Business environmental responsibility assessment method, device, equipment, medium and program product

Abstract

The application discloses an enterprise environmental responsibility assessment method, device, equipment, medium and program product, the method comprises: preprocessing a plurality of source heterogeneous data sets of sample enterprises, obtaining a multi-modal feature vector, classifying the multi-modal feature vector into internal multi-modal feature vectors and external multi-modal feature vectors based on data sources, performing cross-modal feature alignment on the internal multi-modal feature vectors and the external multi-modal feature vectors, performing authenticity evaluation based on target internal information and target external information, and obtaining authenticity evaluation results; inputting the authenticity evaluation results into a target enterprise environmental responsibility scoring model for evaluation, and obtaining enterprise environmental responsibility scores of each sample enterprise, so that internal features and external features of the enterprise are effectively mined from the plurality of source heterogeneous data, accurate evaluation of the true situation of the enterprise performing environmental responsibility is realized, behaviors of the enterprise performing environmental responsibility inconsistently are found in time, and environmental supervision intensity is improved.

Description

Technical Field

[0001] This invention relates to the field of data processing, and more particularly to a method, apparatus, equipment, medium, and program product for corporate environmental responsibility assessment. Background Technology

[0002] Against the backdrop of increasing global environmental awareness, the quantitative assessment of corporate environmental responsibility (CIR) has become a crucial indicator for measuring a company's sustainable development capabilities. As a key benchmark for evaluating corporate sustainability, the differentiation and measurement of the degree of variation in CIR is increasingly important, yet existing CIR measurement methods have not yet provided a solution to this problem. Furthermore, traditional CIR assessment methods primarily rely on internal company data or data provided by environmental authorities. This data is often limited, restrictive, and outdated, making it difficult to comprehensively and accurately reflect a company's environmental performance.

[0003] Traditional measures of corporate environmental data cannot effectively determine the accuracy of information from a technical perspective. Furthermore, reliance on limited data sources leads to incomplete assessment results, and the lack of in-depth data mining and analysis in the assessment methods results in inaccurate outcomes. Therefore, current assessments are limited in scope, failing to effectively extract potential features from multi-source, heterogeneous, and multimodal data. This makes it difficult to accurately assess whether companies are truly fulfilling their environmental responsibilities, resulting in vague quantitative assessments of corporate environmental responsibility and low efficiency in regulating corporate environmental responsibility compliance. Summary of the Invention

[0004] The main objective of this invention is to provide a method, apparatus, equipment, medium, and procedure for assessing corporate environmental responsibility, aiming to solve the technical problems in the prior art that make it impossible to accurately assess whether enterprises are truly fulfilling their environmental responsibilities, and that the quantitative assessment of corporate environmental responsibility is vague, resulting in low efficiency in supervising the fulfillment of corporate environmental responsibilities.

[0005] To achieve the above objectives, this invention provides a corporate environmental responsibility assessment method, which is applied to environmental responsibility assessment participating nodes in a blockchain. The method includes the following steps:

[0006] Collect multi-source heterogeneous datasets from multiple sample enterprises, and preprocess the multi-source heterogeneous datasets to obtain multimodal feature vectors, which include image feature vectors, audio feature vectors, text feature vectors, and numerical feature vectors;

[0007] Based on the data sources corresponding to each feature vector in the multimodal feature vector, the data sources are classified to obtain the internal multimodal feature vectors and external multimodal feature vectors of each sample enterprise;

[0008] The internal multimodal feature vector and the external multimodal feature vector are input into a multimodal feature fusion model for cross-modal feature alignment to obtain target internal information and target external information. The multimodal feature fusion model includes a multi-head attention module. The target internal information includes the enterprise's first environmental monitoring information and first environmental protection investment information. The target external information includes the second environmental monitoring information and second environmental protection investment information collected by external monitoring agencies.

[0009] The authenticity of each sample company is assessed based on the target's internal information and external information, and the authenticity assessment results of each sample company are obtained.

[0010] The authenticity assessment results are input into the target enterprise environmental responsibility scoring model for evaluation, and the environmental responsibility scores of each sample enterprise are obtained. The target enterprise environmental responsibility scoring model is constructed based on the local model parameters provided by each environmental responsibility assessment participating node in the blockchain.

[0011] Optionally, the step of conducting an authenticity assessment of each sample company based on the target's internal information and the target's external information, and obtaining the authenticity assessment results for each sample company, includes:

[0012] Based on the first environmental monitoring information and the second environmental monitoring information, calculate the internal and external difference rate of monitoring indicators for each sample enterprise:

[0013]

[0014] Wherein, DR represents the internal and external difference rate of monitoring indicators, Internal Value represents the monitoring indicators within the enterprise, and External Value represents the monitoring indicators of the external monitoring structure.

[0015] Based on the first environmental monitoring information and the second environmental monitoring information, internal and external monitoring images of the target pollution emission monitoring area are obtained, and the similarity of the pollution emission monitoring images is calculated based on the internal and external monitoring images:

[0016]

[0017] Where, μ x The mean value of internal monitoring images, μ y This represents the mean value of external monitoring images. Indicates the variance of internal monitoring images. σ represents the variance of the external monitoring images. xy Let C1 and C2 represent covariance, and C2 be constant terms. SSIM(x,y) represents the similarity of pollution emission monitoring images.

[0018] Based on the first environmental protection investment information and the second environmental protection investment information, calculate the internal and external environmental protection investment difference rate for each sample enterprise:

[0019]

[0020] Among them, DR invest This indicates the difference rate between internal and external environmental protection investment. Internal Investment represents the amount of environmental protection investment within the enterprise, while External Investment represents the amount of environmental protection investment obtained from external monitoring agencies.

[0021] An authenticity assessment model is constructed, and the internal and external difference rates of the monitoring indicators, the similarity of the pollution emission monitoring images, and the internal and external difference rates of environmental protection investment are input into the authenticity assessment model to conduct an authenticity assessment, obtaining the authenticity assessment results for each sample enterprise. The authenticity assessment model includes:

[0022] EIAS=ω1(1-DR)+ω2SSIM(x,y)+ω3(1-DR invest )

[0023] Here, EIAS represents the authenticity assessment result, and ω1, ω2, and ω3 represent the weight parameters, respectively.

[0024] Optionally, before inputting the authenticity assessment results into the target company's environmental responsibility scoring model for assessment, the method further includes:

[0025] Obtain the local model parameters uploaded to the blockchain by each participating node in the environmental responsibility assessment. The local model parameters are obtained by each participating node in the environmental responsibility assessment through gradient training of its local responsibility scoring model.

[0026] Based on the historical behavior information of each environmental responsibility assessment participant node, abnormal nodes with abnormal behavior are removed from the environmental responsibility assessment participant nodes to obtain candidate nodes;

[0027] Obtain the historical contribution information of each candidate node, and determine the contribution weight of each candidate node based on the historical contribution information;

[0028] The local model parameters of each candidate node are aggregated based on the contribution weights to obtain the global model parameters.

[0029] The pre-built local responsibility scoring model is updated based on the global model parameters to obtain the target company's environmental responsibility scoring model.

[0030] Optionally, the multimodal feature fusion model further includes a projection layer;

[0031] The projection layer is used to map the internal multimodal feature vector and the external multimodal feature vector to the same dimensional space, and output candidate internal vectors and candidate external vectors.

[0032] The multi-head attention module is used to perform a linear transformation on the candidate internal vector and the candidate external vector to generate a query matrix, a key matrix and a value matrix.

[0033] The multi-head attention module is further configured to calculate attention scores among modal feature vectors based on the query matrix, the key matrix, and the value matrix, and to perform weighted aggregation based on the attention scores and the value matrix to obtain the multi-modal fusion result of each attention head.

