A wall water seepage intelligent identification method based on big data

Abstract

The application discloses a kind of wall water seepage intelligent identification method based on big data, method includes obtaining multi-source data, wall water seepage area segmentation, wall water seepage type identification, wall water seepage risk evolution prediction and wall water seepage comprehensive identification.The application relates to the field of data processing, specifically refers to a kind of wall water seepage intelligent identification method based on big data, the scheme innovatively proposes hierarchical identification method of wall water seepage, constructs wall water seepage area segmentation, wall water seepage type identification and wall water seepage evolution trend identification composition three-level hierarchy system, significantly improve the accuracy of wall water seepage identification;By constructing the improved area segmentation clustering algorithm containing noise suppression design clustering target function, design feature weight self-adaptive adjustment mechanism, and introducing dynamic membership update strategy, the accurate identification of water seepage area is realized;Propose the improved convolutional neural network based on local difference difference binary aggregation strategy, effectively improve the water seepage type identification accuracy.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, and particularly refers to a wall water seepage intelligent identification method based on big data. BACKGROUND

[0002] The wall water seepage intelligent identification method based on big data is a method for intelligently identifying the wall water seepage phenomenon of a building. By using big data technology to collect, fuse and analyze multi-source environmental data in real time, the target area of wall water seepage can be accurately identified, so as to improve the identification efficiency of wall water seepage problems, optimize building maintenance schemes and reduce maintenance costs.

[0003] However, in the traditional wall water seepage identification method, there is a technical problem that it relies on a single image recognition method and cannot simultaneously identify the water seepage area, the water seepage type and the future evolution trend, resulting in fragmented identification results, poor real-time performance and insufficient accuracy. In the clustering algorithm of the existing wall water seepage area segmentation model, there is a technical problem that it is sensitive to noise, cannot reasonably distinguish the importance of features, and is difficult to handle the fuzzy boundary of the water seepage area, resulting in unstable water seepage area segmentation results, discontinuous boundaries and insufficient precision. In the existing wall water seepage type identification model, there is a technical problem that it cannot effectively extract local detailed features of wall water seepage, has insufficient discrimination for multiple types of water seepage, and has weak adaptability to complex wall surface scenes, resulting in low accuracy of water seepage type identification. SUMMARY

[0004] In view of the above, in order to overcome the defects of the prior art, the present application provides a wall water seepage intelligent identification method based on big data, in order to solve the technical problems that the traditional wall water seepage identification method relies on a single image recognition method, cannot simultaneously identify the water seepage area, water seepage type and future evolution trend, and leads to fragmented identification results, poor real-time performance and insufficient accuracy, the present application innovatively proposes a hierarchical wall water seepage identification method, constructs a three-level hierarchical system composed of wall water seepage area segmentation, wall water seepage type identification and wall water seepage evolution trend identification, and performs layer-by-layer identification from the spatial dimension, semantic dimension and time dimension, realizes multi-level deep identification of wall water seepage, changes the identification method from single feature judgment to multi-dimensional information fusion collaborative judgment, and thus significantly improves the accuracy, stability and integrity of the identification results of wall water seepage identification; in view of the technical problems that the clustering algorithm suitable for the existing wall water seepage area segmentation model is sensitive to noise, cannot reasonably distinguish the importance of features, and is difficult to handle the fuzzy boundary of the water seepage area, which leads to unstable water seepage area segmentation results, discontinuous boundary and insufficient precision, the present application innovatively constructs an improved area segmentation clustering algorithm containing noise suppression design clustering target function, designs a feature weight self-adaptive adjustment mechanism, and introduces a dynamic membership update strategy, suppresses the interference of noise on water seepage segmentation, can automatically enhance the response to key features of water seepage, significantly enhances the anti-noise ability, improves the water seepage area boundary precision, improves the anti-noise ability of the model, enhances the water seepage edge recognition effect, and realizes the self-adaptive optimization of key feature weights, so that the water seepage area segmentation precision, boundary integrity and multi-scene adaptability are comprehensively improved, and the wall water seepage area is stably, accurately and high-quality identified; in view of the technical problems that the existing wall water seepage type identification model cannot effectively extract local detail features of wall water seepage, has insufficient discrimination for multiple types of water seepage, and has weak adaptability to complex wall surface scenes, which leads to low water seepage type identification accuracy, the present application innovatively proposes an improved convolutional neural network based on a local difference binary aggregation strategy, and applies it to wall water seepage type identification, introduces a local difference matrix to strengthen local water seepage difference features, proposes a local difference adaptive threshold mechanism and designs a local difference binary aggregation strategy, significantly enhances the local feature extraction capability, improves the sensitivity to weak water seepage and small water seepage area, enhances the anti-interference ability and multi-scene adaptability of the model, and thus effectively improves the water seepage type identification accuracy, realizes high-precision, lightweight and intelligent identification of wall water seepage type.

[0005] The technical scheme adopted by the present application is as follows: the present application provides a wall water seepage intelligent identification method based on big data, which comprises the following steps:

[0006] Step S1: acquiring multi-source data;

[0007] Step S2: wall water seepage area segmentation;

[0008] Step S3: wall surface water seepage type identification;

[0009] Step S4: wall surface water seepage risk evolution prediction;

[0010] Step S5: wall surface water seepage comprehensive identification.

[0011] Further, in step S1, the multi-source data is obtained, specifically, wall surface water seepage management original data is obtained through data collection operation, and data optimization operation is performed on the wall surface water seepage management original data to obtain wall surface water seepage management optimized data; the wall surface water seepage management original data includes historical wall surface image data, historical wall surface water seepage evolution data, real-time wall surface image data and real-time wall surface water seepage evolution data;

[0012] The data optimization operation includes image data optimization processing, structure data optimization processing and data synchronization processing.

[0013] Further, in step S2, the wall surface water seepage area is segmented, specifically including the following step process:

[0014] Step S21: wall surface image feature extraction, specifically, the wall surface image in the wall surface water seepage management optimized data is divided into N samples according to the granularity of superpixels, and the features of each sample are extracted, then the features are normalized to obtain a sample vector set;

[0015] Step S22: constructing a wall surface water seepage area segmentation model, specifically including the following step process:

[0016] Step S221: cluster center initialization, specifically, first set C cluster centers, then randomly select a sample from the sample vector set as the first initial cluster center , and determine the remaining initial cluster centers in turn according to the shortest distance probability selection mechanism, integrate all initial cluster centers to obtain an initial cluster center set;

[0017] Step S222: noise parameter initialization, specifically, calculate the initial fuzzy membership of each sample to each cluster by fuzzy C-means algorithm, then for each cluster, combine the initial fuzzy membership to calculate the noise coefficient of the feature distribution of the cluster, and finally calculate the noise distance of each sample to each cluster based on the noise coefficient and the initial fuzzy membership;

[0018] Step S223: design a clustering objective function, specifically, construct a clustering objective function containing a soft subspace clustering term and a noise suppression term, and optimize the feature weight through relaxation constraint; the formula is as follows:

