser radar scan diagram for collecting forest data in the present application;

[0074] Figure 3 A landscape visual quality evaluation system diagram constructed based on the extracted landscape 3D metric features in the present application. DETAILED DESCRIPTION

[0075] The terms "first", "second", "third", "fourth" and the like in the description and in the claims of the present application, and above-mentioned drawings, if any, are used to distinguish between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the use of the terms so-termed, where appropriate, can be interchanged to allow the embodiments of the present application described herein to be carried out in sequences other than those illustrated or described herein. Moreover, the terms "comprises", "comprising", "corresponds" and any variations thereof, are intended to cover non-exclusive inclusions, for example, a process, method, system, product or apparatus that includes a list of steps or units is not necessarily limited to those steps or units that are clearly listed, but can include other steps or units that are not clearly listed or inherent to such processes, methods, products or apparatuses.

[0076] Embodiment one

[0077] The method implemented in this embodiment can be implemented in a system, which can be implemented in a server or in a terminal, and the specific implementation is not limited. From the perspective of system implementation, the forest landscape visual quality evaluation method based on point cloud data and machine learning in the present application will be introduced. Please refer to Figure 1 The method provided by the embodiment of the present application includes the following steps:

[0078] S11. Obtain three-dimensional point cloud data of the forest land by using a laser radar, and perform data preprocessing based on the obtained three-dimensional point cloud data;

[0079] In this embodiment, three-dimensional point cloud data of the forest land is obtained by laser radar technology, and data preprocessing is performed based on the collected three-dimensional point cloud data. Representative forest areas are selected for data collection, covering different stand types, topography and different age structures, etc., to ensure that the collected data can fully reflect the overall characteristics of the forest land. Specifically, according to the forest land conditions and research needs, appropriate scanning angle range is set, the number of times of scanning per unit time, the wavelength and power of the laser are set, etc. During the data collection process, the running state of the equipment is monitored in real time to ensure that the laser radar normally transmits and receives laser pulse signals, and the data can be accurately recorded into the storage device. The collected point cloud data is preprocessed, including noise reduction, coordinate conversion, etc., to generate a high-quality three-dimensional model. Then, the quantitative structure model constructed by the TreeQSM software is used to extract forest parameters, and a 3D voxel information data frame with leaf area density (LAD) value is created. Please refer to Figure 2 From left to right, the pictures represent the forest point cloud data collected by the laser radar, the model diagram of constructing the quantitative structure model to extract the forest parameters, and the schematic diagram of creating the 3D voxel data frame.

[0080] S12. Extract landscape 3D metric features based on the preprocessed data;

[0081] In this embodiment, step S12 includes the following:

[0082] 1. The PointNet model is used to normalize and voxelize the preprocessed data to obtain the global feature information of the point cloud;

[0083] Let the preprocessed point cloud data be P = {p1, p2, ..., p...} n}, where p i =(x i ,y i ,z i Let represent the three-dimensional coordinates of the i-th point. Calculate the minimum and maximum values ​​on each coordinate axis and normalize the coordinates of each point.

[0084] Determine the size s of the voxel based on the spatial extent of the point cloud [x min ,x max ]×[y min ,y max ]×[z min ,z max The space is divided into a three-dimensional voxel mesh, and for each point p′ i Calculate its voxel index (v x ,v y ,v z ),in Count the number of points N within each voxel. v Average coordinates Properties such as average reflectance intensity are used as characteristic values ​​of voxels.

[0085] The normalized point cloud data P′={p1′,p2′,...,p n The input is fed into PointNet, and the initial feature vector f for each point is... i It can be set to f i =(x′) i ,y′ i ,z′ i ,r′ i (). The input features are processed through multiple MLP layers, each consisting of fully connected neurons. Let the input feature dimension of the j-th MLP layer be... The output feature dimension is The weight matrix is ​​W j The bias vector is b j For the eigenvector of the i-th point The output calculation formula for the j-th layer MLP is: Where σ is the activation function. After passing through multiple MLP layers, the feature vector of each point is obtained. A global feature vector G is obtained by a max-pooling operation, and the kth dimension of the global feature vector G is

[0086] 2. The voxelized data is input into a convolutional neural network to extract local feature information of the voxel data;

[0087] A CNN model including multiple convolutional layers, pooling layers and fully connected layers is constructed. For example, the convolutional kernel size of the first convolutional layer is 3x3x3, the step size is 1, the input channel number is the dimension d of the voxel feature v , such as d v =4, including the number of points, the average coordinates and the average reflection intensity, and the output channel number is c1, c1=16.

