Multi-feature-based enterprise freight facility discrimination method and system

Abstract

The invention relates to the technical field of urban large-scale building facility group merging, in particular to a multi-feature-based enterprise freight facility discrimination method and system, which have very strong positive significance for remote sensing images with very large scales by simulating human eye scanning and observing satellite images and gradually observing each part of the images. Detailed information in the image is reserved to a greater extent, a far-end neglect hypothesis is established, the similarity degree between two or more targets is determined through building target similarity calculation, a freight facility building group is established for building areas with the target similarity, and a freight facility building group is established for the building areas with the target similarity. According to the method, core freight vehicle correlation weight and non-core vehicle distance correlation weight of each building and freight facility building group are calculated, meanwhile, a semi-supervised learning freight facility discrimination model is established, the model is continuously improved through a teaching logic regression device and a student logic regression device, and distribution of model discrimination is learned through an integrated learning method.

Description

Technical Field

[0001] The present invention relates to the technical field of merging urban large-scale building facility groups, and in particular to a method and system for distinguishing enterprise freight facilities based on multiple features. Background Art

[0002] In real life, almost all buildings, including residential buildings and commercial and industrial buildings, exist in clusters of building communities and industrial parks. The buildings in these building communities often show certain graphic similarities in satellite images. However, the segmented building samples do not have such clustering information. However, when discussing the parking of freight vehicles in building complexes, if we simply extract the buildings from the building complexes to which they belong for spatial analysis, it will lead to different qualitative analysis results for different building objects in the same building complex, such as Figure 3 The parking of freight vehicles between facilities often has strong local clustering and overall imbalance.

[0003] Traditionally, in the process of spatial regional agglomeration of building facilities, the first step is to verify the distance of different buildings, because information that is closer to the target has a stronger innate correlation with the target itself. This step can help identify the building facilities that are most relevant to the target and provide valuable reference for subsequent analysis. By verifying the distance, we can more effectively focus on the building facilities that are closely related to the target, thereby improving the efficiency and accuracy of the spatial regional agglomeration process. By measuring the Euclidean distance between the centroids of building facilities, we can quantitatively judge whether there is spatial correlation between buildings. This method is easy to calculate, but it obviously ignores the direct impact of the building scale on the calculation results of the centroid Euclidean distance. The larger the building scale, the greater the distance between its centroid and the centroids of other buildings, so this method may lead to misjudgment of spatial correlation. Summary of the invention

[0004] In view of the problems existing in the prior art, the present invention is proposed.

[0005] Therefore, the problem to be solved by the present invention is how to solve the problem that the segmented building samples do not have specific building clustering information. However, when discussing the parking of freight vehicles in a building complex, if the building is simply extracted from the building complex to which it belongs and spatial analysis is performed, it will lead to the problem of different qualitative analysis results for different building objects in the same building complex.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, an embodiment of the present invention provides a method for identifying enterprise freight facilities based on multiple features, comprising: acquiring first image data of a building area, and performing first preprocessing on the first image data to obtain second image data;

[0008] Presetting a first building freight facility discrimination model, and inputting the second image data into the first building freight facility discrimination model;

[0009] Intelligent identification of freight facilities is performed based on the output of the first building freight facility identification model.

[0010] As a preferred solution of the enterprise freight facility identification method based on multiple features of the present invention, the first image data is building area data obtained by segmenting satellite images, and the first preprocessing includes:

[0011] Perform distance verification on the building area and identify the building facilities most relevant to the target;

[0012] The image is divided into multiple local areas through a sliding window with a fixed step size and size, and a sliding window with the centroid of the target freight building facility as the centroid of the sliding window is constructed.

[0013] As a preferred solution of the enterprise freight facility identification method based on multiple features of the present invention, the first building freight facility identification model includes:

[0014] Building target similarity calculation, used to calculate the similarity between building areas;

[0015] Building complex freight weight calculation, used to calculate the core freight vehicle association weight and non-core vehicle distance association weight of the building and freight facility complex;

[0016] Semi-supervised learning discriminative unit for discriminating whether a building is a cargo facility by teaching a logistic regressor and a student logistic regressor.

[0017] As a preferred solution of the enterprise freight facility identification method based on multiple features of the present invention, the building target similarity calculation includes:

[0018] A geometric similarity calculation unit is used to calculate the similarity of geometric features such as shape and size of buildings;

[0019] The texture similarity calculation unit is used to calculate the similarity of texture features such as hue, purity and brightness of the building.