[0019] ;

[0020] wherein, denotes the weighted feature distance of the ith sample to the cth cluster in the tth iteration, J denotes the cluster objective function value, C denotes the number of clusters, denotes the number of features, k denotes the feature dimension index, denotes the weight of the kth feature under the cth cluster in the tth iteration, denotes the hybrid membership of the ith sample to the cth cluster in the tth iteration, denotes the noise distance of the ith sample under the cth cluster, m denotes the membership exponent;

[0021] Step S224: iteratively optimizing the clustering parameters, specifically, by setting the current iteration number t = 1, cyclically updating the membership, cluster center and feature weight, and gradually minimizing the objective function value; including the following steps:

[0022] Step S2241: updating the hybrid membership, specifically, based on the current iteration number, using a dynamic membership fusion strategy, combining the hard membership and fuzzy membership to dynamically update the hybrid membership of each sample to the cluster, the formula used is as follows:

[0023] ;

[0024] ;

[0025] ;

[0026] wherein, denotes the membership fusion weight in the tth iteration, denotes the fuzzy membership of the ith sample to the cth cluster in the tth iteration, denotes the hard membership of the ith sample to the cth cluster in the tth iteration, denotes the weighted feature distance of the ith sample to the lth cluster in the tth iteration, denotes, for the ith sample, calculating its weighted feature distance to all clusters, finding the cluster index with the minimum distance ;

[0027] Step S2242: updating the cluster center, specifically, according to the updated hybrid membership, updating the cluster center of each cluster category according to the weighted average principle; the formula used is as follows:

[0028] ;

[0029] Step S2243: updating the feature weight, specifically, based on the clustering center and the mixed membership of the current iteration, the feature weight of each class is adaptively adjusted, so that the feature more sensitive to the wall seepage identification obtains higher weight; the formula used is as follows:

[0030] ;

[0031] In the formula, represents the dth feature value of the ith sample, represents a relaxation factor, d represents a feature dimension index, represents the dth feature value of the cth clustering center sample in the tth iteration;

[0032] Step S2244: clustering iteration termination judgment, specifically, the clustering objective function value in the current iteration is recalculated , and compared with the objective function value in the last iteration , by calculating the difference between the two, the objective function change value is obtained, finally, according to the preset termination condition, the iteration termination judgment is carried out, if any iteration termination condition is met, the clustering iteration is stopped, and the final clustering center set, mixed membership matrix and feature weight matrix are outputted;

[0033] Step S225: clustering result output, specifically, according to the final mixed membership matrix, for each sample, the class with the maximum mixed membership is selected as the final clustering class, so as to divide all samples into multiple independent clusters, and the image segmentation clustering result is obtained;

[0034] Step S23: wall seepage area segmentation result output, specifically, the historical wall image data and the real-time wall image data are inputted into the wall seepage area segmentation model respectively, the seepage area segmentation processing is performed on all wall images, the corresponding historical image segmentation clustering result and real-time image segmentation clustering result are obtained, and in the respective segmentation clustering result, the cluster set whose clustering class belongs to the seepage area class is selected as the seepage cluster set of the image, finally, the seepage area feature extraction is carried out from the seepage cluster set, so as to obtain the historical seepage area feature set and the real-time seepage area feature set.

[0035] Further, in step S3, the wall seepage type identification specifically includes the following steps:

[0036] Step S31: establishing a wall seepage type identification model, specifically, an improved convolutional neural network based on local difference binary aggregation strategy is introduced to construct a wall seepage type identification model; the improved convolutional neural network includes convolution feature extraction block, pooling layer based on local difference binary aggregation, global average pooling layer and full connection classification layer connected in turn;

[0037] The pooling layer based on local difference binary aggregation specifically includes local difference matrix calculation, local difference adaptive threshold calculation, local difference binary pattern aggregation, final pooling value calculation, and sliding window traversal.

[0038] The local difference matrix calculation specifically defines a local window by sliding a preset window size on the multi-scale convolution feature map, and calculates the gray difference between each pixel and the center pixel in each local window to obtain a local difference matrix.

[0039] The local difference adaptive threshold calculation specifically calculates the average absolute deviation of all elements in the local difference matrix as the local difference adaptive threshold.

[0040] The local difference binary pattern aggregation specifically calculates a binary approximation value for each element in the local difference matrix by function, forms a local difference binary pattern, and then sums all elements of all local difference binary patterns to obtain a local aggregation value.

[0041] The final pooling value calculation specifically calculates a pooling weighting factor based on the local aggregation value , and obtains a final pooling value by weighting and fusing the local window standard deviation and the local window maximum value through the pooling weighting factor.

[0042] The sliding window traversal specifically slides the local window along the width and height directions of the multi-scale convolution feature map by a step size s, and then sequentially performs the local difference matrix calculation, the local difference adaptive threshold calculation, the local difference binary pattern aggregation, and the final pooling value calculation operations on each local window to generate a pooling classification feature map.

[0043] Step S32: wall surface water seepage type recognition model training, specifically using the historical water seepage area feature set as training data to train the recognition model, and finally obtaining the trained wall surface water seepage type recognition model.

[0044] Step S33: real-time wall surface water seepage type recognition, specifically inputting the real-time water seepage area feature set into the trained wall surface water seepage type recognition model to obtain a real-time wall surface water seepage type recognition result.

[0045] Further, in step S4, the wall surface water seepage evolution prediction specifically includes the following steps:

[0046] Step S41: establishing a wall surface water seepage evolution prediction model, specifically establishing a wall surface water seepage evolution prediction model based on a long short-term memory network.

[0047] Step S42: wall surface water seepage evolution prediction model training, specifically, historical wall surface water seepage evolution data is taken as training data to train the prediction model, and finally a trained wall surface water seepage evolution prediction model is obtained.

[0048] Step S43: wall surface water seepage evolution real-time prediction, specifically, real-time wall surface water seepage evolution data is input into the trained wall surface water seepage evolution prediction model to obtain real-time wall surface water seepage evolution results.

[0049] Further, in step S5, the wall surface water seepage comprehensive identification is specifically to comprehensively identify the current wall surface water seepage type and state of the target wall surface and the wall surface water seepage evolution trend according to the real-time wall surface water seepage type identification result and the real-time wall surface water seepage evolution result, to obtain the overall wall surface water seepage condition of the target wall surface, and to realize intelligent identification of the wall surface water seepage condition of the target wall surface.

[0050] The beneficial effects achieved by the above-mentioned scheme are as follows:

[0051] (1) In view of the technical problem that the traditional wall surface water seepage identification method relies on a single image recognition method, cannot simultaneously identify the water seepage area, water seepage type and future evolution trend, and leads to fragmented identification results, poor real-time performance and insufficient accuracy, the present scheme innovatively proposes a hierarchical wall surface water seepage identification method, constructs a three-level hierarchical system composed of wall surface water seepage area segmentation, wall surface water seepage type identification and wall surface water seepage evolution trend identification, and performs layer-by-layer identification from the spatial dimension, semantic dimension and time dimension, to realize multi-level deep identification of wall surface water seepage, change the identification method from single feature judgment to multi-dimensional information fusion collaborative judgment, and thus significantly improve the accuracy, stability and completeness of the identification results.