[0088] If there are labeled forest landscape point cloud data and their corresponding landscape 3D metric feature local feature labels, the CNN is trained. Suppose there are m voxelized point cloud samples V={V1, V2,..., V m} in the training data set, and the corresponding local feature labels are Y={y1, y2,..., y m}. Define a loss function, such as a mean square error loss function , wherein is the predicted local feature of the i-th sample by the CNN. Through a back propagation algorithm, the parameters of the CNN model are updated according to the loss function to minimize the loss function.

[0089] The voxelized point cloud data V={V1, V2,..., V m} is input into the trained CNN model, and after processing by the convolutional layers and the pooling layers, a feature map F of the voxel data is obtained , wherein l is the number of feature maps, the size of each feature map is h x w x c, wherein h and w are the height and width of the feature map, and c is the number of channels.

[0090] 3. The global feature information of the point cloud is fused with the local feature information of the voxel data to obtain a fusion feature vector;

[0091] The first d g dimensions of the global feature vector G extracted from the PointNet are selected as key global features G s , and the dimensions thereof are transformed into d f by a fully connected layer, and the transformed features are denoted as G f , and the calculation formula is: G f =σ(G s W f +b f ), wherein: W f is the weight matrix of the fully connected layer, and b fis the bias vector, and σ is the activation function.

[0092] For the feature maps extracted by CNN, a global average pooling operation is used to convert each feature map into a feature vector where: is the feature value of the jth feature map at the (x, y) position. Then these feature vectors are spliced into a long vector and its dimension is also transformed to d by a fully connected layer f The transformed feature is denoted as The transformed PointNet feature G f and the CNN feature are spliced in dimension to obtain a fusion feature vector with dimension 2d f

[0093] 4. Calculate the landscape 3D metric features based on the fusion feature vector.

[0094] The landscape 3D metric features include landscape diversity, landscape shape, landscape connectivity, color uniformity, openness, and terrain. In terms of landscape diversity, the horizontal, vertical, and distance diversity (HVDD) index is used for quantification. The main components of HVDD include the horizontal, vertical, and distance profiles of landscape visual quality. The horizontal profile shows the landscape structure based on different horizontal segments; the vertical profile reveals the vertical layered structure of the landscape; and the distance curve represents the distance of landscape elements to the center position. The spatial structure of landscape visual quality is described by combining the horizontal, vertical, and distance profiles. The calculation of HVDD requires that the landscape be divided into groups according to the horizontal angle, height, and distance to the viewpoint. The number of groups depends on the size of the voxel and is not fixed. Then, the volumes of each group are compared to calculate the spatial diversity of landscape visual quality. The larger the HVDD value, the higher the spatial diversity and landscape complexity. The calculation formula is as follows:

[0095]

[0096] where: HVDD is the diversity index, N is the number of groups, V i is the volume of the ith group, and V is the average volume of all groups.

[0097] In terms of landscape shape, the shape index and surface-volume ratio are used to quantify shape diversity or regularity. The three-dimensional landscape shape index is the ratio of the actual surface area of the landscape to the minimum surface area of the landscape of the same size. Compared with the 2D shape index based on edge length or 2D area, the 3D index uses 3D surface area or volume for analysis. The value range of the three-dimensional shape index is 1 to infinity. The surface-volume ratio is the ratio of the surface area of the landscape to its volume. The lower the three-dimensional shape index or surface-volume ratio, the more standard and regular the shape of the landscape, and vice versa, the more complex the shape. The shape extraction is carried out based on the landscape shape recognition algorithm of unsupervised learning in machine learning. The calculation formula is as follows:

[0098]

[0099]

[0100] Wherein: MSI is the shape index, SVR is the surface-volume ratio, S is the actual surface area of the landscape, S min is the minimum surface area of the landscape of the same size, and V is the volume of the landscape.