[0034]

[0035] Z i =A i ·V i

[0036] Among them, A i Let Q represent the attention score matrix of the i-th attention head. i K i V i Let d represent the query matrix, key matrix, and value matrix of the i-th attention head, respectively, where T represents the transpose, and d k Z represents the scaling factor. i This represents the multimodal fusion result of the i-th attention head;

[0037] The multi-head attention module is further used to concatenate the features of the multimodal fusion results of each attention head and perform a linear transformation to the target dimension to obtain the multi-head attention output result, which includes target internal information and target external information.

[0038] Z = Concat(Z) 1 ,...,Z h W

[0039] Where Z represents the multi-head attention output, Concat(Z) 1 ,...,Z h ) represents the feature concatenation of the multimodal fusion results of h attention heads, and W represents the linear transformation matrix.

[0040] Optionally, the preprocessing of the multi-source heterogeneous dataset to obtain multimodal feature vectors includes:

[0041] Data cleaning is performed on the multi-source heterogeneous dataset to obtain initial text data, initial numerical data, initial audio data, and initial image data;

[0042] The initial text data is segmented into words, and text feature encoding is performed based on the segmentation results to obtain text feature vectors;

[0043] The initial numerical data is normalized to obtain numerical feature vectors:

[0044]

[0045] Where, x′ i Let x represent the numerical eigenvector obtained after normalizing the i-th initial numerical data. i Let x represent the i-th initial numerical data. max x represents the maximum value in the initial data. min This represents the minimum value in the initial data.

[0046] Mel frequency cepstral coefficients are extracted from the initial audio data, and an audio feature vector is obtained based on the Mel frequency cepstral coefficients;

[0047] Edge detection is performed on the initial image data, and image features are extracted based on the edge detection results to obtain an image feature vector.

[0048] Optionally, the step of performing edge detection on the initial image data and extracting image features based on the edge detection results to obtain an image feature vector includes:

[0049] Edge detection is performed on the initial image data to obtain horizontal and vertical gradient values;

[0050] Calculate the gradient magnitude and gradient direction angle based on the horizontal and vertical gradient values:

[0051]

[0052] Where A represents the initial image data, G represents the gradient magnitude, θ represents the gradient direction angle, and G represents the gradient magnitude. x G represents the gradient value in the horizontal direction. y This represents the gradient value in the vertical direction;

[0053] The initial image data is divided into multiple cells, and a gradient direction histogram for each cell is generated based on the gradient magnitude and the gradient direction angle.

[0054] Based on the gradient histogram, a region of interest analysis for pollution monitoring is performed on each cell, and target pollution emission monitoring cells are selected from the cells based on the results of the pollution monitoring region of interest analysis.

[0055] Adjacent cells in the target pollution emission monitoring cell are stitched together to generate multiple image blocks, and the gradient direction histograms of the target pollution emission monitoring cells in each image block are stitched together to obtain the stitched gradient histogram of the image block.

[0056] The stitched gradient histogram is normalized to obtain the normalized histogram of each image patch:

[0057]

[0058] Among them, H normalized Represents a normalized histogram, h i represents the i-th element in the stitched gradient histogram, H represents the stitched gradient histogram of the image patch, ∈ represents the smoothing term, and n represents the number of elements in the stitched gradient histogram;

[0059] The normalized histograms of each image patch are concatenated to obtain the image feature vector.

[0060] Furthermore, to achieve the above objectives, the present invention also proposes a corporate environmental responsibility assessment device, which is applied to environmental responsibility assessment participating nodes in a blockchain. The corporate environmental responsibility assessment device includes:

[0061] The data acquisition module is used to collect multi-source heterogeneous datasets from multiple sample enterprises and preprocess the multi-source heterogeneous datasets to obtain multimodal feature vectors, which include image feature vectors, audio feature vectors, text feature vectors, and numerical feature vectors.

[0062] The data source classification module is used to classify the data source based on the data source corresponding to each feature vector in the multimodal feature vector, and to obtain the internal multimodal feature vector and external multimodal feature vector of each sample enterprise;

[0063] A cross-modal feature alignment module is used to input the internal multimodal feature vector and the external multimodal feature vector into a multimodal feature fusion model for cross-modal feature alignment to obtain target internal information and target external information. The multimodal feature fusion model includes a multi-head attention module. The target internal information includes the enterprise's first environmental monitoring information and first environmental protection investment information. The target external information includes the second environmental monitoring information and second environmental protection investment information collected by external monitoring agencies.

[0064] The authenticity assessment module is used to assess the authenticity of each sample enterprise based on the target's internal information and the target's external information, and to obtain the authenticity assessment results of each sample enterprise.

[0065] The corporate environmental responsibility assessment module is used to input the authenticity assessment results into the target corporate environmental responsibility scoring model for evaluation, and obtain the corporate environmental responsibility score of each sample company. The target corporate environmental responsibility scoring model is constructed based on the local model parameters provided by each environmental responsibility assessment participating node in the blockchain.

[0066] In addition, to achieve the above objectives, this application also proposes a corporate environmental responsibility assessment device, the device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the corporate environmental responsibility assessment method as described above.

[0067] In addition, to achieve the above objectives, this application also proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the corporate environmental responsibility assessment method described above.

[0068] In addition, to achieve the above objectives, this application also provides a computer program product, which includes a computer program that, when executed by a processor, implements the steps of the corporate environmental responsibility assessment method described above.

[0069] This invention collects multi-source heterogeneous datasets from multiple sample enterprises and preprocesses these datasets to obtain multimodal feature vectors. These multimodal feature vectors include image feature vectors, audio feature vectors, text feature vectors, and numerical feature vectors. Based on the data sources corresponding to each feature vector in the multimodal feature vectors, data source classification is performed to obtain internal and external multimodal feature vectors for each sample enterprise. The internal and external multimodal feature vectors are then input into a multimodal feature fusion model for cross-modal feature alignment to obtain target internal and target external information. The multimodal feature fusion model includes a multi-head attention module. The target internal information includes the enterprise's first environmental monitoring information and first environmental protection investment information, while the target external information includes second environmental monitoring information and second environmental protection investment information collected by external monitoring agencies. Based on the target internal and target external information, the authenticity of each sample enterprise is assessed to obtain the target enterprise's data. The authenticity assessment results of the enterprises are used to input the authenticity assessment results into the target enterprise environmental responsibility scoring model for evaluation, and to obtain the environmental responsibility score of each sample enterprise. The target enterprise environmental responsibility scoring model is constructed based on the local model parameters provided by each environmental responsibility assessment participating node in the blockchain. Because the present invention uses a multimodal feature fusion model to perform cross-modal feature alignment of internal multimodal feature vectors and external multimodal feature vectors, it can accurately mine the internal and external features of sample enterprises, thereby realizing the authenticity assessment of the environmental monitoring information and environmental protection investment information provided by enterprises, accurately detecting whether the pollution emissions of enterprises truly meet the standards and whether the environmental protection investment of enterprises is consistent with the promised investment. The enterprise environmental responsibility scoring model is constructed based on the local model parameters provided by each environmental responsibility assessment participating node in the blockchain to evaluate enterprises, thereby ensuring the authenticity and effectiveness of the assessment, avoiding the problem of the enterprise environmental responsibility assessment results being affected by subjective factors, improving the assessment accuracy, and thus significantly enhancing the environmental supervision. Attached Figure Description

[0070] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0071] Figure 1 This is a schematic diagram of the structure of the enterprise environmental responsibility assessment device for the hardware operating environment involved in the embodiments of the present invention;

[0072] Figure 2 This is a flowchart illustrating an embodiment of the corporate environmental responsibility assessment method of the present invention;

[0073] Figure 3 This is a schematic diagram of the enterprise authenticity assessment process in one embodiment of the present invention;

[0074] Figure 4 This is a schematic diagram illustrating the process of constructing a target enterprise environmental responsibility scoring model in one embodiment of the present invention;

[0075] Figure 5 This is a structural block diagram of an embodiment of the corporate environmental responsibility assessment device of the present invention.