[0052] (2) In view of the technical problem that the existing clustering algorithm suitable for wall surface water seepage area segmentation model is sensitive to noise, cannot reasonably distinguish the importance of features, and is difficult to handle the fuzzy boundary of the water seepage area, thus leading to unstable water seepage area segmentation results, discontinuous boundaries and insufficient precision, the present scheme innovatively constructs an improved area segmentation clustering algorithm containing noise suppression design clustering objective function, designs a feature weight self-adaptive adjustment mechanism, and introduces a dynamic membership update strategy, to suppress the interference of noise on water seepage segmentation, automatically enhance the response to key features of water seepage, significantly enhance the anti-noise ability, improve the boundary precision of water seepage area, and improve the anti-noise ability of the model, enhance the water seepage edge recognition effect, and realize self-adaptive optimization of key feature weights, so as to comprehensively improve the water seepage area segmentation precision, boundary integrity and multi-scene adaptability, and realize stable, accurate and high-quality identification of the wall surface water seepage area.

[0053] (3) In view of the technical problems that in the existing wall water seepage type identification model, local detail features of wall water seepage cannot be effectively extracted, the discrimination degree for multiple types of water seepage is insufficient, and the adaptability to complex wall surface scenes is weak, resulting in low water seepage type identification accuracy, the scheme innovatively proposes an improved convolutional neural network based on local difference binary aggregation strategy, and applies it to wall water seepage type identification, by introducing a local difference matrix, strengthening local water seepage difference features, proposing a local difference adaptive threshold mechanism and designing a local difference binary aggregation strategy, the local feature extraction capability is significantly enhanced, the sensitivity to weak water seepage and small water seepage area is improved, the model anti-interference ability and multi-scene adaptability are enhanced, thereby effectively improving the water seepage type identification accuracy, realizing high-precision, lightweight and intelligent identification of wall water seepage type. BRIEF DESCRIPTION OF DRAWINGS

[0054] Figure 1 A flowchart of a wall water seepage intelligent identification method based on big data provided by the present application is shown in the figure.

[0055] Figure 2 A flowchart of step S2 is shown in the figure.

[0056] Figure 3 A flowchart of step S3 is shown in the figure.

[0057] Figure 4 A flowchart of step S4 is shown in the figure.

[0058] Figure 5 A flowchart of step S22 is shown in the figure.

[0059] The accompanying drawings are used to provide a further understanding of the present application, and constitute a part of the specification, together with the embodiments of the present application, to explain the present application, and do not constitute a limitation on the present application. DETAILED DESCRIPTION

[0060] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of the present application.

[0061] In the description of the present application, it should be understood that the terms "upper", "lower", "front", "back", "left", "right", "top", "bottom", "inner", "outer" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the systems or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the present application.

[0062] Embodiment one, refer to Figure 1 The technical scheme adopted by the present application is as follows: the present application provides a wall water seepage intelligent identification method based on big data, which comprises the following steps:

[0063] Step S1: acquiring multi-source data, specifically, obtaining wall water seepage management optimization data through data acquisition operation and data optimization operation;

[0064] Step S2: wall water seepage area segmentation, used for automatic identification and extraction of water seepage area in target wall image, realizing fine separation of water seepage area and non-water seepage area, specifically, first extracting wall image features, then improving clustering algorithm through noise suppression design clustering target function, feature weight self-adaptive adjustment and dynamic membership updating mechanism, completing wall water seepage area segmentation model construction, finally, inputting historical wall image data and real-time wall image data into the wall water seepage area segmentation model respectively to obtain historical water seepage area feature set and real-time water seepage area feature set;

[0065] Step S3: wall water seepage type identification, used for automatic classification of water seepage type of each wall water seepage area, distinguishing different water seepage forms, specifically, first establishing a wall water seepage type identification model by introducing an improved convolutional neural network based on local difference binary aggregation strategy, training the model using the historical water seepage area feature set to generate a trained wall water seepage type identification model, and finally inputting the real-time water seepage area feature set into the trained identification model to obtain real-time wall water seepage type identification result;

[0066] Step S4: wall water seepage evolution prediction, used for predicting the water seepage evolution trend of the target wall in the future time period on the premise that water seepage has occurred on the wall; specifically, constructing a wall water seepage evolution prediction model based on long short-term memory network and training the model, and finally inputting real-time data into the trained prediction model to obtain real-time wall water seepage evolution result;

[0067] Step S5: wall water seepage comprehensive identification, specifically, comprehensively identifying the wall water seepage condition according to the real-time wall water seepage type identification result and the real-time wall water seepage evolution result.

[0068] By performing the above operation, in order to solve the technical problems that the conventional wall seepage identification method relies on a single image identification method, cannot identify the seepage area, seepage type and future evolution trend at the same time, and leads to fragmented identification results, poor real-time performance and insufficient accuracy, the scheme innovatively proposes a hierarchical wall seepage identification method, constructs a three-level hierarchical system composed of wall seepage area segmentation, wall seepage type identification and wall seepage evolution trend identification, and identifies layer by layer from the spatial dimension, semantic dimension and time dimension, realizes multi-level deep identification of wall seepage, changes the identification method from single feature judgment to multi-dimensional information fusion collaborative judgment, and thus significantly improves the accuracy, stability and integrity of the identification results of wall seepage identification.

[0069] In the second embodiment, referring to Figure 1 The embodiment is based on the above-mentioned embodiment, in step S1, the multi-source data is obtained, specifically by performing data collection operation on the current building management platform to obtain wall seepage management original data, and performing data optimization operation on the wall seepage management original data to obtain wall seepage management optimized data;

[0070] The data optimization operation includes image data optimization processing, structure data optimization processing and data synchronization processing;

[0071] The wall seepage management original data includes historical wall image data, historical wall seepage evolution data, real-time wall image data and real-time wall seepage evolution data;

[0072] The historical wall image data and real-time wall image data both include wall image, image collection time and image position information;

[0073] The historical wall image data further includes historical wall seepage type identification result;

[0074] The historical wall seepage evolution data and real-time wall seepage evolution data both include environmental monitoring data;

[0075] The historical wall seepage evolution data further includes historical wall seepage evolution result, historical wall seepage type identification result and historical wall seepage area data;

[0076] The historical wall seepage evolution result includes stable type, diffusion type, aggregation type, migration type and attenuation type;

[0077] The wall image data is specifically collected by a mobile terminal inspection terminal, a fixed camera and a unmanned aerial vehicle to collect the images of the inner and outer walls of the building;

[0078] The historical wall seepage type identification result is used to indicate the seepage type of the wall at the historical moment, and is obtained by semi-automatic labeling of historical wall image data, including wet stain diffusion type, edge immersion type, point leakage type, mold spot complex type and crack immersion type.