[0101] In terms of landscape connectivity, four indexes are included, namely, cluster number, cluster density, cluster size and cluster distance. The cluster number is the number of independent clusters, and the nearest cluster must have a volume and distance greater than a certain size and distance; the cluster density is the ratio of the cluster number to the spatial volume of the landscape; the cluster size refers to the average volume of the cluster; and the cluster distance refers to the average distance between the different cluster centroid voxel points. The fewer the cluster number, the lower the density, the larger the volume and the shorter the distance, the higher the landscape connectivity. The calculation formula is as follows:

[0102] NC=M

[0103]

[0104] Wherein: NC is the cluster number, CD is the cluster density, CS is the cluster size, CDist is the cluster distance, M is the number of clusters obtained after statistical clustering, H a is the horizontal area of the landscape, V h is the maximum height of the landscape, |C i | is the number of voxels contained in each cluster C i , d(C i , C j ) is the distance between the i-th cluster and the j-th cluster centroid voxel point.

[0105] The color uniformity is quantified by the color change between voxels. First, the color distance between the voxel and the baseline (black-RGB(0, 0, 0)) is calculated; then, the color change index is calculated by using the color distance. High color change indicates that the landscape color difference is significant, and vice versa, the landscape color is similar. The formula is as follows:

[0106]

[0107]

[0108] where CD v is the color distance, Dave is the average color distance, C v is the color variation index, v r , v g , and v b are the values of the red, green, and blue color channels of a voxel, N v is the total number of voxels.

[0109] Openness is a quantitative measure of visual scale. In terms of landscape openness, the visual distance is used to represent the depth and range of the landscape, i.e., the degree of openness is quantified based on the central viewpoint. The visual distance is the distance between the viewpoint and the surrounding obstacles. According to the horizontal and vertical angles, the space is divided into groups, and the visual distance of all groups (the minimum distance between the viewpoint and any voxel in each group) is measured, and the average visual distance is calculated as the visual distance of the landscape unit. A longer visual distance indicates a higher landscape openness, and vice versa, indicating a closed and compact landscape. The calculation formula is as follows:

[0110]

[0111] where VD is the visual distance, K is the number of groups, d min (g k ) is the minimum distance between the kth group of viewpoints and any voxel in the group.

[0112] In terms of terrain, the average slope and gradient variation are used for quantification. According to the generated ground grid data, the grid slope is extracted, and the average value of all slopes is the average slope of the ground steepness. The gradient variation reflects the difference in grid gradient. A high gradient variation index indicates that the terrain changes greatly in the entire plot, and vice versa, indicating a flat ground. The calculation formula is as follows:

[0113]

[0114]

[0115] where AS is the average slope, GV is the gradient variation, N g is the number of ground grids, s i is the slope of the ith ground grid, g i is the gradient of the ith ground grid, g ave is the average value of all grid gradients.

[0116] S13. Construct a landscape visual quality evaluation index system based on the extracted landscape 3D metric features, and determine the visual quality evaluation scores or visual quality evaluation levels corresponding to each landscape 3D metric feature according to the landscape visual quality evaluation index system, and determine the sample data set of machine learning based on the visual quality evaluation scores or visual quality evaluation levels.

[0117] The landscape visual quality evaluation index system is constructed based on the extracted landscape 3D metric features, please refer to Figure 3 The index system includes target layer, criterion layer and index layer, wherein the target layer is landscape visual quality, the criterion layer includes landscape 3D metric features, i.e. landscape diversity, landscape shape, landscape connectivity, color uniformity, openness and terrain; the index layer includes a key feature subset obtained by analyzing and screening the indicators in each landscape 3D metric feature. Specifically, it includes the following:

[0118] 1. Analyze and screen the extracted landscape 3D metric features to determine a key feature subset related to landscape visual quality in each landscape 3D metric feature;

[0119] As described above, the landscape 3D metric features include landscape diversity, landscape shape, landscape connectivity, color uniformity, openness, terrain and other features, each feature includes multiple indicators, such as horizontal, vertical, distance and other indicators for landscape diversity; shape index, volume and other indicators for landscape shape. The key feature subset is the part of indicators that best reflects the essence and core of landscape visual quality from a large number of indicators.