[0076] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0077] It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0078] Reference Figure 1 , Figure 1 This is a schematic diagram of the structure of the enterprise environmental responsibility assessment device for the hardware operating environment involved in the embodiments of the present invention.

[0079] like Figure 1 As shown, the corporate environmental responsibility assessment device may include: a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. The communication bus 1002 is used to enable communication between these components. The user interface 1003 may include a display screen and an input unit such as a keyboard; optionally, the user interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wireless-Fidelity (Wi-Fi) interface). The memory 1005 may be high-speed random access memory (RAM) or stable non-volatile memory (NVM), such as a disk storage device. The memory 1005 may also optionally be a storage device independent of the aforementioned processor 1001.

[0080] Those skilled in the art will understand that Figure 1 The structure shown does not constitute a limitation on the equipment for corporate environmental responsibility assessment and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0081] like Figure 1As shown, the memory 1005, which is a computer-readable storage medium, may include an operating system, a network communication module, a user interface module, and a corporate environmental responsibility assessment program.

[0082] exist Figure 1 In the corporate environmental responsibility assessment device shown, the network interface 1004 is mainly used for data communication with the network server; the user interface 1003 is mainly used for data interaction with the user; the processor 1001 and the memory 1005 in the corporate environmental responsibility assessment device of the present invention can be set in the corporate environmental responsibility assessment device, and the corporate environmental responsibility assessment device calls the corporate environmental responsibility assessment program stored in the memory 1005 through the processor 1001 and executes the corporate environmental responsibility assessment method provided in the embodiment of the present invention.

[0083] This invention provides a method for corporate environmental responsibility assessment, referring to... Figure 2 , Figure 2 This is a flowchart illustrating an embodiment of the corporate environmental responsibility assessment method of the present invention.

[0084] In this embodiment, the corporate environmental responsibility assessment method is applied to environmental responsibility assessment participating nodes in a blockchain, and the corporate environmental responsibility assessment method includes the following steps:

[0085] Step S10: Collect multi-source heterogeneous datasets from multiple sample enterprises and preprocess the multi-source heterogeneous datasets to obtain multimodal feature vectors.

[0086] It should be noted that this embodiment is applied to the quantitative assessment of corporate environmental responsibility. It evaluates the authenticity of corporate pollution emissions and environmental protection investment based on environmental monitoring information and environmental protection investment information, respectively. It constructs a target corporate environmental responsibility scoring model by combining local model parameters provided by other environmental responsibility assessment participating nodes in the blockchain. Based on the target corporate environmental responsibility scoring model, the enterprise is quantitatively scored, thereby achieving accurate quantification of corporate environmental responsibility performance and effectively strengthening corporate environmental supervision.

[0087] It should be understood that the executing entity of this embodiment can be a computing service device with data processing, network communication, and program execution functions, such as a tablet computer, personal computer, or mobile phone, or a terminal electronic device capable of performing the above functions. The following description uses a corporate environmental responsibility assessment device (hereinafter referred to as the assessment device) as an example to illustrate this embodiment and the subsequent embodiments.

[0088] It should be noted that the enterprise's environmental data includes internal data and external data. The internal data includes the enterprise's first environmental monitoring indicator data and first environmental protection investment data, while the external data includes the second environmental monitoring indicator data and second environmental protection investment data collected by external monitoring agencies.

[0089] It should be noted that multi-source heterogeneous data can be data from different sources with different structures and formats. Multi-source heterogeneous data comes from multiple channels, such as internal enterprise systems, publicly available data from environmental protection agencies, third-party monitoring platforms, social media, etc.; heterogeneous means that the data has diverse types and structures, including structured data (such as database tables), semi-structured data (such as XML, JSON), and unstructured data (such as text, images, and videos).

[0090] In some embodiments, the assessment device may select a data source that reflects the company’s environmental responsibility, wherein the data source may include environmental monitoring data, environmental protection investment records, resource consumption statistics, etc. from within the company, as well as environmental monitoring data and monitoring data on the company’s environmental protection investment from external sources.

[0091] The assessment equipment collects various types of data, including but not limited to text, numerical values, voice, and images, based on the aforementioned data sources, to construct a multi-source heterogeneous dataset for corporate environmental responsibility assessment.

[0092] It should be noted that the multimodal feature vectors include image feature vectors, audio feature vectors, text feature vectors, and numerical feature vectors.

[0093] Understandably, preprocessing includes data cleaning, feature extraction, and normalization of multimodal data in multi-source heterogeneous datasets. Evaluation devices can select corresponding processing strategies based on data types, such as word segmentation for text data and object detection and edge detection for image data.

[0094] Furthermore, in order to accurately extract multimodal features from multi-source heterogeneous data, step S10 above may include:

[0095] Step S11: Perform data cleaning on the multi-source heterogeneous dataset to obtain initial text data, initial numerical data, initial audio data, and initial image data;

[0096] Step S12: Segment the initial text data into words, and encode the text features based on the segmentation results to obtain the text feature vector.

[0097] In some embodiments, the evaluation device performs relevant processing on the collected text data, and the main steps are as follows:

[0098] Data cleaning includes: removing irrelevant information from text; correcting spelling errors; deleting stop words; and removing irrelevant information such as HTML tags.

[0099] Preprocessing: Merge synonyms in the text and convert all letters in the text data to lowercase; convert words in different forms into a unified form through operations such as stemming and word correction.

[0100] Standardization: The TF-IDF model is used to perform full-text search and evaluation of the relative importance of word frequency texts, and the text data is then converted into vector representations based on this.

[0101] Step S13: Normalize the initial numerical data to obtain numerical feature vectors;

[0102] In some embodiments, the evaluation device performs relevant processing on the collected numerical data, and the main steps are as follows:

[0103] Data cleaning and preprocessing: Carefully handle missing or outlier values.

[0104] Standardization: Using the "maximum-minimum normalization" method, the data is transformed and reduced to the same dimensionality range to facilitate model training and comparison. Standardization is based on the following formula:

[0105]

[0106] Where, x′ i Let x represent the numerical eigenvector obtained after normalizing the i-th initial numerical data. i Let x represent the i-th initial numerical data. max x represents the maximum value in the initial data. min This represents the minimum value in the initial data.

[0107] Step S14: Extract Mel frequency cepstral coefficients from the initial audio data, and obtain an audio feature vector based on the Mel frequency cepstral coefficients.

[0108] In some embodiments, the evaluation device performs relevant processing on the acquired audio data, and the main steps are as follows:

[0109] Data cleaning: Remove silence and noise, keeping only the speech; remove audio with a very low signal-to-noise ratio to improve data quality.

[0110] Preprocessing: Extract features such as MFCC (Mel frequency cepstral coefficients), LFBE, or PLP from the audio to identify relevant features of the speech data.

[0111] Standardization: The extracted features are standardized, such as by subtracting the mean or dividing by the standard deviation, so that the mean is 0 and the variance is 1 (or other specified mean and variance).

[0112] Furthermore, in order to accurately extract audio features, step S14 above may include:

[0113] Step S141: Pre-emphasize the initial audio data;

[0114] Step S142: Divide the pre-emphasized initial audio data into multiple audio frames, and perform windowing processing on each audio frame based on the window value;

[0115] Step S143: Perform Fast Fourier Transform on each audio frame after windowing, and map the Fast Fourier Transform result to the Mel frequency scale to obtain the Mel frequency cepstral coefficients.

[0116] Step S144: Obtain initial audio features based on the Mel frequency cepstral coefficients, and perform standardization processing on the initial audio features to obtain an audio feature vector.