[0079] The environmental monitoring data is used to represent the environmental state of the area where the building is located, including temperature, humidity, rainfall, wind direction and wind speed.

[0080] The wall seepage area data is specifically a geometric time sequence feature set obtained by quantifying the wall seepage state at each moment based on a seepage area feature set, including seepage area, seepage area perimeter, area number and area relative position distribution.

[0081] The image data optimization processing is used to obtain wall image optimization data, specifically including image denoising, image enhancement and image registration. The image denoising specifically removes random noise and background interference in the image through a Gaussian filter algorithm to ensure the clarity of the image. The image enhancement specifically enhances the visual features of the wall image through contrast enhancement, brightness adjustment and edge enhancement technology. The image registration is used to ensure the data consistency and accuracy of multi-angle images, specifically through a feature point matching algorithm to perform image alignment operation on wall images taken at different angles.

[0082] The structure data optimization processing is used to optimize the environmental monitoring data and wall seepage area data, to obtain environmental monitoring optimization data and wall seepage area optimization data, specifically including data cleaning and standardization processing.

[0083] The data cleaning is to handle missing values and outliers in the structure data, including missing value filling and outlier removal.

[0084] The missing value filling specifically fills the missing values through the mean filling method. The outlier removal specifically detects and removes extreme values and logical outliers in the original data through the IQR method.

[0085] The standardization processing is based on the maximum and minimum normalization method for data standardization processing, and the one-hot encoding method is used for encoding processing of the category field in the original data.

[0086] Data synchronization processing is used to time-align wall image data, environmental monitoring data and wall seepage area data. Specifically, based on the time stamp corresponding to each data, the wall image optimization data, environmental monitoring optimization data and wall seepage area optimization data are time-aligned, and heterogeneous data from different data sources are mapped to a unified time axis, thereby constructing wall seepage management optimization data with time sequence as the index dimension.

[0087] Embodiment three, refer to Figure 1 , Figure 2 and Figure 5 , based on the above embodiment, in step S2, the wall seepage area segmentation is used for automatic identification and extraction of the seepage area in the target wall image, realizing the fine separation of seepage area and non-seepage area, which specifically includes the following steps:

[0088] Step S21: wall image feature extraction, for extracting the feature vector set of the input wall seepage area segmentation model from the wall image in the wall image optimization data, specifically, dividing the wall image in the wall seepage management optimization data into N samples according to the granularity of superpixels, and extracting color features, texture features and spatial position features for each sample, then, normalizing the color features, texture features and spatial position features to obtain the sample vector set;

[0089] The division according to the granularity of superpixels specifically adopts the superpixel segmentation algorithm based on SLIC to divide the wall image into N samples; the sample is specifically a local area corresponding to the wall image;

[0090] The color feature is obtained by converting the wall image from the RGB color space to the HSV color space, and calculating the hue component, saturation component and brightness component of the pixels in the sample area;

[0091] The texture feature is obtained by calculating the local binary pattern code of the preset neighborhood of the pixels of the i-th sample, and counting the frequency of the LBP pattern in the sample area to form an LBP histogram, and taking the LBP histogram as the texture feature vector;

[0092] The spatial position feature is obtained by obtaining the center of gravity coordinates of the sample in the wall image, and normalizing the center of gravity coordinates according to the width and height of the wall image to obtain the spatial position feature;

[0093] The normalization processing specifically unifies the features of different dimensions and value ranges into a preset numerical interval;

[0094] The formula used is as follows:

[0095] ;

[0096] In the formula, represents the i-th sample, represents the color feature vector of the i-th sample, represents the texture feature vector of the i-th sample, which is calculated by local binary pattern, represents the spatial position feature of the i-th sample;

[0097] Step S22: constructing a wall seepage area segmentation model, for constructing a model for automatically segmenting the wall seepage area, specifically, improving the clustering algorithm through noise suppression design, feature weight self-adaptive adjustment, and dynamic membership update mechanism, so as to complete the wall seepage area segmentation model construction, including the following step process:

[0098] Step S221: clustering center initialization, for ensuring that the initial clustering center adapts to different categories of regions in the wall image; specifically, first set C clustering centers, then randomly select a sample from the sample vector set as the first initial clustering center , and sequentially determine the remaining initial clustering centers according to the shortest distance probability selection mechanism, integrate all initial clustering centers to obtain an initial clustering center set;

[0099] The shortest distance probability selection mechanism specifically calculates the Euclidean distance of each sample to all selected initial clustering centers, takes the minimum value as the shortest distance of the sample to the selected center, calculates the probability of each sample being selected as the c-th clustering center based on the shortest distance, and selects the sample with the maximum probability as the c-th initial clustering center ;

[0100] The formula used is as follows:

[0101] ;

[0102] In the formula, Pi,c represents the probability of the i-th sample being selected as the c-th clustering center, D(i,c) represents the shortest Euclidean distance of the i-th sample to the selected initial clustering center;

[0103] Step S222: noise parameter initialization, for quantifying the noise properties of each sample to different clusters in the presence of stains, light spot noise, etc. in the wall image, specifically, through the fuzzy C-means algorithm, calculate the initial fuzzy membership of each sample to each cluster, then for each cluster, combine the initial fuzzy membership to calculate the noise coefficient of the feature distribution of the cluster, and finally based on the noise coefficient and the initial fuzzy membership, calculate the noise distance of each sample to each cluster; the formula used is as follows:

[0104] ;

[0105] ;

[0106] ;

[0107] In the formula, denotes the initial fuzzy membership of the i-th sample to the c-th cluster, i.e., the probability of the i-th sample belonging to the c-th cluster, , denotes the Euclidean distance calculation function, m denotes the membership exponent for controlling the fuzzy degree of membership, and the value range is , denotes the noise coefficient sensitivity parameter, and the value range is , denotes the square of the Euclidean distance, denotes the noise distance of the i-th sample under the c-th cluster, and the smaller the value is, the more likely the pixel is misjudged as noise, denotes the l-th initial cluster center, and C denotes the number of clusters, denotes the noise coefficient of the c-th cluster.