[0120] First, collect the visual quality scores of experts on each landscape 3D metric feature.

[0121] Collect the visual quality scores of experts on landscape 3D metric features. Specifically, corresponding to the extracted landscape 3D metric features described above, set up a forest landscape visual quality evaluation questionnaire based on all landscape 3D metric features including landscape diversity, landscape shape, landscape connectivity, color uniformity, openness and terrain. Collect forest landscape pictures in each forest land where point cloud data is collected, select a time period with high brightness and visibility (9:00-11:00, 15:00-17:00) in a day, and use the same person and camera to take pictures, which fully reflects the characteristics of vegetation and ensures the consistency of each sample picture. The collected pictures are first processed by Adobe Photoshop for background removal and pixel uniformity, then the subjects are invited to score after observing the pictures for 5-7 seconds, and the pictures are evaluated by first intuition. The questionnaire is distributed by combining field research with random distribution on the network platform. The questionnaire results are analyzed for validity and reliability by Cronbach test and Pearson correlation analysis, and then the rationality of the number of effective questionnaires is determined.

[0122] To eliminate the influence of individual differences in landscape evaluation, the original questionnaire results were standardized, and the calculation formula is as follows:

[0123]

[0124] in: Let be the standardized value of the landscape visual quality score given by the p-th subject to the 3D landscape metric feature X. Let be the landscape visual quality score given by the p-th subject to the 3D landscape metric feature X. Let be the average landscape visual quality score given by the p-th subject to the landscape 3D metric feature X. Let N be the standard deviation of the landscape visual quality score given by the p-th subject to the landscape 3D metric feature X, and N be the standard deviation of the score given by the subject. p For the total number of subjects, LVD X The final standardized landscape visual quality value for the 3D landscape metric feature X.

[0125] Then, the Pearson correlation coefficient between each index in each extracted landscape 3D metric feature and the aforementioned known landscape visual quality score is calculated, and a subset of key features related to landscape visual quality in each landscape 3D metric feature is selected. Let the landscape diversity vector in the landscape 3D metric feature be X1={x1,x2,…,x n}, where n is the number of indicators, and the corresponding landscape visual quality score is Y={y1,y2,…,y m}, where m is the sample size. For the i-th indicator x i Correlation coefficient between score y and Pearson coefficient The calculation formula is:

[0126]

[0127] Where: x ij For the j-th sample, the i-th index value is... It is the average value of the i-th indicator. This is the average score. Based on the absolute value of the correlation coefficient, indicators with high correlation are selected, for example, choosing... Indicators >0.5 are used as a preliminary subset of key features.

[0128] 2. Collect expert scores for the selected key feature subsets and determine the visual quality level or visual quality score corresponding to each landscape 3D metric feature;

[0129] Collect the landscape visual quality expert scores of the key feature subsets corresponding to the landscape 3D metric features: for example, the key feature subsets obtained based on the above-mentioned landscape diversity screening include horizontal, vertical and distance diversity, and the key feature subsets of other landscape 3D metric features are determined in the same way, and then a forest landscape visual quality evaluation questionnaire is set based on the key feature subsets corresponding to all the landscape 3D metric features, the questionnaire pictures are collected, post-processed and tested as above.

[0130] Similarly, in order to eliminate the influence of individual differences in landscape evaluation, the original questionnaire results are standardized, and the calculation formula is as follows:

[0131]

[0132] Wherein: is the standardized value of the landscape visual quality score of the pth subject to the ith key feature subset of the landscape 3D metric feature X, is the landscape visual quality score of the pth subject to the ith key feature subset of the landscape 3D metric feature X, is the average value of the landscape visual quality scores of the pth subject to all subsets of the landscape 3D metric feature X, is the standard deviation of the landscape visual quality scores of the pth subject to all subsets of the landscape 3D metric feature X, N p is the total number of subjects, is the final standardized landscape visual quality value of the ith key feature subset of the landscape 3D metric feature X.