[0020] As a preferred solution of the enterprise freight facility identification method based on multiple features of the present invention, the freight weight calculation of the building complex includes:

[0021] A core vehicle association weight calculation unit is used to calculate the vehicle weight within the core area of ​​a building / building complex;

[0022] The non-core vehicle distance association weight calculation unit is used to calculate the vehicle weight in the non-core area of ​​the building / building complex.

[0023] As a preferred solution of the enterprise freight facility identification method based on multiple features of the present invention, the semi-supervised learning identification unit includes:

[0024] Teach a logistic regressor, including an encoder, a noisy decoder, and a clean decoder;

[0025] Student logistic regressor, used to learn the data distribution from the taught model and make predictions.

[0026] As a preferred solution of the multi-feature-based enterprise freight facility identification method described in the present invention, the model operation module performs feature characterization on the input data, including multi-dimensional feature data of building area, density, rectangularity, eccentricity, main axis direction angle, number of vertices, and clustering information.

[0027] In a second aspect, an embodiment of the present invention provides a system for distinguishing enterprise freight facilities based on multiple features, which includes an image acquisition module for acquiring first image data of a building area and performing a first preprocessing on the first image data to obtain second image data;

[0028] An input module, presetting a first building freight facility discrimination model, and inputting the second image data into the first building freight facility discrimination model;

[0029] The discrimination output module performs intelligent discrimination of freight facilities according to the output of the first building freight facility discrimination model.

[0030] In a third aspect, an embodiment of the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program instructions are executed by the processor, the steps of the method for identifying enterprise freight facilities based on multiple features as described in the first aspect of the present invention are implemented.

[0031] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: the computer program instructions, when executed by a processor, implement the steps of the method for identifying enterprise freight facilities based on multiple features as described in the first aspect of the present invention.

[0032] The beneficial effects of the present invention are as follows: the present invention simulates human eye scanning when observing satellite images, gradually observes each part of the image, has a very strong positive significance for remote sensing images of very large scale, retains the detail information in the image to a greater extent, and establishes a far-end neglect hypothesis. By calculating the similarity of building targets, the similarity between two or more targets is determined, and a freight facility building complex is established in the building area with target similarity. The core freight vehicle association weight and the non-core vehicle distance association weight of each building and freight facility building complex are calculated, and a semi-supervised learning freight facility discrimination model is established. The model is continuously improved by teaching a logistic regressor and a student logistic regressor, and an integrated learning method is used to learn the distribution of model discrimination, thereby solving the problem that simply extracting a building from its building complex for spatial analysis will inevitably lead to incorrect classification and cognition of other samples. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0034] Figure 1 It is a flowchart of the enterprise freight facility identification method based on multiple features;

[0035] Figure 2 A computer equipment diagram for a method of identifying enterprise freight facilities based on multiple features;

[0036] Figure 3 This is an example diagram of a city freight building facility complex based on a multi-feature enterprise freight facility identification method;

[0037] Figure 4 A schematic diagram of a similarity-based urban freight building facility cluster algorithm based on a multi-feature-based enterprise freight facility identification method;

[0038] Figure 5 It is a diagram showing the core associated area of ​​urban freight building facilities based on the multi-feature enterprise freight facility identification method;

[0039] Figure 6 A flowchart of the teaching model architecture and training process for the enterprise freight facility identification method based on multiple features;

[0040] Figure 7 Flowchart of the process of training a student model for a teaching model of a multi-feature based enterprise freight facility discrimination method. DETAILED DESCRIPTION

[0041] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0042] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0043] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor is it a separate or selective embodiment that is mutually exclusive with other embodiments.

[0044] Example 1

[0045] Reference Figure 1 to Figure 7 , which is the first embodiment of the present invention, and provides a method for identifying enterprise freight facilities based on multiple features, including:

[0046] S100: Acquire first image data of a building area, and perform first preprocessing on the first image data to obtain second image data;

[0047] Preferably, the building area is obtained by segmenting the buildings in the building settlement through satellite images, and the distance verification is performed on different buildings to identify the building facilities most relevant to the target;

[0048] S101: The first image data is building area data obtained by performing building segmentation on a satellite image. The first preprocessing includes:

[0049] Perform distance verification on the building area and identify the building facilities most relevant to the target;

[0050] The image is divided into multiple local areas through a sliding window with a fixed step size and size, and a sliding window with the centroid of the target freight building facility as the centroid of the sliding window is constructed.