[0108] Step S223: design a clustering objective function for providing a quantitative optimization target for wall seepage area multi-cluster segmentation, significantly reducing missegmentation caused by wall reflection, wall skin spots and noise points, and improving the purity and stability of the seepage area boundary, specifically constructing a clustering objective function containing a soft subspace clustering term and a noise suppression term, and optimizing the distribution of feature weights through relaxation constraints;

[0109] The soft subspace clustering term is used to measure the weighted distance of each pixel in different category subspaces, so that the wall seepage area pixels are closer to the corresponding cluster center in different feature dimensions;

[0110] The noise suppression term is used to jointly constrain the membership and noise distance, and improve the robustness of wall seepage area segmentation in complex environments;

[0111] The formula used is as follows:

[0112] ;

[0113] ;

[0114] In the formula, denotes the weighted feature distance of the i-th sample to the c-th cluster in the t-th iteration, and J denotes the value of the clustering objective function, denotes the number of features, and k denotes the feature dimension index, denotes the weight of the k-th feature under the c-th cluster in the t-th iteration, and the weight is initialized as a uniform distribution, denotes the relaxation factor, which is set to 2, and is used to amplify the difference between feature weights and strengthen the contribution of important features to clustering, denotes the weight of the d-th feature under the c-th cluster in the t-th iteration, and is used to adjust the influence degree of different features on the distance of the cluster, denotes the kth feature value of the ith sample, denotes the kth feature value of the cth cluster center sample in the tth iteration, denotes the hybrid membership of the ith sample to the cth cluster in the tth iteration, denotes the constraint condition that must be met in the explicit objective function optimization process, d denotes the feature dimension index;

[0115] Step S224: iteratively optimize the clustering parameters, specifically, by setting the current iteration number t = 1, cyclically updating the membership, cluster center and feature weight, and gradually minimizing the objective function value; including the following steps:

[0116] Step S2241: update the hybrid membership, which is used to update the membership of each sample to each cluster, thereby adaptively adjusting the class attribution of each wall surface water seepage area, solving the problems of seepage edge blur, color gradient and diffusion transition, making the segmentation result smoother and the boundary more stable; specifically, based on the current iteration number, a dynamic membership fusion strategy is used to dynamically update the hybrid membership of each sample to the cluster by combining the hard membership and fuzzy membership, and the formula used is as follows:

[0117] ;

[0118] ;

[0119] ;

[0120] ;

[0121] In the formula, denotes the maximum iteration number, denotes the membership fusion weight in the tth iteration, denotes the fuzzy membership of the ith sample to the cth cluster in the tth iteration, denotes the hard membership of the ith sample to the cth cluster in the tth iteration, if , it indicates that the ith sample belongs to the cth cluster in the tth iteration, if , it indicates that the ith sample does not belong to the cth cluster in the tth iteration, denotes the weighted feature distance of the ith sample to the lth cluster in the tth iteration, denotes the calculation of the weighted feature distance of the ith sample to all clusters, and the cluster index with the minimum distance is found ;

[0122] Step S2242: updating the cluster center, for making the cluster center always fit the actual feature distribution of each category of the wall surface, specifically, updating the cluster center of each cluster category according to the updated hybrid membership degree according to the weighted average principle; the formula used is as follows:

[0123] ;

[0124] Step S2243: updating the feature weight, for adaptively adjusting the relative importance of the feature in the clustering process, thereby improving the distinguishing ability of the feature subspace to the water seepage area of the wall surface; specifically, based on the cluster center and the hybrid membership degree of the current iteration, adaptively adjusting the feature weight of each category, so that the feature more sensitive to the water seepage recognition obtains a higher weight; the formula used is as follows:

[0125] ;

[0126] In the formula, represents the dth feature value of the ith sample, represents the dth feature value of the cth cluster center sample in the tth iteration;

[0127] Step S2244: cluster iteration termination judgment, for determining whether the model reaches a convergence state according to the change degree of the objective function, thereby ensuring the stability of the wall surface water seepage area segmentation result; specifically, recalculating the cluster objective function value in the current iteration , and comparing it with the objective function value in the last iteration , obtaining the objective function change value by calculating the difference between the two, and finally, according to the preset termination condition, the iteration termination judgment is made, if any of the iteration termination conditions is met, the clustering iteration is stopped, and the final cluster center set, hybrid membership degree matrix and feature weight matrix are outputted;

[0128] The iteration termination conditions include that the objective function change value is less than the iteration convergence threshold and the current iteration number reaches the maximum iteration number; the formula used is as follows:

[0129] ;

[0130] Step S225: cluster result output, specifically, according to the final hybrid membership degree matrix, for each sample, selecting the category with the maximum hybrid membership degree as the final cluster category, thereby dividing all samples into multiple independent clusters, obtaining the image segmentation clustering result; the cluster category refers to the category division based on the clustering result to the wall surface image sample, for identifying the category mark of different wall surface water seepage states;

[0131] Preferably, the clustering categories include two categories of wall seepage areas and normal walls; optionally, when different seepage degrees need to be distinguished, the clustering categories can also include light seepage areas, moderate seepage areas, heavy seepage areas and normal walls, so as to realize multi-category fine division of wall seepage areas; the cluster corresponds to a group of wall image samples with high similarity in the feature space; the category division is specifically that, according to the distribution of each clustering center in the feature dimension in the clustering result, feature analysis is performed on each clustering center, and it is one-to-one mapped with the preset wall state category; preferably, the clustering center with low gray value, high texture entropy and large gradient amplitude is mapped to the wall seepage area category, and the clustering center with high gray value, low texture entropy and small gradient amplitude is mapped to the wall non-seepage category; optionally, when multiple seepage degree categories are set, the seepage degree can also be graded according to the strength of the above features;

[0132] Step S23: wall seepage area segmentation result output, specifically, historical wall image data and real-time wall image data are input into the wall seepage area segmentation model, seepage area segmentation processing is performed on all wall images, corresponding historical image segmentation clustering results and real-time image segmentation clustering results are obtained, and in the respective segmentation clustering results, the cluster set whose clustering category belongs to the seepage area category is selected as the seepage cluster set of the wall image, and finally seepage area feature extraction is performed from the seepage cluster set, so as to obtain a historical seepage area feature set and a real-time seepage area feature set;

[0133] The seepage area feature extraction is specifically that, for each seepage cluster, based on the spatial position of the samples in the original wall image, connectivity analysis is performed, samples adjacent to each other and belonging to the cluster are merged into multiple seepage connected regions, and a region feature vector for representing the seepage form is calculated on each seepage connected region, the feature vectors of all seepage connected regions in the same wall image are combined according to the region dimension, forming the seepage area feature set corresponding to the wall image, realizing the conversion from the cluster-level clustering result to the region-level seepage feature set;

[0134] The region features include shape geometric features, texture features, gray statistical features and spatial distribution features;

[0135] The shape geometric features include seepage area, perimeter, aspect ratio, circumscribed rectangle area, circularity and edge directionality;

[0136] The gray statistical features include average gray value, gray variance, gray skewness and gray kurtosis.

[0137] By performing the above operation, for the technical problems of existing clustering algorithms suitable for wall water seepage area segmentation model are sensitive to noise, cannot reasonably distinguish feature importance, and are difficult to handle the fuzzy boundary of water seepage area, resulting in unstable water seepage area segmentation result, discontinuous boundary and insufficient precision, the scheme innovatively constructs a noise suppression design clustering objective function, designs a feature weight self-adaptive adjustment mechanism, and introduces a dynamic membership update strategy to improve the region segmentation clustering algorithm. The interference of noise on water seepage segmentation is suppressed, the response to key features of water seepage can be automatically enhanced, the anti-noise ability is significantly enhanced, the water seepage boundary precision is improved, the anti-noise ability of the model is improved, the water seepage edge recognition effect is enhanced, and the adaptive optimization of key feature weight is realized. The water seepage area segmentation precision, boundary integrity and multi-scene adaptability are comprehensively improved, and the robust, accurate and high-quality identification of the wall water seepage area is realized.