[0133] After determining the scores of the key feature subsets corresponding to each landscape 3D metric feature, the visual quality evaluation results of the landscape 3D metric features can be obtained in two ways, one is to determine the evaluation results by calculating the quality evaluation score through setting the weight distribution of each subset, and the other is to determine the evaluation results by calculating the quality evaluation grade based on the evaluation score.

[0134] In terms of evaluation score, the calculation formula of the landscape visual quality score of each landscape 3D metric feature is as follows:

[0135]

[0136] Wherein: S X is the visual quality score of the landscape 3D metric feature X, k is the number of key feature subsets in the landscape 3D metric feature X, w Xi is the weight of the ith key feature subset of the landscape 3D metric feature X, the weight is determined by the proportion of the score of each key feature subset in the landscape 3D metric feature X in the total sum of all subset scores.

[0137] In terms of evaluation grades, the landscape visual quality is divided into five grades, i.e. very good, good, general, poor and very poor, according to the equal interval method.

[0138] Very good: [(LVD max ―LVD min )×0.8+LVD min ,LVD max ]

[0139] Good: [(LVD max ―LVD min )×0.6,(LVD max ―LVD min )×0.8+LVD min ]

[0140] General:

[0141] [(LVD max ―LVD min )×0.4+LVD min ,(LVD max ―LVD min )×0.6+LVD min ]

[0142] Poor:

[0143] [(LVD max ―LVD min )×0.2+LVD min ,(LVD max ―LVD min )×0.4+LVD min ]

[0144] Very poor: [LVD min ,(LVD max ―LVD min )×0.2+LVD min ]

[0145] Wherein, LVD max represents the maximum landscape visual quality value, and LVD min represents the minimum landscape visual quality value.

[0146] 3. The landscape 3D measurement feature vector extracted in each sample is taken as the feature data of the data set, and the visual quality score or rating of each landscape 3D measurement feature is taken as the label data of the data set, so as to obtain the data set for machine learning.

[0147] Specifically, the obtained machine learning dataset is divided into a training set, a validation set, and a test set. The training set serves as the sample dataset for training the model, the validation set is used to adjust model parameters during training, and the test set is used to evaluate the performance of the trained model. The landscape 3D metric feature vectors extracted from each sample and corresponding to the selected key feature subsets are used as the feature data of the dataset. The visual quality score or rating S of each landscape 3D metric feature is used as the label data to construct a dataset {(X1,S1),(X2,S2),…,(X... m ,S m )}, where X j Let S be the feature vector value of the j-th sample. j Assign it a corresponding visual quality score or rating.

[0148] The labeled data in this machine learning sample dataset can be either visual quality ratings or visual quality scores for various landscape 3D metric features, providing two types of evaluation results to meet the training needs of different models. The machine learning models involved in visual quality evaluation include five types: Decision Tree (DT), Random Forest (RF), Gaussian Process Regression (GPR), Support Vector Machine Regression (SVR), and Neural Network (NN).

[0149] S14. Use the sample dataset to train the preset landscape visual quality assessment model, and use the trained landscape visual quality assessment model to assess the visual quality of any new 3D point cloud data of the forest landscape.

[0150] In this embodiment, step S14 further includes the following:

[0151] The model learns the mapping relationship between key feature subsets in each landscape 3D measurement feature and the corresponding landscape 3D measurement feature, as well as the mapping relationship between each landscape 3D measurement feature and landscape visual quality, and determines the trained landscape visual quality assessment model based on the learned mapping relationship.

[0152] Assume the landscape 3D metric feature vector corresponding to the sample dataset is X. target =[x t1 ,x t2 ,...,x tn ], where n is the number of features, x ti Let X be the i-th feature value. The landscape 3D metric feature vector X... targetand the corresponding visual quality grades or visual quality scores are input into the constructed landscape visual quality assessment model to learn the mapping relationship between different degrees of each landscape 3D metric feature and landscape visual quality, where the different degrees include the mapping relationship between different visual quality grades or different visual quality scores of the corresponding landscape 3D metric feature and the key feature subset of each landscape 3D metric feature; and the mapping relationship between each landscape 3D metric feature and different grades of the determined landscape visual quality or the mapping relationship between different range score values of the landscape visual quality. In addition, different types of models have different input layer structures, but all receive the feature vector as the initial input data.