[0117] It should be noted that the initial audio data is pre-emphasized to enhance high-frequency details and improve the resolution of subsequent frequency domain analysis. The audio data pre-emphasis is performed according to the following formula:

[0118] s′(n)=s(n)-α·s(n-1)

[0119] Where s′(n) represents the nth data point in the initial audio data after pre-emphasis, s(n) represents the nth data point in the initial audio data, and α represents the pre-emphasis coefficient.

[0120] It should be noted that this embodiment can perform frame-by-frame windowing on the audio data to improve the stability of the audio data. The continuous speech signal is divided into short frames (usually 20-40ms), and each frame is windowed to reduce spectral leakage. The window value is calculated based on the following formula:

[0121]

[0122] Where w(n) represents the window value of the i-th audio frame, n represents the frame index, and N represents the frame length. Used to calculate angle values, where 'a' represents a constant offset and 'b' represents a cosine scaling factor, which is used to control the dynamic range of the window shape.

[0123] Step S15: Perform edge detection on the initial image data, and extract image features based on the edge detection results to obtain the image feature vector.

[0124] In some embodiments, the evaluation device performs relevant processing on the acquired image data, and the main steps are as follows:

[0125] Image preprocessing: Image scaling and cropping are performed using image processing libraries (such as OpenCV); noise reduction algorithms such as median filtering and Gaussian filtering are used.

[0126] Image analysis: Edge detection is performed using Canny, Sobel, etc.; image feature extraction is performed using HOG, SIFT, SURF, etc.

[0127] Standardization: Converting images to a uniform format (such as JPEG, PNG); adjusting images to a uniform resolution.

[0128] Furthermore, in order to accurately extract image features, step S15 above may include:

[0129] Step S151: Perform edge detection on the initial image data to obtain horizontal and vertical gradient values;

[0130] Step S152: Calculate the gradient magnitude and gradient direction angle based on the horizontal gradient value and the vertical gradient value;

[0131] Step S153: Divide the initial image data into multiple cells, and generate a gradient direction histogram for each cell based on the gradient magnitude and the gradient direction angle;

[0132] Step S154: Perform pollution monitoring region of interest analysis on each cell based on the gradient histogram, and select target pollution emission monitoring cells from the cells based on the pollution monitoring region of interest analysis results;

[0133] Step S155: Stitch together adjacent cells in the target pollution emission monitoring cell to generate multiple image blocks, and stitch together the gradient direction histograms of the target pollution emission monitoring cells in each image block to obtain the stitched gradient histogram of the image block.

[0134] Step S156: Normalize the stitched gradient histogram to obtain the normalized histogram of each image block;

[0135] Step S157: Concatenate the normalized histograms of each image patch to obtain the image feature vector.

[0136] It should be noted that the evaluation device can calculate from both the horizontal and vertical directions using convolution kernels to detect vertical and horizontal edges, thereby accurately describing the edge direction of the image and distinguishing between strong and weak edges. The formulas for calculating the gradient magnitude and gradient direction angle are as follows:

[0137]

[0138] Where A represents the initial image data, G represents the gradient magnitude, θ represents the gradient direction angle, and G represents the gradient magnitude. x G represents the gradient value in the horizontal direction. y This represents the gradient value in the vertical direction.

[0139] It should be noted that, in order to eliminate amplitude deviations caused by differences in illumination or contrast in different areas and to preserve directional distribution characteristics, the evaluation device can normalize the stitched gradient histogram. The histogram normalization is performed using the following formula:

[0140]

[0141] Among them, H normalized Represents a normalized histogram, h i Let represent the i-th element in the stitched gradient histogram, H represent the stitched gradient histogram of the image patch, ∈ represent the smoothing term, and n represent the number of elements in the stitched gradient histogram.

[0142] Understandably, in order to focus on image areas that may emit pollutants, this embodiment performs pollution monitoring region of interest analysis on each cell and selects target pollution emission monitoring cells from the cells based on the results of the pollution monitoring region of interest analysis, thereby identifying the target pollution emission area in the image and performing targeted feature analysis on the pollution emission area.

[0143] Step S20: Based on the data source corresponding to each feature vector in the multimodal feature vector, classify the data source to obtain the internal multimodal feature vector and external multimodal feature vector of each sample enterprise.

[0144] It should be noted that, in order to accurately assess the authenticity of corporate environmental responsibility performance, this embodiment can classify the data sources of multimodal data, classifying the data into internal multimodal feature vectors and external multimodal feature vectors, thereby combining external factors to assess the authenticity of the internal data provided by the company.

[0145] It should be noted that internal data sources can be data sources provided within the enterprise, while external data sources can be data sources provided by external monitoring agencies (such as third-party environmental monitoring agencies or official monitoring agencies).

[0146] Step S30: Input the internal multimodal feature vector and the external multimodal feature vector into the multimodal feature fusion model to perform cross-modal feature alignment, and obtain the target internal information and the target external information.

[0147] It should be noted that the multimodal feature fusion model includes a multi-head attention module, the target internal information includes the enterprise's first environmental monitoring information and first environmental protection investment information, and the target external information includes the second environmental monitoring information and second environmental protection investment information collected by external monitoring agencies.

[0148] It should be noted that environmental monitoring information can be monitoring information on enterprise pollutant emissions, which may include pollutant emission monitoring indicators and pollutant emission monitoring images. Environmental protection investment information can be information on the resources invested by enterprises in controlling pollutant emissions, such as the amount of money invested, manpower invested, and materials invested.

[0149] In some embodiments, the multimodal feature fusion model further includes a projection layer;

[0150] The projection layer is used to map the internal multimodal feature vector and the external multimodal feature vector to the same dimensional space, and output candidate internal vectors and candidate external vectors.

[0151] The multi-head attention module is used to perform a linear transformation on the candidate internal vector and the candidate external vector to generate a query matrix, a key matrix and a value matrix.

[0152] The multi-head attention module is further configured to calculate attention scores among modal feature vectors based on the query matrix, the key matrix, and the value matrix, and to perform weighted aggregation based on the attention scores and the value matrix to obtain the multi-modal fusion result of each attention head.

[0153]

[0154] Z i =A i ·V i

[0155] Among them, A i Let Q represent the attention score matrix of the i-th attention head. i K i V i Let d represent the query matrix, key matrix, and value matrix of the i-th attention head, respectively, where T represents the transpose, and d k Z represents the scaling factor. i This represents the multimodal fusion result of the i-th attention head;

[0156] The multi-head attention module is further used to concatenate the features of the multimodal fusion results of each attention head and perform a linear transformation to the target dimension to obtain the multi-head attention output result, which includes target internal information and target external information.

[0157] Z = Concat(Z) 1 ,...,Z h W

[0158] Where Z represents the multi-head attention output, Concat(Z) 1 ,...,Z h ) represents the feature concatenation of the multimodal fusion results of h attention heads, and W represents the linear transformation matrix.

[0159] Step S40: Based on the target's internal information and the target's external information, conduct an authenticity assessment of each sample enterprise and obtain the authenticity assessment results for each sample enterprise.

[0160] It should be noted that the authenticity assessment result can refer to the degree of consistency between the corporate environmental responsibility disclosure information and the actual activities. If the company's "sayings" and "actions" regarding environmental responsibility are consistent, then the company's environmental data is deemed to be true and reliable.

[0161] In some embodiments, the authenticity assessment model can be a neural network model built on a semi-supervised generative adversarial network; the assessment device sets up a semi-supervised generative adversarial network to conduct comparative assessments based on multi-source heterogeneous data, and distinguishes the real dataset of environmental data by comparing different types of data of the same environmental information, so as to carry out authenticity judgment.

[0162] In some embodiments, the assessment device compares the pollution monitoring indicators provided by the enterprise with the pollution monitoring indicators collected by external monitoring agencies based on the first environmental monitoring information and the second environmental monitoring information, thereby assessing the authenticity of the enterprise's pollutant emissions; and compares the pollutant treatment investment resources provided by the enterprise with the actual pollutant treatment investment resources collected by external monitoring agencies based on the first environmental protection investment information and the second environmental protection investment information, thereby assessing the authenticity of the enterprise's environmental protection investment.