[0051] Preferably, the image is simulated by human eye scanning in observing satellite images, by sliding a window with a fixed step length and size on the image, gradually covering the entire image, dividing the image into multiple local areas, and applying a target detection algorithm or classifier in each building area to identify the target, and constructing a sliding window with the centroid of the target freight building facility as the centroid of the sliding window;

[0052] S200: Presetting a first building freight facility discrimination model, and inputting the second image data into the first building freight facility discrimination model;

[0053] S201: The first building freight facility identification model includes:

[0054] Building target similarity calculation, used to calculate the similarity between building areas;

[0055] Building complex freight weight calculation, used to calculate the core freight vehicle association weight and non-core vehicle distance association weight of the building and freight facility complex;

[0056] Semi-supervised learning discriminative unit for discriminating whether a building is a cargo facility by teaching a logistic regressor and a student logistic regressor.

[0057] S202: Building target similarity calculation includes:

[0058] A geometric similarity calculation unit is used to calculate the similarity of geometric features such as shape and size of buildings;

[0059] The texture similarity calculation unit is used to calculate the similarity of texture features such as hue, purity and brightness of the building.

[0060] Preferably, the building target similarity calculation of the building area is to establish a freight facility building complex in the building area with target similarity. The calculation of target similarity mainly consists of geometric similarity and texture similarity, and the formula is as follows:

[0061] ΔSim=ρ geom +ρ tex (4.6);

[0063] S203: Building complex freight weight calculation includes:

[0064] A core vehicle association weight calculation unit is used to calculate the vehicle weight within the core area of ​​a building / building complex;

[0065] The non-core vehicle distance association weight calculation unit is used to calculate the vehicle weight in the non-core area of ​​the building / building complex.

[0066] Preferably, the calculation formula for the core freight vehicle association weight and the non-core vehicle distance association weight of each building and freight facility building complex is as follows:

[0067]

[0068] For freight facility complexes, it is necessary to perform a secondary division of freight vehicle weights internally. The calculation formula for the secondary division is as follows:

[0069]

[0070] S204: The semi-supervised learning discriminant unit includes:

[0071] Teach a logistic regressor, including an encoder, a noisy decoder, and a clean decoder;

[0072] Student logistic regressor, used to learn the data distribution from the taught model and make predictions.

[0073] S205: The model operation module performs feature characterization on the input data, including multi-dimensional feature data of building area, density, rectangularity, eccentricity, main axis direction angle, number of vertices, and clustering information.

[0074] S300: Perform intelligent identification of freight facilities according to the output of the freight facility identification model of the first building.

[0075] Preferably, a semi-supervised learning freight facility discrimination model is established, and the distribution of model discrimination is learned by teaching the logistic regressor and the student logistic regressor, and a soft label is used to mark whether the building facility is an active freight building facility, and the problem is converted into a logistic regression problem represented by a probability distribution. The output of the model is shown in the following formula:

[0076] Output m =[1-P(A),P(A)](4.12).

[0078] The calculation formula of geometric similarity is as follows:

[0079] ρ geom =sim Τ -δ area (4.7.1),

[0081]

[0082]

[0083] in, is the geometric similarity; sim T represents the similarity of the inner angle; δ area Indicates the area difference, T a and T b Represents the dimensionless internal angle matrix of the two building targets a and b respectively; ||T|| represents the modulus of the tensor; T a ·T b represents the dot product, S a With S b Represents the area of ​​two building targets a and b, Represents the 85% quantile of the building area distribution curve within the remote sensing image area.

[0084] The texture similarity is calculated as follows:

[0085]

[0086] in, is the texture similarity, δ H is the hue difference, δ S is the purity difference, δ V is the brightness difference, Respectively represent the hue, purity and brightness of the target building area.

[0087] The hue has a special periodicity. The colors represented by hue number 0 and hue number 360 are both pure red.

[0088] Scale it to The cosine value is calculated in the interval to achieve periodic calculation.

[0089] The non-core vehicle distance association weight is as follows:

[0090]

[0091] I i is the non-core vehicle distance association weight of the building / building complex, t is the freight vehicle target set that can be retrieved in the direction quadrant of the building / building complex, and u is the parked freight vehicle target p j The set of cargo buildings / building complexes retrieved in the directional quadrant, η j The default weight for non-core vehicle distance in the region. Output m is the output of the logistic regressor, and event A is that the building is an active freight building facility. The taught logistic regressor of includes an encoder, a noisy decoder, and a clean decoder.