[0138] In an embodiment four, referring to Figure 1 and Figure 3 , the embodiment is based on the above-mentioned embodiments, in step S3, the wall water seepage type identification is used to identify the water seepage type of each water seepage area in real time; specifically including the following steps:

[0139] Step S31: Establish a wall water seepage type identification model, specifically by introducing an improved convolutional neural network based on a local difference binary aggregation strategy to construct a wall water seepage type identification model; the improved convolutional neural network includes convolutional feature extraction blocks, a pooling layer based on local difference binary aggregation, a global average pooling layer and a fully connected classification layer connected in turn;

[0140] The convolutional feature extraction block is used for multi-scale feature extraction of the input water seepage area, specifically each convolutional feature extraction block sequentially passes through a convolution layer, a batch normalization layer and a nonlinear activation layer to obtain a multi-scale convolutional feature map, wherein the convolution layer uses a preset size of convolution kernel to perform local convolution operation on the input data, the batch normalization layer is used to normalize the convolution output result, and the nonlinear activation layer preferably uses a ReLU activation function to enhance the nonlinear representation ability of the network; the formula used is:

[0141] ;

[0142] In the formula, denotes a multi-scale convolutional feature map, denotes a batch normalization layer operation, denotes a convolution operation, denotes model input data;

[0143] The pooling layer based on local difference binary aggregation is used to enhance the response capability to wall surface detail features while reducing the spatial resolution of multi-scale convolution feature maps, improve the discrimination of the wall surface water seepage type recognition model for different water seepage forms, and specifically includes local difference matrix calculation, calculation of local difference adaptive threshold, local difference binary mode aggregation, calculation of final pooling value, and sliding window traversal.

[0144] The local difference matrix calculation is used to highlight the water seepage difference of the wall surface water seepage area, specifically, a local window is defined by sliding a preset window size on the multi-scale convolution feature map, and in each local window, the gray difference between each pixel and the center pixel is calculated to obtain a local difference matrix; the formula is as follows:

[0145] ;

[0146] In the formula, represents the gray difference value between the pixel and the center pixel, that is, the element value of the local difference matrix at , represents the feature value of in the local window, represents the feature value of the center pixel of the local window, represents any pixel coordinate in the local window, represents the center pixel coordinate of the local window;

[0147] The calculation of the local difference adaptive threshold is used to adapt to the gray difference characteristics of different wall surface materials and different water seepage types, and avoid missed detection and false detection caused by fixed threshold; specifically, the average absolute deviation of all elements of the local difference matrix is calculated as the local difference adaptive threshold; the formula is as follows:

[0148] ;

[0149] In the formula, represents the local difference adaptive threshold, represents a local window with a size of , represents the total number of pixels in the window;

[0150] The local difference binary mode aggregation is used to quantify the water seepage feature density in the local window and adapt to the feature difference of different water seepage types; specifically, for each element in the local difference matrix, a binary approximation value is calculated by function to form a local difference binary mode, and then the sum of all elements of all local difference binary modes is calculated to obtain a local aggregation value; the formula is as follows:

[0151] ;

[0152] ;

[0153] wherein, denotes the feature binarization approximation value at the local window coordinate ; denotes the slope control parameter, used to regulate the sharpness of the binarization approximation, and adapt to the dynamic range of the gray difference of each wall seepage area, The larger the value is, the closer the binarization is to the hard threshold, denotes the local aggregation value, used to distinguish the feature density of different seepage types;

[0154] The final pooling value is calculated, which is used to fuse the contrast features and dominant intensity features of the local window, and adapt to the multi-dimensional distinguishing needs of the wall seepage type; specifically, the pooling weighting factor is calculated based on the local aggregation value The local window standard deviation and the local window maximum value are weighted and fused by the pooling weighting factor to obtain the final pooling value; the formula used is as follows:

[0155] ;

[0156] ;

[0157] ;

[0158] wherein, denotes the maximum value of the pixel intensity range, which is set to 255, denotes the local window standard deviation, denotes the local window maximum value, denotes the final pooling value;

[0159] The sliding window traversal is specifically sliding the local window along the width and height of the multi-scale convolution feature map by a step size s, and then performing the local difference matrix calculation, the local difference adaptive threshold calculation, the local difference binarization mode aggregation, and the final pooling value calculation operations on each local window in turn to generate a pooling classification feature map with a size of , wherein, denotes the ceiling operation;

[0160] The global average pooling layer is used to retain the overall seepage pattern discrimination information while reducing the feature dimension and suppressing overfitting, and specifically performs a global average pooling operation on the pooling classification feature map to obtain a global seepage type classification feature;

[0161] The full connection classification layer is configured to realize final type identification of the water seepage area; specifically, the global water seepage type classification features are sequentially passed through a full connection layer and an output layer adopting a Softmax activation function, so as to obtain a wall surface water seepage type identification result.

[0162] The full connection layer is configured to perform linear transformation and feature fusion on the global water seepage type classification features; and the output layer is configured to calculate a prediction probability of each water seepage type and output a wall surface water seepage type identification result of a target water seepage area according to a probability maximum principle.

[0163] Step S32: wall surface water seepage type identification model training, specifically, the historical water seepage area feature set is taken as training data to train the identification model, and finally a trained wall surface water seepage type identification model is obtained.

[0164] The identification model training specifically adopts a cross-entropy loss function as an optimization target in the training process, and utilizes a back propagation algorithm and a gradient descent optimization method to iteratively update the weight matrix and bias parameters of the model. In the training process, the identification model parameters are continuously optimized through multiple iterations. When a preset maximum training number is reached or a loss function value converges to a set identification model iteration termination threshold, the iteration training is stopped.

[0165] Step S33: real-time wall surface water seepage type identification, specifically, the real-time water seepage area feature set is input into the trained wall surface water seepage type identification model to obtain a real-time wall surface water seepage type identification result.

[0166] By performing the above operations, the technical problems of the existing wall surface water seepage type identification model, such as the inability to effectively extract local water seepage detail features, the insufficient distinction of multiple water seepage types, and the weak adaptability to complex wall surface scenes, which result in low water seepage type identification accuracy, are solved. The present application innovatively proposes an improved convolutional neural network based on a local difference binary aggregation strategy, and applies it to wall surface water seepage type identification. By introducing a local difference matrix, the local water seepage difference features are strengthened. A local difference adaptive threshold mechanism is proposed and a local difference binary aggregation strategy is designed, which significantly enhances the local feature extraction capability, improves the sensitivity to weak water seepage and small water seepage areas, enhances the model's anti-interference ability and multi-scene adaptability, and thus effectively improves the water seepage type identification accuracy, achieving high-precision, lightweight and intelligent identification of wall surface water seepage types.