[0163] In some embodiments, to improve the accuracy of authenticity assessment, the assessment device can construct a generator to generate a sample dataset and a real dataset based on multimodal feature vectors. A discriminator is then constructed, and the sample dataset and the real dataset are input into the discriminator for data classification and authenticity determination. Classification loss information and authenticity determination loss information are obtained. Based on the classification loss information and the authenticity determination loss information, the total loss information of the discriminator is obtained. A pre-constructed semi-supervised generative adversarial network is trained based on the total loss information to obtain an authenticity assessment model. By constructing a semi-supervised generative adversarial network, the authenticity determination of the aforementioned corporate environmental responsibility data can be carried out, that is, by analyzing data to determine whether the performance data of the same corporate environmental responsibility is consistent across different types of data presentation.

[0164] It is understandable that the evaluation equipment can build a realistic evaluation model based on the following steps:

[0165] Step 1: Build a generator to produce realistic data samples (text, images, numbers, etc.).

[0166] The second step is to build a discriminator to distinguish between real and generated data, and to classify real data (distinguishing between "consistent" and "inconsistent"), and to conduct supervised learning using a small amount of labeled data.

[0167] Step 3: By separating the generator loss and the discriminator loss, design the loss function and perform a weighted summation of the total loss. The calculation formula is as follows:

[0168] L total =αL 真实性判别 +βL 分类

[0169] Among them, L total L represents the total loss after weighted summation. 真实性判别 L represents the authenticity of loss information. 分类 This represents the classification loss information, where α and β represent the weight coefficients, respectively.

[0170] Step 4: Input multi-source heterogeneous data into a semi-supervised generative adversarial network, and use semi-supervised learning to train the generator and discriminator respectively.

[0171] Step 5: Use the test set to evaluate the classification performance of the model (accuracy, recall, F1 score, etc.), and use cross-validation to select the best model.

[0172] Step 6: Conduct authenticity assessment, which involves inputting heterogeneous data from multiple sources from the same company and using a trained discriminator to classify the data to determine whether the company's environmental responsibility performance is "consistent with its words" and to distinguish between real and non-real datasets.

[0173] In some embodiments, in order to determine whether a company’s environmental responsibility performance is “consistent with its words and deeds,” this embodiment can classify multimodal feature vectors into real datasets and non-real datasets, determine the proportion of internal and external data of the company in the real dataset, and the proportion of internal and external data of the company in the non-real dataset; if the proportion of non-real data within the company is large, it is determined that the company has inconsistent words and deeds; if the proportion of real data within the company is large, it is determined that the company’s publicity and actions in fulfilling its environmental responsibility are consistent.

[0174] Furthermore, in order to accurately assess the authenticity of corporate environmental responsibility fulfillment, refer to Figure 3 , Figure 3As a schematic diagram of the enterprise authenticity assessment process in one embodiment, step S40 above may include:

[0175] Step S401: Calculate the internal and external difference rate of monitoring indicators for each sample enterprise based on the first environmental monitoring information and the second environmental monitoring information;

[0176] Step S402: Obtain internal and external monitoring images of the target pollution emission monitoring area based on the first environmental monitoring information and the second environmental monitoring information, and calculate the similarity of the pollution emission monitoring images based on the internal and external monitoring images;

[0177] Step S403: Calculate the internal and external environmental investment difference rate of each sample enterprise based on the first environmental investment information and the second environmental investment information;

[0178] Step S404: Construct an authenticity assessment model, and input the internal and external difference rate of the monitoring indicators, the similarity of the pollution emission monitoring images, and the internal and external difference rate of environmental protection investment into the authenticity assessment model to conduct an authenticity assessment and obtain the authenticity assessment results of each sample enterprise.

[0179] It should be noted that, in order to assess the true extent of a company's pollutant emissions, the internal and external discrepancy rate of monitoring indicators can be calculated based on the following formula, thereby enabling a discrepancy analysis between the company's advertised pollution emissions and its actual pollution emissions:

[0180]

[0181] Wherein, DR represents the internal and external difference rate of monitoring indicators, Internal Value represents the monitoring indicators within the enterprise, and External Value represents the monitoring indicators of the external monitoring structure.

[0182] It should be noted that, in order to analyze the differences between the pollutant emission images advertised by enterprises and the actual emission images, this implementation can calculate the similarity of pollution emission monitoring images based on the following formula, thereby assessing whether the emission images advertised by enterprises are consistent with the actual emission images:

[0183]

[0184] Where, μ x The mean value of internal monitoring images, μ y This represents the mean value of external monitoring images. Indicates the variance of internal monitoring images. σ represents the variance of the external monitoring images. xy Let C1 and C2 represent the covariance, and C2 be constant terms. SSIM(x,y) represents the similarity of pollution emission monitoring images.

[0185] It should be noted that, in order to check whether the amount of money companies advertise for pollution control and environmental protection investment is consistent with their actual investment, the internal and external difference rate of environmental protection investment can be calculated based on the following formula:

[0186]

[0187] Among them, DR invest This indicates the difference rate between internal and external environmental protection investment. Internal Investment represents the amount of environmental protection investment within the enterprise, while External Investment represents the amount of environmental protection investment obtained from external monitoring agencies.

[0188] It should be noted that, in order to assess the authenticity of corporate environmental responsibility fulfillment from multiple dimensions, the authenticity assessment model includes:

[0189] EIAS=ω1(1-DR)+ω2SSIM(x,y)+ω3(1-DR invest )

[0190] Here, EIAS represents the authenticity assessment result, and ω1, ω2, and ω3 represent the weight parameters, respectively.

[0191] Step S50: Input the authenticity assessment results into the target enterprise environmental responsibility scoring model for evaluation, and obtain the enterprise environmental responsibility score of each sample enterprise.

[0192] It should be noted that the target enterprise environmental responsibility scoring model is constructed based on the local model parameters provided by each environmental responsibility assessment participating node in the blockchain. The blockchain can be a federated learning-based enterprise responsibility assessment blockchain, comprising multiple environmental responsibility assessment participating nodes.

[0193] Understandably, in order to ensure the accuracy and credibility of corporate environmental responsibility assessment results, this embodiment aggregates the local model parameters of each participating node in the environmental responsibility assessment in the blockchain, and updates the local scoring model based on the aggregated global model parameters, thereby improving the generalization ability and accuracy of the scoring model.

[0194] Furthermore, in order to accurately quantify and assess a company's environmental responsibility, refer to Figure 4 , Figure 4 This is a schematic diagram of the process for constructing a target enterprise's environmental responsibility scoring model in one embodiment. Step S50 above may include:

[0195] Step S501: Obtain the local model parameters uploaded to the blockchain by each participating node in the environmental responsibility assessment;

[0196] Step S502: Based on the historical behavior information of each environmental responsibility assessment participating node, remove abnormal nodes with abnormal behavior from the environmental responsibility assessment participating nodes to obtain candidate nodes;

[0197] Step S503: Obtain the historical contribution information of each candidate node, and determine the contribution weight of each candidate node based on the historical contribution information;

[0198] Step S504: Aggregate the local model parameters of each candidate node based on the contribution weight to obtain the global model parameters;

[0199] Step S505: Update the pre-built local responsibility scoring model according to the global model parameters to obtain the target enterprise's environmental responsibility scoring model.

[0200] It should be noted that the local model parameters are obtained by each participating node in the environmental responsibility assessment through gradient training of the local responsibility scoring model.