[0092] The spatial correlation caused by the building scale is proposed based on the maximum and minimum distances of the building area shape:

[0093]

[0094] Where M and N represent the point sets of the two target building areas, P M With P N represents a point in the point set, D EExpressed as Euclidean distance. Although this method can well characterize the distance relationship between two building areas, it is found in practice that this distance characterization method needs to consider the Euclidean distance of all points on the boundaries of the two polygonal areas and compare them, and the time cost of the operation is too high. At the same time, the shortest or longest distance is easily affected by outliers because only the distance between the extreme points of the two polygons is considered. If there are abnormalities such as outliers or incomplete segmentation in the recognition results, the shortest or longest distance may not be robust enough. At the same time, for polygons with complex shapes, especially building area polygons with concave and convex parts or internal holes, the calculation of the shortest or longest distance will be more complicated, and the result may not reflect the required real distance information.

[0095] The scale of the building complex is complex, ranging from 500 meters to 2000 meters. In order to ensure that the merged model has good dynamic adaptability to larger and smaller building complexes, the human eye scanning when observing satellite images is simulated. By sliding a window with a fixed step size and size on the image, the entire image is gradually covered, the image is divided into multiple local areas, and the target detection algorithm or classifier is applied in each building area to identify the target. The centroid of the target freight building facility is used as the centroid of the sliding window to construct a sliding window; the behavior of the human eye when observing the image, the human eye will move on the image at a certain speed and step size to obtain the overall picture and detailed information of the image.

[0096] The core idea of ​​sliding window technology is to divide the image into multiple local areas and apply target detection algorithms or classifiers in each area to identify the target of interest. It simulates the scanning method of the human eye when observing remote sensing images. The human eye will scan the image at a certain step size and speed, so as to gradually observe each part of the image. The observation of local areas can retain the detailed information in the image to a greater extent, and establish the far-end neglect hypothesis, that is, if the spatial distance between the two objects to be discussed is too far, it is a priori assumed that the connection between the two is too small and can be subjectively set to zero. The recognition results of urban freight buildings are reviewed with a sliding window of a certain scale. This behavior relies on an a priori assumption: when the sliding window is reviewed, all objects within the same window range have spatial correlation, and the spatial correlation is equivalent to the entire window domain, that is, the spatial correlation of any two objects is equal and does not change with the spatial distance between the two under the window perspective. At this time, the spatial correlation of any object is converted into a Boolean variable coefficient, and its value is shown in the following formula:

[0097]

[0098] Where: M and N represent any two building facility space areas in the space, S i，j Represents the window range with row and column numbers (i, j), RM,N，i，j It represents the spatial correlation coefficient between these two building areas within the window range.

[0099] The sliding window mode can equally consider the similarity of neighboring buildings in all directions for any sample. For any target sample m, the expectation is expressed by the following formula:

[0100] ∑s i′，j′ ∈ST m R m，n，i′，j′ ≥1 (4.4.1);

[0102] N m ={n|D E (m, n)≤L′} (4.4.2);

[0104] ST m ={S i′，j′ |m∈S i′，j′} (4.4.3)

[0105] Where: m is the target sample, N m S is the set of all other targets whose centroid Euclidean distance to the target sample m is less than the judgment threshold L (usually 250500m). i′，j′ Represents all sliding windows containing the target sample m.

[0106] The sliding window is constructed with the centroid of the target freight building facility as the sliding window centroid. Because buildings have strong regional clustering, each time a sliding window is constructed with the target centroid as the center to discuss the relationship between the target sample and the surrounding samples, the constructed sliding window repetition rate is too high, and the many-to-one calculation is not convenient to convert into matrix operation to improve the algorithm efficiency. Therefore, the following formula is used here to construct the sliding window system to avoid the above defects:

[0107]

[0108] Where: s x With s y Indicates the size of the sliding window in the horizontal and vertical directions, p x With p y Represents the step size in the horizontal and vertical directions of the sliding window, W and L are the length and width of the overall minimum horizontal circumscribed rectangle of the target remote sensing image area.

[0109] Through the above series of construction methods, the number of sliding windows actually used in the model can be effectively reduced and the size characteristics of the original remote sensing image area can be fully matched. It can be seen from the formula that for a square research area, the sliding window system can maximize the effectiveness in the vertical direction to the horizontal direction.