[0167] Embodiment five, referring to Figure 1 and Figure 4 This embodiment is based on the above-mentioned embodiments. In step S4, the wall surface water seepage evolution prediction is configured to predict the water seepage development trend of the target wall surface in a future preset time period on the premise that water seepage has occurred on the wall surface; specifically including the following steps:

[0168] Step S41: establishing a wall seepage evolution prediction model, specifically, establishing a wall seepage evolution prediction model based on a long short-term memory network; the wall seepage evolution prediction model comprises a time sequence feature sequence input layer, a long short-term memory network structure layer, and a risk prediction output layer;

[0169] The time sequence feature sequence input layer is configured to input multi-dimensional features in time sequence to the evolution prediction model, and construct a time sequence data structure, specifically, wall seepage area data, wall seepage type identification results, and environmental monitoring data are sorted and assembled into a time sequence feature sequence according to time stamps.

[0170] The long short-term memory network structure layer is specifically configured to set up an LM long short-term memory network unit, update the state of the input time sequence feature sequence, and extract long-term dependence, so as to obtain high-dimensional time sequence risk prediction features capable of representing the development trend of wall seepage;

[0171] The risk prediction output layer is specifically configured to map the high-dimensional time sequence risk prediction features through a fully connected layer, calculate the probability distribution of each risk level through a Softmax function, select the risk level corresponding to the maximum probability as the final output result, and obtain the wall seepage evolution result;

[0172] Step S42: wall seepage evolution prediction model training, specifically, using historical wall seepage evolution data as training data to train the prediction model, and finally obtaining the trained wall seepage evolution prediction model;

[0173] The prediction model training is specifically configured to use a cross-entropy loss function as the optimization target in the training process, and use a back propagation algorithm and a gradient descent optimization method to train and iteratively update the weight matrix and bias parameters of the model. In the training process, the prediction model parameters are continuously optimized through multiple iterations. When the maximum training number or the loss function value converges to the set prediction model iteration termination threshold, the iteration training is stopped.

[0174] Step S43: wall seepage evolution real-time prediction, specifically, inputting wall seepage evolution real-time data into the trained wall seepage evolution prediction model to obtain real-time wall seepage evolution results.

[0175] The wall seepage evolution real-time data comprises real-time wall seepage type identification results, real-time wall seepage area data, and environmental monitoring data in real-time wall seepage evolution data at the same time.

[0176] The real-time wall seepage area data is specifically a geometric time sequence feature set obtained by quantifying the wall seepage state corresponding to each time based on a real-time seepage area feature set.

[0177] Embodiment six,Figure 1 The embodiment is based on the above embodiment, and in step S5, the wall water seepage comprehensive identification is specifically comprehensive identification of the current wall water seepage type and state and the wall water seepage evolution trend of the target wall according to the real-time wall water seepage type identification result and the real-time wall water seepage evolution result, to obtain the overall wall water seepage condition of the target wall, and realize intelligent identification of the target wall water seepage condition.

[0178] It should be noted that, in this document, the terms such as first and second are used only to distinguish one entity or operation from another, and do not necessarily require or imply any actual such relationship or order between these entities or operations. Moreover, the terms "include", "contain" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such a process, method, article or device.

[0179] Although the embodiments of the present application have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present application.

[0180] The above describes the present application and its embodiments, which is not restrictive, and the drawings only show one of the embodiments of the present application, and the actual structure is not limited thereto. In general, if a person skilled in the art is inspired by it, without departing from the purpose of the present application, without creative design, similar structure and embodiments of the technical solution can be designed, which should belong to the protection scope of the present application.

Claims

1. A wall water seepage intelligent identification method based on big data, characterized by: The method comprises the following steps: Step S1: acquiring multi-source data, specifically, obtaining wall seepage management optimization data through data acquisition and data optimization operations; Step S2: wall seepage area segmentation, which is used for automatic identification and segmentation of the seepage area in the wall image, specifically, first extracting the wall image features, then improving the clustering algorithm through noise suppression design, clustering target function, feature weight adaptive adjustment and dynamic membership update mechanism, completing the wall seepage area segmentation model construction, finally, inputting the historical wall image data and real-time wall image data into the wall seepage area segmentation model to obtain the historical seepage area feature set and the real-time seepage area feature set; Step S3: wall seepage type identification, which is used for automatic classification of the seepage type of each wall seepage area, specifically, first introducing an improved convolutional neural network based on a local difference binary aggregation strategy to establish a wall seepage type identification model, training the model using the historical seepage area feature set to generate a trained wall seepage type identification model, and finally inputting the real-time seepage area feature set into the trained identification model to obtain real-time wall seepage type identification results; The improved convolutional neural network comprises a convolution feature extraction block, a local difference binary aggregation-based pooling layer, a global average pooling layer and a full connection classification layer connected in sequence; The local difference binary aggregation-based pooling layer specifically comprises local difference matrix calculation, local difference adaptive threshold calculation, local difference binary pattern aggregation, final pooling value calculation and sliding window traversal; The local difference matrix calculation specifically defines a local window by sliding a multi-scale convolution feature map according to a preset window size, and in each local window, calculates the gray difference between each pixel and the center pixel to obtain a local difference matrix; The local difference adaptive threshold calculation specifically calculates the average absolute deviation of all elements in the local difference matrix as the local difference adaptive threshold; The local difference binary pattern is aggregated, specifically for each element in the local difference matrix, a binary value is calculated by The function calculates a binary approximation value, forms a local difference binary pattern, and then sums all elements of all local difference binary patterns to obtain a local aggregation value. The final pooling value is calculated, specifically, a pooling weighting factor is calculated based on the local aggregation value The final pooling value is obtained by weighted fusion of the local window standard deviation and the local window maximum value through the pooling weighting factor. The sliding window traversal specifically slides the local window along the width and height directions of the multi-scale convolution feature map with a step size s, and then performs the local difference matrix calculation, local difference adaptive threshold calculation, local difference binary pattern aggregation and final pooling value calculation operations on each local window in sequence to generate a pooling classification feature map; Step S4: wall seepage evolution prediction, which is used for predicting the seepage evolution trend of the target wall in a future time period; specifically, a wall seepage evolution prediction model is constructed based on a long short-term memory network, and the model is trained, and finally, the real-time wall seepage evolution data is input into the trained prediction model to obtain real-time wall seepage evolution results; Step S5: wall seepage comprehensive identification, specifically, comprehensively identifying the wall seepage condition according to the real-time wall seepage type identification result and the real-time wall seepage evolution result. 2.The wall water seepage intelligent identification method based on big data according to claim 1, characterized in that: In step S2, the wall seepage area segmentation specifically comprises the following step process: Step S21: wall surface image feature extraction, specifically, the wall surface image in the wall surface water seepage management optimization data is divided into N samples according to the granularity of superpixels, and the features of each sample are extracted, then the features are normalized to obtain a sample vector set; Step S22: constructing a wall surface water seepage area segmentation model; Step S23: wall surface water seepage area segmentation result output, specifically, the historical wall surface image data and the real-time wall surface image data are input into the wall surface water seepage area segmentation model, the water seepage area segmentation processing is performed on all wall surface images, the corresponding historical image segmentation clustering result and real-time image segmentation clustering result are obtained, and in the respective segmentation clustering result, the cluster set whose clustering category belongs to the water seepage area category is selected as the water seepage cluster set of the image, and finally the water seepage area feature extraction is performed from the water seepage cluster set, thereby obtaining the historical water seepage area feature set and the real-time water seepage area feature set. 3.The wall water seepage intelligent identification method based on big data according to claim 2, characterized in that: In step S22, the wall surface water seepage area segmentation model is constructed, specifically including the following step process: Step S221: cluster center initialization, specifically, first set C cluster centers, then randomly select 1 sample from the sample vector set as the first initial cluster center , and sequentially determine the remaining initial cluster centers according to the shortest distance probability selection mechanism, integrate all initial cluster centers to obtain an initial cluster center set; Step S222: noise parameter initialization, specifically, the initial fuzzy membership of each sample to each cluster is calculated by fuzzy C-means algorithm, then the noise coefficient of the feature distribution of each cluster is calculated combined with the initial fuzzy membership, and finally the noise distance of each sample to each cluster is calculated based on the noise coefficient and the initial fuzzy membership; Step S223: design a clustering objective function, specifically, a clustering objective function containing a soft subspace clustering term and a noise suppression term is constructed, and the feature weight is optimized and distributed through relaxation constraint; the formula is as follows: ； In the formula, represents the weighted feature distance of the ith sample to the cth cluster in the tth iteration, J represents the cluster objective function value, and C represents the number of clusters, represents the number of features, N represents the number of samples, and k represents the feature dimension index, represents the weight of the kth feature under the cth cluster in the tth iteration, represents the hybrid membership of the ith sample to the cth cluster in the tth iteration, represents the noise distance of the ith sample under the cth cluster, and m represents the membership index. Step S224: iterative optimization of clustering parameters; Step S225: clustering result output, specifically, according to the final mixed membership matrix, the sample selects the category with the maximum mixed membership as the final clustering category, thereby dividing all samples into multiple independent clusters to obtain the image segmentation clustering result.