[0201] In some embodiments, let the set of nodes participating in the environmental responsibility assessment be N = [N1, N2, ..., Nm], and define the historical behavior information matrix H ∈ R. m×k Each row H i The k-dimensional behavioral characteristics of node Ni (such as data upload frequency, model training accuracy fluctuation, data compliance score, etc.) are represented. The evaluation device can set an abnormal behavior threshold, and the mean μ is calculated for each feature dimension j. j and standard value σ j Set the threshold for abnormal behavior to μ j ±ασ j (α is a hyperparameter) If the eigenvector H of node Ni i If any dimension value exceeds the abnormal behavior threshold, then Ni is determined to be an abnormal node, denoted as A, and the final set of candidate nodes is C = NA.

[0202] In some embodiments, the evaluation device can calculate the entropy value of each candidate node by combining the historical contribution behavior of each candidate node with the current network communication performance of each candidate node using the entropy weight method, and calculate the contribution weight based on the entropy value.

[0203] In some embodiments, the evaluation device may aggregate the local model parameters of each candidate node based on the following formula:

[0204]

[0205] Where, θ global θ represents the global model parameters. i ω represents the local model parameters of candidate node i. i This represents the contribution weight of candidate node i.

[0206] In some embodiments, the local responsibility scoring model can be a support vector machine (SVM) model. The assessment device can use the SVM model to process real datasets to measure the environmental responsibility level of sample enterprises and the degree of difference between them and other related enterprises. The specific steps are as follows:

[0207] Step 1: Organize the dataset into structured data, and divide it into 80% training set and 20% test set.

[0208] Step 2: Use the regression SVM (SVR) model to predict the environmental responsibility score of the sample companies, and select radial basis kernel function parameters to carry out the analysis.

[0209] Step 3: Input the training set data into the SVM model to train and confirm the optimal parameter combination of the model.

[0210] Step 4: Input the test set data into the trained SVM model and confirm the model's test results.

[0211] Step 5: Utilize the support vectors of the SVM model to calculate the distance between sample firms in the feature space. A larger distance indicates a greater difference in environmental responsibility performance among firms.

[0212] Step 6: Based on the output of the SVM model, select the K-Means clustering method to perform cluster analysis on the sample companies, and divide the companies into different environmental responsibility levels or categories.

[0213] Step 7: Identify the features that have the greatest impact on corporate environmental responsibility differentiation through SVM model weight coefficient or feature contribution analysis.

[0214] In some embodiments, the assessment device may determine the degree of responsibility differentiation among sample companies based on the corporate environmental responsibility score.

[0215] It should be noted that the degree of responsibility differentiation can refer to the differences in the performance of different companies in fulfilling their environmental responsibilities. The assessment of the degree of differentiation is usually based on multi-dimensional data, combined with factors such as industry characteristics, company size, and regional differences, to form a comprehensive evaluation result.

[0216] In some embodiments, the assessment device can conduct a comprehensive assessment of the corporate environmental responsibility of each sample enterprise based on the corporate environmental responsibility score and the degree of responsibility differentiation, obtain the comprehensive assessment results of corporate environmental responsibility, compile the corporate environmental responsibility score and the degree of responsibility differentiation into an assessment report, and use tools such as radar charts and scatter plots for visualization.

[0217] In some embodiments, in order to accurately determine whether an enterprise is truly fulfilling its environmental responsibilities, the assessment device can determine the proportion of genuine internal data and the proportion of non-genuine internal data for each sample enterprise based on the data sources of each feature data in the genuine dataset and the non-genuine dataset. The data sources include external data sources and internal data sources. Based on the proportion of genuine internal data and the proportion of non-genuine internal data, the device can conduct a consistency assessment of each sample enterprise to obtain a consistency assessment result. Based on the consistency assessment result, the enterprise's environmental responsibility score, and the degree of responsibility differentiation, the device can conduct a comprehensive assessment of the enterprise's environmental responsibility to obtain a comprehensive assessment result.

[0218] It should be noted that, in order to determine whether a company's environmental responsibility performance is "consistent with its words and actions," this embodiment can classify multimodal feature vectors into real datasets and non-real datasets, determine the proportion of internal and external data of the company in the real dataset, and the proportion of internal and external data of the company in the non-real dataset; if the proportion of non-real data within the company is large, it is determined that the company has inconsistencies between its words and actions; if the proportion of real data within the company is large, it is determined that the company's publicity and actions in fulfilling its environmental responsibility are consistent.

[0219] It is understandable that this embodiment not only assesses the actual environmental responsibility performance of enterprises and the degree of difference between enterprises, but also evaluates the authenticity of enterprises' environmental responsibility performance and the consistency between their words and deeds, thereby accurately assessing the credibility of enterprises' environmental responsibility performance and achieving a comprehensive assessment of enterprises' environmental responsibility from multiple dimensions and in all aspects.

[0220] This embodiment collects multi-source heterogeneous datasets from multiple sample enterprises and preprocesses these datasets to obtain multimodal feature vectors. These multimodal feature vectors include image feature vectors, audio feature vectors, text feature vectors, and numerical feature vectors. Based on the data sources corresponding to each feature vector in the multimodal feature vectors, data source classification is performed to obtain internal and external multimodal feature vectors for each sample enterprise. The internal and external multimodal feature vectors are then input into a multimodal feature fusion model for cross-modal feature alignment to obtain target internal and target external information. The multimodal feature fusion model includes a multi-head attention module. The target internal information includes the enterprise's first environmental monitoring information and first environmental protection investment information, while the target external information includes second environmental monitoring information and second environmental protection investment information collected by external monitoring agencies. Based on the target internal and target external information, the authenticity of each sample enterprise is assessed to obtain the results for each sample enterprise. The authenticity assessment results of enterprises are used to input the results into the target enterprise environmental responsibility scoring model for evaluation, thereby obtaining the environmental responsibility score of each sample enterprise. The target enterprise environmental responsibility scoring model is constructed based on the local model parameters provided by each environmental responsibility assessment participating node in the blockchain. Because this invention uses a multimodal feature fusion model to perform cross-modal feature alignment of internal and external multimodal feature vectors, it accurately mines the internal and external features of sample enterprises, thereby realizing the authenticity assessment of the environmental monitoring information and environmental protection investment information provided by enterprises. It accurately detects whether the pollution emissions of enterprises truly meet the standards and whether the environmental protection investment of enterprises is consistent with the promised investment. The enterprise environmental responsibility scoring model is constructed based on the local model parameters provided by each environmental responsibility assessment participating node in the blockchain to evaluate enterprises, thereby ensuring the authenticity and effectiveness of the evaluation, avoiding the problem of the enterprise environmental responsibility evaluation results being affected by subjective factors, improving the evaluation accuracy, and thus significantly enhancing the environmental supervision.

[0221] Furthermore, embodiments of the present invention also propose a computer-readable storage medium storing a corporate environmental responsibility assessment program, wherein the corporate environmental responsibility assessment program, when executed by a processor, implements the steps of the corporate environmental responsibility assessment method described above.

[0222] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0223] The aforementioned computer-readable storage medium may be included in the corporate environmental responsibility assessment equipment; or it may exist independently and not be assembled into the corporate environmental responsibility assessment equipment.

[0224] Furthermore, embodiments of the present invention also propose a computer program product, including a corporate environmental responsibility assessment program, which, when executed by a processor, implements the steps of the corporate environmental responsibility assessment method described above.

[0225] The specific implementation of the computer program product of the present invention is basically the same as the embodiments of the above-mentioned corporate environmental responsibility assessment method, and will not be repeated here.

[0226] Reference Figure 5 , Figure 5 This is a structural block diagram of an embodiment of the corporate environmental responsibility assessment device of the present invention.