[0153] According to the learned mapping relationship between different degrees of each landscape 3D metric feature and landscape visual quality, the specific visual quality score or grade of the forest landscape can be determined. Here, whether the model outputs the visual quality score or the visual quality grade is determined according to the selected training model. If the selected model is trained based on a regression task, a specific numerical value will be output; if the selected model is trained according to a classification task, the grade to which the landscape visual quality belongs will be output. During the training process, a validation set with part of the data with landscape 3D metric features and scores is used to monitor the training effect of the model and prevent overfitting. By comparing the predicted results of the model on the validation set with the actual scores, the performance of the model is evaluated, and the trained visual quality assessment model is finally determined.

[0154] When the 3D metric feature data of a new forest landscape is input into the trained model, the model calculates and judges according to the learned mapping relationship, and outputs the corresponding visual quality quantitative value or the quality grade to which it belongs (such as very good, good, general, poor, and very poor). For example, for a forest landscape with a specific combination of landscape diversity, shape, connectivity, etc., the model can give an accurate score, such as 75 points, or classify it into the "good" grade category, thereby achieving objective and accurate assessment of the visual quality of the forest landscape and overcoming the subjectivity and fuzziness of traditional qualitative assessment methods.

[0155] Finally, the trained model is evaluated and optimized using a test set. Here, multiple evaluation indicators such as root mean square error, R 2 , accuracy, and recall rate are used to comprehensively evaluate the performance of the model. According to the evaluation results, the model is optimized, including but not limited to adjusting hyperparameters, selecting different feature combinations, increasing training data, etc. The selection of hyperparameters is crucial in the training process of machine learning models. Hyperparameter tuning can effectively improve the accuracy and stability of the model. The tuning methods include grid search, random search, or Bayesian optimization, etc.

[0156] The above embodiment is based on the construction of the landscape visual quality evaluation system of the landscape 3D measurement feature, and determines the sample data set suitable for machine learning, which lays a foundation for subsequent training of an accurate landscape visual quality prediction model. Finally, any new three-dimensional data obtained in real time is input into the trained landscape visual quality evaluation model to obtain the corresponding visual quality evaluation result. Whether it is a quantitative score or a classified level, it can provide a reference basis for forest landscape management, ecological tourism planning and other related applications.

[0157] Embodiment two

[0158] An embodiment of a forest landscape visual quality evaluation system based on point cloud data and machine learning in the present application comprises the following steps:

[0159] A data acquisition and preprocessing unit is configured to acquire three-dimensional point cloud data of a forest land using a laser radar, and to perform data preprocessing based on the acquired three-dimensional point cloud data;

[0160] A landscape 3D measurement feature extraction unit is configured to extract landscape 3D measurement features based on the preprocessed data;

[0161] A sample data set determination unit is configured to construct a landscape visual quality evaluation system based on the extracted landscape 3D measurement features, to determine the visual quality evaluation score or visual quality evaluation level corresponding to each landscape 3D measurement feature according to the landscape visual quality evaluation system, and to determine the sample data set for machine learning based on the visual quality evaluation score or visual quality evaluation level;

[0162] A landscape visual quality evaluation model training unit is configured to train a preset landscape visual quality evaluation model using the sample data set, and to perform forest landscape visual quality evaluation on any new three-dimensional point cloud data using the trained landscape visual quality evaluation model.

[0163] The specific limitations of the system can be referred to the limitations of the method in the above, which will not be repeated here. Each module in the above system can be realized by software, hardware and their combination in whole or in part. The above modules can be embedded in or independent of the processor in the computer device in hardware form, or can be stored in the memory in the computer device in software form, so as to be called and executed by the processor to perform the operations corresponding to each module.