4. The wall water seepage intelligent identification method based on big data according to claim 3, characterized in that: In step S224, the iterative optimization of clustering parameters, specifically, by setting the current iteration number t=1, the membership, cluster center and feature weight are updated in a loop to gradually minimize the objective function value; including the following steps: Step S2241: update the mixed membership, specifically, based on the current iteration number, the dynamic membership fusion strategy is adopted to dynamically update the mixed membership of each sample to the cluster combined with the hard membership and the fuzzy membership; the formula is as follows: ； ； ； wherein, denotes the membership fusion weight in the tth iteration, denotes the fuzzy membership of the ith sample to the cth cluster in the tth iteration, denotes the hard membership of the ith sample to the cth cluster in the tth iteration, denotes the weighted feature distance of the ith sample to the lth cluster in the tth iteration, denotes for the ith sample, compute its weighted feature distance to all clusters, find the cluster index with the minimum distance ; Step S2242: update the cluster center, specifically, according to the updated mixed membership, the cluster center of each cluster category is updated according to the weighted average principle; the formula is as follows: ； wherein, denotes the kth feature value of the ith sample, denotes the kth feature value of the cth cluster center sample in the tth iteration; Step S2243: update the feature weight, specifically, based on the current iteration of the cluster center and the mixed membership, the feature weight of each category is adjusted adaptively, so that the features more sensitive to wall surface water identification obtain higher weight; the formula is as follows: ； wherein, represents the dth feature value of the ith sample, represents a relaxation factor, d represents a feature dimension index, represents the dth feature value of the cth cluster center sample in the tth iteration, represents the weight of the kth feature under the cth cluster in the tth iteration. Step S2244: Clustering iteration termination judgment, specifically, re-computing the clustering objective function value in the current iteration , and comparing it with the objective function value in the last iteration , obtaining the objective function change value by calculating the difference between the two . Finally, according to the preset termination condition, the iteration termination judgment is made. If any iteration termination condition is met, the clustering iteration is stopped, and the final clustering center set, mixed membership matrix and feature weight matrix are output.

5. The wall water seepage intelligent identification method based on big data according to claim 1, characterized in that: In step S3, the wall surface water type identification, specifically including the following steps: Step S31: establishing a wall surface water type identification model, specifically, an improved convolutional neural network based on local difference binary aggregation strategy is introduced to construct a wall surface water type identification model; Step S32: wall surface water seepage type recognition model training, specifically, the historical water seepage area feature set is taken as training data to train the recognition model, and finally the trained wall surface water seepage type recognition model is obtained; Step S33: real-time wall surface water seepage type recognition, specifically, the real-time water seepage area feature set is input into the trained wall surface water seepage type recognition model to obtain the real-time wall surface water seepage type recognition result.

6. The wall water seepage intelligent identification method based on big data according to claim 1, characterized in that: In step S4, the wall surface water seepage evolution prediction, specifically including the following steps: Step S41: establishing a wall surface water seepage evolution prediction model, specifically, a wall surface water seepage evolution prediction model is established based on a long short-term memory network; Step S42: wall surface water seepage evolution prediction model training, specifically, the historical wall surface water seepage evolution data is taken as training data to train the prediction model, and finally the trained wall surface water seepage evolution prediction model is obtained; Step S43: real-time wall surface water seepage evolution prediction, specifically, the real-time wall surface water seepage evolution data is input into the trained wall surface water seepage evolution prediction model to obtain the real-time wall surface water seepage evolution result.

7. The wall water seepage intelligent identification method based on big data according to claim 1, characterized in that: In step S5, the wall surface water seepage comprehensive recognition is specifically a comprehensive recognition of the current wall surface water seepage type and state of the target wall surface and the wall surface water seepage evolution trend according to the real-time wall surface water seepage type recognition result and the real-time wall surface water seepage evolution result, to obtain the overall wall surface water seepage condition of the target wall surface, and to realize the intelligent recognition of the target wall surface water seepage condition. 8.The wall water seepage intelligent identification method based on big data according to claim 1, characterized in that: In step S1, the multi-source data is obtained by data collection operation, and the wall surface water seepage management original data is obtained, and the wall surface water seepage management original data is subjected to data optimization operation to obtain wall surface water seepage management optimized data; the wall surface water seepage management original data includes historical wall surface image data, historical wall surface water seepage evolution data, real-time wall surface image data and real-time wall surface water seepage evolution data; the data optimization operation includes image data optimization processing, structure data optimization processing and data synchronization processing.

Images