[0227] like Figure 5 As shown, in this embodiment, the device is applied to an environmental responsibility assessment participating node in a blockchain. The enterprise environmental responsibility assessment device proposed in this embodiment includes:

[0228] The data acquisition module 10 is used to collect multi-source heterogeneous datasets from multiple sample enterprises and preprocess the multi-source heterogeneous datasets to obtain multimodal feature vectors, which include image feature vectors, audio feature vectors, text feature vectors and numerical feature vectors.

[0229] The data source classification module 20 is used to classify the data source based on the data source corresponding to each feature vector in the multimodal feature vector, and obtain the internal multimodal feature vector and external multimodal feature vector of each sample enterprise;

[0230] The cross-modal feature alignment module 30 is used to input the internal multimodal feature vector and the external multimodal feature vector into the multimodal feature fusion model for cross-modal feature alignment to obtain target internal information and target external information. The multimodal feature fusion model includes a multi-head attention module. The target internal information includes the enterprise's internal first environmental monitoring information and first environmental protection investment information. The target external information includes the second environmental monitoring information and second environmental protection investment information collected by external monitoring agencies.

[0231] The authenticity assessment module 40 is used to conduct authenticity assessments on each sample enterprise based on the target's internal information and the target's external information, and to obtain the authenticity assessment results for each sample enterprise.

[0232] The corporate environmental responsibility assessment module 50 is used to input the authenticity assessment results into the target corporate environmental responsibility scoring model for evaluation, and obtain the corporate environmental responsibility score of each sample company. The target corporate environmental responsibility scoring model is constructed based on the local model parameters provided by each environmental responsibility assessment participating node in the blockchain.

[0233] This embodiment collects multi-source heterogeneous datasets from multiple sample enterprises and preprocesses these datasets to obtain multimodal feature vectors. These multimodal feature vectors include image feature vectors, audio feature vectors, text feature vectors, and numerical feature vectors. Based on the data sources corresponding to each feature vector in the multimodal feature vectors, data source classification is performed to obtain internal and external multimodal feature vectors for each sample enterprise. The internal and external multimodal feature vectors are then input into a multimodal feature fusion model for cross-modal feature alignment to obtain target internal and target external information. The multimodal feature fusion model includes a multi-head attention module. The target internal information includes the enterprise's first environmental monitoring information and first environmental protection investment information, while the target external information includes second environmental monitoring information and second environmental protection investment information collected by external monitoring agencies. Based on the target internal and target external information, the authenticity of each sample enterprise is assessed to obtain the results for each sample enterprise. The authenticity assessment results of enterprises are used to input the results into the target enterprise environmental responsibility scoring model for evaluation, thereby obtaining the environmental responsibility score of each sample enterprise. The target enterprise environmental responsibility scoring model is constructed based on the local model parameters provided by each environmental responsibility assessment participating node in the blockchain. Because this invention uses a multimodal feature fusion model to perform cross-modal feature alignment of internal and external multimodal feature vectors, it accurately mines the internal and external features of sample enterprises, thereby realizing the authenticity assessment of the environmental monitoring information and environmental protection investment information provided by enterprises. It accurately detects whether the pollution emissions of enterprises truly meet the standards and whether the environmental protection investment of enterprises is consistent with the promised investment. The enterprise environmental responsibility scoring model is constructed based on the local model parameters provided by each environmental responsibility assessment participating node in the blockchain to evaluate enterprises, thereby ensuring the authenticity and effectiveness of the evaluation, avoiding the problem of the enterprise environmental responsibility evaluation results being affected by subjective factors, improving the evaluation accuracy, and thus significantly enhancing the environmental supervision.

[0234] The corporate environmental responsibility assessment device provided in this application, employing the corporate environmental responsibility assessment method described in the above embodiments, can solve the technical problems of corporate environmental responsibility assessment. Compared with the prior art, the beneficial effects of the corporate environmental responsibility assessment device provided in this application are the same as those of the corporate environmental responsibility assessment method provided in the above embodiments, and other technical features in the corporate environmental responsibility assessment device are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0235] It should be understood that the above are merely illustrative examples and do not constitute any limitation on the technical solutions of the present invention. In specific applications, those skilled in the art can make settings as needed, and the present invention does not impose any restrictions on this.

[0236] It should be noted that the workflow described above is merely illustrative and does not limit the scope of protection of this invention. In practical applications, those skilled in the art can select some or all of the workflow to achieve the purpose of this embodiment according to actual needs, and no restrictions are imposed here.

[0237] In addition, for technical details not described in detail in this embodiment, please refer to the corporate environmental responsibility assessment method provided in any embodiment of the present invention, which will not be repeated here.

[0238] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0239] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0240] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as read-only memory / random access memory, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0241] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A method for assessing corporate environmental responsibility, characterized in that, The corporate environmental responsibility assessment method is applied to environmental responsibility assessment participating nodes in a blockchain, and the corporate environmental responsibility assessment method includes: Collect multi-source heterogeneous datasets from multiple sample enterprises, and preprocess the multi-source heterogeneous datasets to obtain multimodal feature vectors, which include image feature vectors, audio feature vectors, text feature vectors, and numerical feature vectors; Based on the data sources corresponding to each feature vector in the multimodal feature vector, the data sources are classified to obtain the internal multimodal feature vectors and external multimodal feature vectors of each sample enterprise; The internal multimodal feature vector and the external multimodal feature vector are input into a multimodal feature fusion model for cross-modal feature alignment to obtain target internal information and target external information. The multimodal feature fusion model includes a multi-head attention module. The target internal information includes the enterprise's first environmental monitoring information and first environmental protection investment information. The target external information includes the second environmental monitoring information and second environmental protection investment information collected by external monitoring agencies. The process of conducting authenticity assessments on each sample company based on the target's internal and external information, and obtaining the authenticity assessment results for each sample company, includes: Based on the first environmental monitoring information and the second environmental monitoring information, calculate the internal and external difference rate of monitoring indicators for each sample enterprise: in, This indicates the rate of difference between internal and external monitoring indicators. These represent internal monitoring indicators for the company. Monitoring indicators representing the external monitoring structure; Based on the first environmental monitoring information and the second environmental monitoring information, internal and external monitoring images of the target pollution emission monitoring area are obtained, and the similarity of the pollution emission monitoring images is calculated based on the internal and external monitoring images: in, This represents the mean value of internal monitoring images. This represents the mean value of external monitoring images. Indicates the variance of internal monitoring images. Indicates the variance of external monitoring images. Describing covariance, and All are constant terms. Indicates the similarity of pollution emission monitoring images; Based on the first environmental protection investment information and the second environmental protection investment information, calculate the internal and external environmental protection investment difference rate for each sample enterprise: in, This indicates the difference rate between domestic and international environmental protection investment. This indicates the amount of environmental protection investment within the company. This indicates the amount of environmental protection investment obtained by external monitoring agencies; An authenticity assessment model is constructed, and the internal and external difference rates of the monitoring indicators, the similarity of the pollution emission monitoring images, and the internal and external difference rates of environmental protection investment are input into the authenticity assessment model to conduct an authenticity assessment, obtaining the authenticity assessment results for each sample enterprise. The authenticity assessment model includes: in, This indicates the results of the authenticity assessment. These represent the weight parameters; Obtain the local model parameters uploaded to the blockchain by each participating node in the environmental responsibility assessment. The local model parameters are obtained by each participating node in the environmental responsibility assessment through gradient training of its local responsibility scoring model. Based on the historical behavior information of each environmental responsibility assessment participant node, abnormal nodes with abnormal behavior are removed from the environmental responsibility assessment participant nodes to obtain candidate nodes; Obtain the historical contribution information of each candidate node, and determine the contribution weight of each candidate node based on the historical contribution information; The local model parameters of each candidate node are aggregated based on the contribution weights to obtain the global model parameters. The pre-built local responsibility scoring model is updated based on the global model parameters to obtain the target company's environmental responsibility scoring model; The authenticity assessment results are input into the target enterprise environmental responsibility scoring model for evaluation, and the environmental responsibility scores of each sample enterprise are obtained. The target enterprise environmental responsibility scoring model is constructed based on the local model parameters provided by each environmental responsibility assessment participating node in the blockchain.