[0164] Those skilled in the art can understand that the units of the examples described in combination with the embodiments disclosed herein can be realized in electronic hardware, computer software or a combination of both. In order to clearly illustrate the interchangeability of hardware and software, the components of the examples have been described in general terms in the above description. Whether the functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. The skilled person can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0165] In the embodiments provided by the present application, it should be understood that the division of units is only a logical functional division, and there can be another division manner in actual implementation, for example, multiple units can be combined into one unit, one unit can be split into multiple units, or some features can be ignored, etc. In addition, each functional unit in each embodiment of the present application can be integrated in one processing unit, or each unit can exist physically independently, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of a software functional unit.

[0166] The integrated unit, if realized in the form of a software functional unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a number of instructions for causing a computer device (which can be a personal computer, a server or a network device, etc.) to execute all or part of the steps of the methods described in the embodiments of the present application. The aforementioned storage medium includes: a U disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a mobile hard disk, a magnetic disk or an optical disk, and various media that can store program codes.

[0167] It can be understood that the above embodiments are only used to illustrate the technical solutions of the present application, but not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that the technical solutions recorded in the foregoing embodiments can be modified, or some or all of the technical features can be replaced by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application, and they should be covered in the scope of the claims and the description of the present application.

Claims

1. A forest landscape visual quality assessment method based on point cloud data and machine learning, characterized in that, The application relates to a method for evaluating forest landscape visual quality based on three-dimensional (3D) point cloud data. The method comprises the following steps: acquiring three-dimensional point cloud data of a forest land by using a laser radar, and performing data preprocessing based on the acquired three-dimensional point cloud data; extracting landscape 3D metric features based on the preprocessed data; the step of extracting landscape 3D metric features based on the preprocessed data comprises the following steps: normalizing and voxelizing the preprocessed data by using a PointNet model to obtain global feature information of the point cloud; inputting the voxelized data into a convolutional neural network to extract local feature information of the voxel data; fusing the global feature information of the point cloud and the local feature information of the voxel data to obtain a fusion feature vector; calculating landscape 3D metric features based on the fusion feature vector; the step of calculating landscape 3D metric features based on the fusion feature vector comprises the following steps: the landscape 3D metric features include landscape diversity, landscape shape, landscape connectivity, color uniformity, openness and terrain; wherein: is the diversity index, is the number of groups, is the volume of the group, is the average volume of all groups; landscape diversity: wherein: is the shape index, is the surface to volume ratio, is the actual surface area of the landscape, is the minimum surface area of the landscape of the same size, is the volume of the landscape; landscape shape: wherein: is the number of clusters, is the cluster density, is the cluster size, is the cluster distance, is the number of clusters obtained after statistical clustering, is the horizontal area of the landscape, is the maximum height of the landscape, is the number of voxels contained in each cluster , is the distance between the th cluster and the th cluster centroid voxel. landscape connectivity: wherein: is the color distance, is the average color distance, is the color variation index, , and are the values of the red, green and blue color channels of the voxel, respectively, is the total number of voxels; color uniformity: in: For field of view distance, Number of groups For the first The minimum distance between the viewpoint of a group and any voxel within the group; openness: wherein: is the average slope, is the change in gradient, is the number of ground grids, is the slope of the th ground grid, is the gradient of the th ground grid, is the average of all grid gradients; terrain: constructing a landscape visual quality evaluation system based on the extracted landscape 3D metric features, determining visual quality evaluation scores or visual quality evaluation grades corresponding to each landscape 3D metric feature according to the landscape visual quality evaluation system, and determining a sample data set for machine learning based on the visual quality evaluation scores or visual quality evaluation grades; 2. The forest landscape visual quality assessment method based on point cloud data and machine learning according to claim 1, characterized in that, training a preset landscape visual quality evaluation model by using the sample data set, and performing forest landscape visual quality evaluation on any new three-dimensional point cloud data by using the trained landscape visual quality evaluation model. wherein: is the fused feature vector, is the transformed PointNet feature, is the transformed feature by the convolutional neural network, is an activation function, is the key global feature, is the weight matrix of the fully connected layer, is the bias vector, is the feature vector transformed by the global average pooling, and is the height and width of the feature map, is the feature value of the th feature map at the th position.