[0110] The building target similarity calculation in the building area is to establish a freight facility building complex in the building area with target similarity. The calculation of target similarity is mainly composed of geometric similarity and texture similarity. The formula is as follows:

[0111] ΔSim=ρ geom +ρ tex (4.6);

[0113] Geometric similarity mainly involves the comparison of geometric properties such as shape, size and position of the target. The geometric similarity calculation method includes calculating the geometric features of the target such as distance, angle, area, perimeter, etc., and comparing these features with the reference object. It is found that the area and dimensionless internal angle matrix have good measurement of the similarity of buildings. In addition, if too many parameter indicators are selected, it will be impossible to balance the values ​​of geometric similarity and texture similarity, which will cause the judgment standard to be too inclined to a certain similarity indicator.

[0114] The calculation formula of geometric similarity is as follows:

[0115] ρ geom =sim T -δ area (4.7.1);

[0117]

[0118] When calculating the area difference, a special numerical normalization is performed by global normalization of the area to avoid numerical fluctuations caused by local normalization in the sliding window area. Global normalization ensures that the area of ​​the entire data set is uniformly scaled, thereby more stably comparing the area differences of different regions or different time points, and maintaining the consistency and comparability of the data.

[0119] The metric related to the dimensionless interior angle matrix is ​​cosine similarity. Cosine similarity is a common similarity measurement method for measuring the cosine value of the angle between two vectors, which is used to reflect their directional similarity. Cosine similarity can be well adapted to the similarity comparison task between high-dimensional matrices. Since the polygons obtained by thinning out the two building areas may have different numbers of vertices, it is necessary to normalize the interior angle matrix. After interpolating the two dimensionless interior angle matrices, the last column marking the corner position score is removed so that the two dimensionless interior angle matrices have the same number of data points.

[0120] Secondly, texture similarity mainly involves the comparison of the texture features, color distribution, texture structure, etc. of the target. Common texture similarity calculation methods include using statistical features, frequency domain features, local texture features, etc. to describe the texture information of the target, and evaluating the texture similarity between targets by comparing these feature vectors or histograms. The structural similarity index is selected as the calculation indicator of texture similarity.

[0121] Structural similarity index is an indicator for evaluating the similarity between two images, which is used to measure the degree of similarity between them. Compared with traditional pixel difference indicators (such as mean square error), SSIM is more in line with the perception of the human visual system and can more accurately reflect the structural and content similarity of images. SSIM evaluates the similarity between images based on three main factors: brightness difference, contrast difference, and structural difference. Brightness difference measures the brightness difference between objects, contrast difference measures the contrast difference between objects, and structural difference measures the structural difference between objects. These factors are calculated by comparing the brightness, contrast, and structural information between local blocks of the image.

[0122] For SSIM, except for structural similarity, the remaining brightness difference and contrast difference calculation methods are to calculate the mean of all pixels in a certain area on their indicators. Therefore, these two indicators are shape-independent, and structural similarity requires that the two compared objects are divided into local blocks of equal size, and a fixed-size window is used to calculate the difference at the corresponding position. Therefore, the structural difference has a significant shape correlation, that is, it requires that the two objects / areas involved in the comparison have complete identity in shape / size. However, the building objects involved in the comparison in engineering often do not have complete identity, and as shown in the previous picture, the building roof is a weak texture, and the color distribution of pixels in the area is often not structural. The texture features that can be mined are limited in most cases. In order to avoid the requirement of complete identity of shape and retain its main features to the greatest extent for comparison. Considering that brightness and saturation are representation parameters belonging to the HSV space, the texture features are simplified to hue difference. The simplified texture similarity calculation is as follows:

[0123]

[0124]

[0125] The building complexes based on the above rule-based merging are divided as follows Figure 3 As shown in the figure (the yellow area is the segmentation result of the freight building, and the red area is the core associated area after the segmentation area of ​​the freight building is expanded by 20m in equal length), because the hue has a special periodicity, the colors represented by hue number 0 and hue number 360 are both pure red, so here we choose to scale it to The cosine value is calculated in the interval to achieve periodic calculation.

[0126] The building areas with target similarity are used to establish freight facility building groups. The buildings in the same group are discussed as a whole building object. The building group and general independent buildings are objects of equal weight.

[0127] Each building object and building complex object will have two weight calculation objects: core freight vehicle association weight and non-core vehicle distance association weight. The first level of the secondary spatial structure is the core association area, which is defined as follows: the effective space area 20m outside the actual area of ​​the building / building complex. Figure 5 As shown in the figure, a 20m space can cover the situation where a freight vehicle weighing more than 12 tons is parked vertically around the facility. Although this expansion method may be affected by the building height and loading and unloading methods, it can maintain good effectiveness overall considering that the length of mainstream freight vehicles is around 9.6m.