2. The corporate environmental responsibility assessment method as described in claim 1, characterized in that, The multimodal feature fusion model also includes a projection layer; The projection layer is used to map the internal multimodal feature vector and the external multimodal feature vector to the same dimensional space, and output candidate internal vectors and candidate external vectors. The multi-head attention module is used to perform a linear transformation on the candidate internal vector and the candidate external vector to generate a query matrix, a key matrix and a value matrix. The multi-head attention module is further configured to calculate attention scores among modal feature vectors based on the query matrix, the key matrix, and the value matrix, and to perform weighted aggregation based on the attention scores and the value matrix to obtain the multi-modal fusion result of each attention head. in, Indicates the first Attention score matrix of each attention head. They represent the first The query matrix, key matrix, and value matrix of each attention head. Indicates transpose. Indicates the scaling factor. Indicates the first The multimodal fusion result of each attention head; The multi-head attention module is further used to concatenate the features of the multimodal fusion results of each attention head and perform a linear transformation to the target dimension to obtain the multi-head attention output result, which includes target internal information and target external information. in, This represents the output result of multi-head attention. Indicates will Feature concatenation is performed on the multimodal fusion results of each attention head. This represents a linear transformation matrix.

3. The corporate environmental responsibility assessment method as described in claim 1, characterized in that, The preprocessing of the multi-source heterogeneous dataset to obtain multimodal feature vectors includes: Data cleaning is performed on the multi-source heterogeneous dataset to obtain initial text data, initial numerical data, initial audio data, and initial image data; The initial text data is segmented into words, and text feature encoding is performed based on the segmentation results to obtain text feature vectors; The initial numerical data is normalized to obtain numerical feature vectors: in, Indicates the first The numerical feature vector obtained after normalizing the initial numerical data. Indicates the first Initial numerical data, This represents the maximum value in the initial data. This represents the minimum value in the initial data. Mel frequency cepstral coefficients are extracted from the initial audio data, and an audio feature vector is obtained based on the Mel frequency cepstral coefficients; Edge detection is performed on the initial image data, and image features are extracted based on the edge detection results to obtain an image feature vector.

4. The corporate environmental responsibility assessment method as described in claim 3, characterized in that, The step of performing edge detection on the initial image data and extracting image features based on the edge detection results to obtain an image feature vector includes: Edge detection is performed on the initial image data to obtain horizontal and vertical gradient values; Calculate the gradient magnitude and gradient direction angle based on the horizontal and vertical gradient values: in, This represents the initial image data. Indicates the gradient magnitude. Indicates the gradient direction angle. This represents the gradient value in the horizontal direction. This represents the gradient value in the vertical direction; The initial image data is divided into multiple cells, and a gradient direction histogram for each cell is generated based on the gradient magnitude and the gradient direction angle. Based on the gradient histogram, a region of interest analysis for pollution monitoring is performed on each cell, and target pollution emission monitoring cells are selected from the cells based on the results of the pollution monitoring region of interest analysis. Adjacent cells in the target pollution emission monitoring cell are stitched together to generate multiple image blocks, and the gradient direction histograms of the target pollution emission monitoring cells in each image block are stitched together to obtain the stitched gradient histogram of the image block. The stitched gradient histogram is normalized to obtain the normalized histogram of each image patch: in, Represents a normalized histogram. Indicates the th in the concatenated gradient histogram One element, The histogram representing the gradient of the stitched image patches. Indicates the smoothing term. This indicates the number of elements in the concatenated gradient histogram; The normalized histograms of each image patch are concatenated to obtain the image feature vector.

5. A corporate environmental responsibility assessment device, characterized in that, The device is used as a participating node in an environmental responsibility assessment within a blockchain, and the enterprise environmental responsibility assessment device includes: The data acquisition module is used to collect multi-source heterogeneous datasets from multiple sample enterprises and preprocess the multi-source heterogeneous datasets to obtain multimodal feature vectors, which include image feature vectors, audio feature vectors, text feature vectors, and numerical feature vectors. The data source classification module is used to classify the data source based on the data source corresponding to each feature vector in the multimodal feature vector, and to obtain the internal multimodal feature vector and external multimodal feature vector of each sample enterprise; A cross-modal feature alignment module is used to input the internal multimodal feature vector and the external multimodal feature vector into a multimodal feature fusion model for cross-modal feature alignment to obtain target internal information and target external information. The multimodal feature fusion model includes a multi-head attention module. The target internal information includes the enterprise's first environmental monitoring information and first environmental protection investment information. The target external information includes the second environmental monitoring information and second environmental protection investment information collected by external monitoring agencies. The authenticity assessment module is used to calculate the internal and external difference rate of monitoring indicators for each sample enterprise based on the first environmental monitoring information and the second environmental monitoring information. in, This indicates the rate of difference between internal and external monitoring indicators. These represent internal monitoring indicators for the company. Monitoring indicators representing the external monitoring structure; Based on the first environmental monitoring information and the second environmental monitoring information, internal and external monitoring images of the target pollution emission monitoring area are obtained, and the similarity of the pollution emission monitoring images is calculated based on the internal and external monitoring images: in, This represents the mean value of internal monitoring images. This represents the mean value of external monitoring images. Indicates the variance of internal monitoring images. Indicates the variance of external monitoring images. Describing covariance, and All are constant terms. Indicates the similarity of pollution emission monitoring images; Based on the first environmental protection investment information and the second environmental protection investment information, calculate the internal and external environmental protection investment difference rate for each sample enterprise: in, This indicates the difference rate between domestic and international environmental protection investment. This indicates the amount of environmental protection investment within the company. This indicates the amount of environmental protection investment obtained by external monitoring agencies; An authenticity assessment model is constructed, and the internal and external difference rates of the monitoring indicators, the similarity of the pollution emission monitoring images, and the internal and external difference rates of environmental protection investment are input into the authenticity assessment model to conduct an authenticity assessment, obtaining the authenticity assessment results for each sample enterprise. The authenticity assessment model includes: in, This indicates the results of the authenticity assessment. These represent the weight parameters; The corporate environmental responsibility assessment module is used to input the authenticity assessment results into the target corporate environmental responsibility scoring model for evaluation, and obtain the corporate environmental responsibility score of each sample company. The target corporate environmental responsibility scoring model is constructed based on the local model parameters provided by each environmental responsibility assessment participating node in the blockchain. The corporate environmental responsibility assessment module is further configured to: acquire local model parameters uploaded to the blockchain by each participating node in the environmental responsibility assessment, wherein the local model parameters are obtained by each participating node through gradient training of its local responsibility scoring model; remove abnormal nodes with abnormal behavior from the participating nodes based on their historical behavior information to obtain candidate nodes; acquire historical contribution information of each candidate node and determine the contribution weight of each candidate node based on the historical contribution information; aggregate the local model parameters of each candidate node based on the contribution weight to obtain global model parameters; and update the pre-constructed local responsibility scoring model according to the global model parameters to obtain the target corporate environmental responsibility scoring model.

6. A corporate environmental responsibility assessment device, characterized in that, The corporate environmental responsibility assessment device includes: a memory, a processor, and a corporate environmental responsibility assessment program stored on the memory and executable on the processor, the corporate environmental responsibility assessment program being configured to implement the corporate environmental responsibility assessment method as described in any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a corporate environmental responsibility assessment program, which, when executed by a processor, implements the corporate environmental responsibility assessment method as described in any one of claims 1 to 4.

8. A computer program product, characterized in that, The computer program product includes a corporate environmental responsibility assessment program, which, when executed by a processor, implements the steps of the corporate environmental responsibility assessment method as described in any one of claims 1 to 4.

Images