3. The forest landscape visual quality assessment method based on point cloud data and machine learning according to claim 1, characterized in that, the expression for fusing the global feature information of the point cloud and the local feature information of the voxel data to obtain a fusion feature vector comprises the following steps: the step of constructing a landscape visual quality evaluation system based on the extracted landscape 3D metric features, determining visual quality evaluation scores or visual quality evaluation grades corresponding to each landscape 3D metric feature according to the landscape visual quality evaluation system, and determining a sample data set for machine learning based on the visual quality evaluation scores or visual quality evaluation grades comprises the following steps: analyzing and screening the extracted landscape 3D metric features to determine a key feature subset related to landscape visual quality in each landscape 3D metric feature; collecting expert scores of the screened key feature subset, and determining visual quality grades or visual quality scores corresponding to each landscape 3D metric feature; 4. The forest landscape visual quality assessment method based on point cloud data and machine learning according to claim 3, characterized in that, taking the landscape 3D metric feature vector corresponding to the key feature subset screened from each sample as feature data of the data set, taking the scores or grades of the landscape visual quality of each landscape 3D metric feature as label data of the data set, and obtaining a sample data set for machine learning. in: For the first Individual subjects' 3D landscape measurement features The Middle Standardized values ​​of landscape visual quality scores for a subset of key features. For the first Individual subjects' 3D landscape measurement features The Middle Landscape visual quality score for a subset of key features For the first Individual subjects' 3D landscape measurement features The average landscape visual quality score of all subsets. For the first Individual subjects' 3D landscape measurement features The standard deviation of the landscape visual quality score for all subsets. The total number of subjects, 3D measurement features of landscape The Middle The final standardized landscape visual quality value of a subset of key features.

5. The forest landscape visual quality assessment method based on point cloud data and machine learning according to claim 4, characterized in that, the expression for collecting expert scores of the screened key feature subset comprises the following steps: the step of collecting expert scores of the screened key feature subset and determining visual quality grades or visual quality scores corresponding to each landscape 3D metric feature comprises the following steps: dividing the landscape visual quality into five grades, i.e. very good, good, general, poor and very poor, according to the equal interval method: Very good: Good: Generally: Poor: Very poor: wherein, represents the maximum landscape visual quality value, represents the minimum landscape visual quality value.

6. The forest landscape visual quality assessment method based on point cloud data and machine learning according to claim 5, characterized in that, The expert score of the collected key feature subset is screened, and the visual quality grade or visual quality score corresponding to each landscape 3D measurement feature is determined, including: wherein: is a visual quality score, is a number of key feature subsets, is a weight of the th key feature subset.

7. The forest landscape visual quality assessment method based on point cloud data and machine learning according to claim 1, characterized in that, The preset landscape visual quality evaluation model is trained using the sample data set, and the forest landscape visual quality of any new three-dimensional point cloud data is evaluated using the trained landscape visual quality evaluation model, including: The model learns the mapping relationship between the key feature subset in each landscape 3D measurement feature and the corresponding landscape 3D measurement feature, and learns the mapping relationship between each landscape 3D measurement feature and the landscape visual quality, and determines the trained landscape visual quality evaluation model according to the learned mapping relationship. 8.A forest landscape visual quality assessment system based on point cloud data and machine learning, characterized in that, The method of any one of claims 1-7, comprising: A data acquisition and preprocessing unit for acquiring three-dimensional point cloud data of a forest land using a laser radar, and performing data preprocessing based on the acquired three-dimensional point cloud data; A landscape 3D measurement feature extraction unit for extracting landscape 3D measurement features based on the preprocessed data; A sample data set determination unit for constructing a landscape visual quality evaluation system based on the extracted landscape 3D measurement features, and determining the visual quality evaluation score or visual quality evaluation grade corresponding to each landscape 3D measurement feature according to the landscape visual quality evaluation system, and determining the sample data set for machine learning based on the visual quality evaluation score or visual quality evaluation grade; A landscape visual quality evaluation model training unit for training a preset landscape visual quality evaluation model using the sample data set, and evaluating the forest landscape visual quality of any new three-dimensional point cloud data using the trained landscape visual quality evaluation model.

Images