[0128] When a freight vehicle is located in the core area of ​​one or more freight buildings / buildings, it will no longer be considered to have a spatial association with other freight buildings / buildings, that is, the vehicle will no longer participate in the calculation of the non-core distance weight proposed later. At this time, the core freight vehicle association weight divided by each building or building group is calculated as follows:

[0129]

[0130] Among them, K i is the core freight vehicle association weight of the ith building / building complex, and {t} is all the freight vehicles existing in the core area of ​​the building / building complex. j is the preset core area vehicle weight, n j It is the cumulative number of times the vehicle is in the core areas of different buildings / building complexes.

[0131] For building complexes, it is necessary to perform secondary division of freight vehicle weights within the building complex. The calculation formula for the secondary division is as follows:

[0132]

[0133] Where: K i is the core freight vehicle association weight of the i-th building in the building complex, {r} is all the freight vehicles in the core area of ​​the building complex, and p j Represented as the jth parked freight vehicle, β j is the preset core area vehicle weight, Dis m i nE (i, j) is the calculated minimum spatial distance between the vehicle and the facility.

[0134] When a freight vehicle is not in the core area of ​​any freight building facility / building complex, it will pass the non-core distance weight. Simplify the calculation of this weight to avoid excessive calculation when only using the distance as a single judgment criterion. Use the direction quadrant to describe the spatial retrieval rules of freight vehicles.

[0135] Directional quadrants are a common concept of orientation used to describe the direction and position of a point or area in geographic space relative to a reference point. East refers to the direction east of the reference point, south refers to the direction south of the reference point, west refers to the direction west of the reference point, and north refers to the direction north of the reference point. Based on these four main directions, there are four secondary directions, namely northeast, southeast, southwest, and northwest.

[0136] Freight vehicles search for the nearest building / building complex to the vehicle target in eight different directions (the quadrants are divided according to the centroid of the building / building complex area), and the distance must not exceed a certain multiple of the spatial distance between the vehicle target and the nearest building / building complex. The multiple value needs to be determined in combination with regional conditions. The non-core vehicle distance association weight is shown in the following formula:

[0137]

[0138] Where: i The non-core vehicle distance association weight of the building / building complex, {t} is the set of all freight vehicle targets that can be retrieved in the direction quadrant of the building / building complex, and {u} is the parked freight vehicle target p j The set of cargo buildings / building complexes retrieved in the directional quadrant, η j The default weight for non-core vehicle distance in the region.

[0139] Furthermore, this embodiment also provides a system for identifying enterprise freight facilities based on multiple features, including:

[0140] An image acquisition module acquires first image data of a building area and performs a first preprocessing on the first image data to obtain second image data;

[0141] An input module, presetting a first building freight facility discrimination model, and inputting the second image data into the first building freight facility discrimination model;

[0142] The discrimination output module performs intelligent discrimination of freight facilities according to the output of the freight facility discrimination model of the first building.

[0143] In summary, the present invention simulates the human eye scanning when observing satellite images, and gradually observes each part of the image. It has a very strong positive significance for remote sensing images with very large scales, retains the detail information in the image to a greater extent, and establishes a far-end neglect hypothesis. By calculating the similarity of building targets, the similarity between two or more targets is determined, and a freight facility building complex is established in the building area with target similarity. The core freight vehicle association weight and the non-core vehicle distance association weight of each building and freight facility building complex are calculated, and a semi-supervised learning freight facility discrimination model is established. The model is continuously improved by teaching the logistic regressor and the student logistic regressor, and the integrated learning method is used to learn the distribution of model discrimination, thereby solving the problem that simply extracting the building from the building complex to which it belongs for spatial analysis will inevitably lead to incorrect classification and cognition of other samples.

[0144] Example 2

[0145] Reference Figure 2 - Figure 7 , which is the second embodiment of the present invention, provides a method for identifying enterprise freight facilities based on multiple features. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0146] The broadcast of non-core vehicle distance association weights in the building complex. Through such distance topological relationship modeling, the weight calculation of parked freight vehicles with high correlation with facilities and freight vehicles with low correlation is fundamentally distinguished, and the subjective prior knowledge of the manual annotator on the overall situation of the area is fully considered. The retrieval mode based on the direction quadrant can be applied to most cases, and the maximum number of building target objects involved in the calculation can be limited to 8. This model can more accurately capture the spatial relationship between freight facilities and parked freight vehicles, providing a more reliable basis for subsequent data analysis and decision-making. Through reasonable weight calculation and distance topological relationship modeling, the present invention can deeply characterize the parking pattern of freight vehicles.

[0147] Labeling and labeled training of freight facilities is an important part of discriminant model construction. Through labeling and labeled training, the model can better understand the relationship between input data and output labels. In order to better describe the spatial characteristics of freight building facilities, the present invention uses the following features to characterize freight facilities and their surrounding environment, as shown in Table 1 below.

[0148] Table 1 Format of freight facility identification feature data entries

[0149]

[0150] In supervised learning, there is usually a set of labeled training data and a model whose goal is to predict unlabeled data. In the current practical application scenario of freight facility identification, labeled data is very scarce, while unlabeled data exists in large quantities. The goal of using semi-supervised learning is to use the numerical spatial distribution of these unlabeled data and labeled data to comprehensively improve the performance of the model.

[0151] In practical applications of the present invention, unlabeled data may not always come from the same distribution, which may cause the model to be biased or have insufficient generalization ability when using these data. However, for efficiency reasons, the present invention assumes that at least the characteristic values ​​proposed by freight building facilities in the study area (core urban area of ​​Shanghai) have a clear and identical distribution. In addition, the present invention infers that the facility type discrimination based on satellite image information conforms to the continuity hypothesis and the clustering hypothesis. The present invention selects the ensemble learning method of the voting regressor to learn the distribution of model discrimination.

[0152] The present invention uses soft labels to mark whether a building facility is an active freight building facility, and converts the problem into a logistic regression problem represented by a probability distribution. The output of the model is shown in the following formula:

[0153] Output m =[1-P(A),P(A)](4.12).

[0155] Where: Output m is the output of the logistic regressor, and event A is that the building is an active freight building facility.

[0156] The model building and training process of teaching logistic regressor is as follows Figure 6 As shown in the figure, the teaching model architecture and training process, the teaching logistic regressor is mainly divided into three parts: encoder, noisy decoder and clean decoder. The encoder is used to convert to representations of other dimensions. It can compress the input data into a more representative feature vector. The clean decoder is only responsible for generating the encoder and remapping the feature output into the probability distribution output expected by the model. The output of the noisy decoder is also a probability distribution output, but the noisy decoder is responsible for combining the output of the encoder with a certain degree of noise to generate a more robust feature representation. This process of adding noise helps the model learn features with more generalization capabilities, thereby improving the performance of the model. The noise-adding method of the noisy decoder is to offset the encoder output tensor f(x) with Gaussian noise. In order to ensure that the feature offset does not exceed a reasonable range, the Gaussian noise is added as shown in the following formula:

[0157]

[0158]

[0159] Formula 4.13.1 shows the numerical relationship between the probability of the Gaussian random error ε taking a value and the hyperparameter Gaussian distribution mean μ and Gaussian distribution variance σ. is the object value to which the error is added, and α is the allowable parameter used to describe ε i The maximum value of is in the range of 0 to 1.

[0160] Semi-supervised training is used in the process of training and teaching the model. The loss function consists of two parts. The first part is the supervised loss when the input sample is a labeled sample, and the unsupervised loss that does not rely on the label. The supervised loss is used to measure the difference between the predicted output of the model and the true label, and guide the model to directly learn the distribution pattern of the labeled samples. The supervised loss enables the model to be trained in a targeted manner based on the labeled data, improving the performance of the model on known categories. The unsupervised loss is mainly used to use unlabeled samples to further guide the training of the model, although the unsupervised loss will also impose a penalty on the model when running labeled samples. The unsupervised loss mainly enhances the smoothness of the model output by measuring the difference between the noisy output and the clean output. As mentioned in the previous content, the degree of change in the noise is strictly limited by the allowable parameters. Under this restriction, the output of the model should not show obvious output distribution differences due to small perturbations, whether it is a labeled or unlabeled sample.

[0161] The loss function calculation of the teaching model is as follows:

[0162]

[0163] Where: t is the event probability distribution output by the clean decoder, is the event probability distribution output by the noise decoder; q(t) is its corresponding actual expected value.

[0164] The ensemble learning method of the voting regressor is used to learn the distribution of model discrimination. Because under the nonlinear feature fitting mechanism, although multiple models with basically the same output distribution can be obtained through training, the deep feature learning inside the model is completely different. This difference often stems from the different ways of deep feature learning inside the model, including the choice of activation function, the design of network structure, parameter initialization and other factors. These factors will affect the model's representation learning process of the data, and then affect the performance of the model on different data. Therefore, although the output distributions of multiple models may be similar, there may be significant differences in their internal feature learning. However, this difference may bring differences in the generalization of the model, but it can also better simulate the personal preferences of human researchers when faced with different features.

[0165] Select the training of the student model, such as Figure 7As shown, the student model learns the distribution of data pseudo labels from the teaching model, and then becomes an independent regressor for prediction. The linear feature reconstructor is used to strengthen the internal differences between different student models. The integrated probability distribution after the voting system is calculated as follows:

[0166]

[0167] Where: y i is the probability distribution of the output of the i-th student model, and n is the total number of student models. The structural parameters of the student model are generally smaller than the sum of the structural parameters of the teaching parameters.

[0168] Example 3

[0169] This embodiment also provides a computer device, which is applicable to a method for identifying enterprise freight facilities based on multiple features, and includes a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement a forced oscillation detection and positioning method for a distribution network as proposed in the above embodiment.

[0170] This embodiment further provides a storage medium on which a computer program is stored. When the program is executed by a processor, a forced oscillation detection and positioning method for a distribution network is implemented as proposed in the above embodiment.

[0171] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0172] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0173] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0174] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering, or processing in another suitable manner as necessary, and then stored in a computer memory.

[0175] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used to implement: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0176] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for identifying enterprise freight facilities based on multiple features, characterized by: include, Acquire first image data of a building area, and perform first preprocessing on the first image data to obtain second image data; Presetting a first building freight facility discrimination model, and inputting the second image data into the first building freight facility discrimination model; Intelligent identification of freight facilities is performed based on the output of the first building freight facility identification model.

2. The enterprise freight facility identification method based on multiple features as claimed in claim 1, characterized in that: The first image data is building area data obtained by performing building segmentation on a satellite image, and the first preprocessing includes: Perform distance verification on the building area and identify the building facilities most relevant to the target; The image is divided into multiple local areas through a sliding window with a fixed step size and size, and a sliding window with the centroid of the target freight building facility as the centroid of the sliding window is constructed.

3. The enterprise freight facility identification method based on multiple features as claimed in claim 2, characterized in that: The first building freight facility discrimination model includes: Building target similarity calculation, used to calculate the similarity between building areas; Building complex freight weight calculation, used to calculate the core freight vehicle association weight and non-core vehicle distance association weight of the building and freight facility complex; Semi-supervised learning discriminative unit for discriminating whether a building is a cargo facility by teaching a logistic regressor and a student logistic regressor.

4. The enterprise freight facility identification method based on multiple features as claimed in claim 3, characterized in that: The building target similarity calculation includes: A geometric similarity calculation unit is used to calculate the similarity of geometric features such as shape and size of buildings; The texture similarity calculation unit is used to calculate the similarity of texture features such as hue, purity and brightness of the building.

5. The enterprise freight facility identification method based on multiple features as claimed in claim 4, characterized in that: The building complex freight weight calculation includes: A core vehicle association weight calculation unit is used to calculate the vehicle weight within the core area of ​​a building / building complex; The non-core vehicle distance association weight calculation unit is used to calculate the vehicle weight in the non-core area of ​​the building / building complex.

6. The enterprise freight facility identification method based on multiple features as claimed in claim 5, characterized in that: The semi-supervised learning discrimination unit comprises: Teach a logistic regressor, including an encoder, a noisy decoder, and a clean decoder; Student logistic regressor, used to learn the data distribution from the taught model and make predictions.

7. The enterprise freight facility identification method based on multiple features as claimed in claim 6, characterized in that: The model operation module performs feature characterization on the input data, including multi-dimensional feature data of building area, density, rectangularity, eccentricity, main axis direction angle, number of vertices, and clustering information.

8. A system for identifying enterprise freight facilities based on multiple features, based on the method for identifying enterprise freight facilities based on multiple features according to any one of claims 1 to 7, characterized in that: Also includes, An image acquisition module acquires first image data of a building area and performs a first preprocessing on the first image data to obtain second image data; An input module, presetting a first building freight facility discrimination model, and inputting the second image data into the first building freight facility discrimination model; The discrimination output module performs intelligent discrimination of freight facilities according to the output of the first building freight facility discrimination model.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, it implements the steps of the enterprise freight facility identification method based on multiple features as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the enterprise freight facility identification method based on multiple features described in any one of claims 1 to 7 are implemented.

Images