Signal system and method for urban streetscape visual emotion prediction and application

Abstract

The invention provides a signal system and method for urban street view visual emotion prediction and application. The signal system comprises a monitoring module, a processing module, a first construction module, a second construction module and an output module. The monitoring module is used for collecting electroencephalogram signals and emotion evaluation reaction indexes when the testee watches the streetscape image; the processing module obtains relative power of electroencephalogram signals of a single-image presentation section and a multi-image sequence section; the first construction module is used for constructing an electroencephalogram signal emotion prediction model by using multi-image sequence segment electroencephalogram signal relative power and emotion evaluation reaction indexes, and is applied to single-image presentation segment electroencephalogram signals to obtain an emotion reaction of any streetscape image; the second construction module is used for constructing a machine learning model, obtaining a streetscape image emotion prediction model based on electroencephalogram signal recognition, and applying the streetscape image emotion prediction model to the output module to obtain a city streetscape visual emotion map; the method has the advantages of being sensitive in expression and accurate in emotion capture.

Description

Technical Field

[0001] The present disclosure relates to the technical field of electroencephalogram processing, and particularly to a signal system, method and application for visual emotion prediction of urban street scenes based on electroencephalogram signals and image recognition. Background Art

[0002] Electroencephalogram (EEG) is a method of recording brain activity using electrophysiological indicators. When the brain is active, it is formed by the summation of postsynaptic potentials generated synchronously by a large number of neurons. It records the electrical wave changes during brain activity and is the overall reflection of the electrophysiological activities of brain nerve cells on the cerebral cortex or the scalp surface.

[0003] As an integrated carrier of urban form and human activities, the visual environmental quality of urban street scenes is directly related to the physical and mental health of residents and the spatial perception efficiency. Existing research shows that scientific street scene design can effectively improve the livability of urban space and have a positive impact on the mental health of residents. However, with the acceleration of urbanization and the rapid increase in building density, the proportion of the original natural landscape has decreased significantly, and the street scene environment shows characteristics such as dense building interfaces, fragmented green spaces, and overloaded artificial colors, resulting in a decline in residents' visual comfort and frequent occurrence of derived psychological problems. Urban designers need a precise visual emotion evaluation model of residents to improve urban street scenes. In this context, constructing a precise visual environment-emotion response correlation model has become an urgent need for humanistic urban design.

[0004] Accurately capturing the emotional state under the influence of the urban visual environment is the key to constructing the model. Existing research evaluates the emotional evaluation level of urban residents through online questionnaires or text analysis surveys. However, traditional evaluation methods mainly rely on subjective evaluation systems, and there are technical bottlenecks such as insensitive expression of emotional experience, inaccurate emotional capture, and low processing efficiency.

[0005] In view of this, there is an urgent need to develop an emotion prediction system based on electroencephalogram signals that is sensitive in expression and accurate in emotional capture under the influence of the urban visual environment. Summary of the Invention

[0006] In order to solve the deficiencies of the prior art, the present disclosure provides a signal system, signal processing method and application based on electroencephalogram signals, which are used to solve at least one technical problem in the background art.

[0007] The technical solution adopted by the present disclosure is as follows:

[0008] A signal system for visual emotion prediction of urban street scenes, which predicts the visual emotion of urban street scenes based on electroencephalogram signals and image recognition, includes:

[0009] The acquisition module is used to acquire any street view image and obtain the street view environment indicators corresponding to the street view image; the street view environment indicators include: building density, green view rate, openness, color quantity, and color entropy; among them, the building density, green view rate, and openness are obtained through a semantic segmentation model, and the color quantity and color entropy are obtained through image color features;

[0010] The monitoring module conducts data interaction with the acquisition module and is used to collect the electroencephalogram (EEG) signals and emotional evaluation response indicators of the subject when watching the street view image, including: the subject is tested in at least 2 working conditions, four groups of street view pictures are played in each working condition, a fourteen-channel electroencephalograph is used to measure the EEG signals, and a discrete emotional evaluation form and the SAM scale are used to record the emotional responses of each street view group; the street view image groups are composed of street view images of the same urban environment type;

[0011] The processing module conducts data interaction with the monitoring module and is used to perform signal preprocessing and segmented cutting on the EEG signals, obtain the power spectral density, and obtain the relative power of the full-time EEG signals, including: importing EEG information, filtering, removing artifact signals, and independent component analysis; the segmented cutting divides the full-time EEG signals into a single-picture presentation segment and a multi-picture sequence segment in the time domain; the power spectral density obtains the relative power of the EEG signals in the single-picture presentation segment and the multi-picture sequence segment through fast Fourier transform;

[0012] The first construction module conducts data interaction with the processing module and the monitoring module and is used to construct an EEG signal emotion prediction model using the relative power of the multi-picture sequence segment EEG signals and the emotional evaluation response indicators, and obtain the emotional response of any street view image, including: the EEG signal emotion prediction model is applied to the EEG signals in the single-picture presentation segment to obtain the emotional response of each street view image; moreover, the EEG signal emotion prediction model is constructed using the random forest algorithm;

[0013] The second construction module conducts data interaction with the first construction module and the acquisition module and is used to construct a machine learning model based on the emotional response of the street view image and the street view environment indicators to obtain a street view image emotion prediction model based on EEG signal recognition; the machine learning model is constructed using the gradient boosting tree algorithm;

[0014] The output module conducts data interaction with the second construction module and is used to obtain the scalar of different emotional response indicators corresponding to each street view image through the street view image emotion prediction model for the street view image to be recognized, and use ArcgisPro to plot it on the map to obtain a visual emotional map of the urban street view.

[0015] The obtaining of the street view environment indicators corresponding to the street view image includes:

[0016] Based on the geographic information system platform, equidistant sampling method is used to set spatial sampling points on the road network nodes in the research area, and the sampling density is set to the preset interval;

[0017] At the coordinates of each sampling point, street view images of the four azimuths of due north, due east, due south, and due west are obtained. The image acquisition parameters are set to a 90° horizontal viewing angle and a resolution of 1024×800 pixels, and the time, longitude, and latitude data automatically recorded during the image acquisition process are recorded.

[0018] The steps for obtaining the building density, green view rate, openness, color quantity, and color entropy include:

[0019] The building density is the ratio of the number of building element pixels to the total number of image pixels, and the specific calculation formula is as follows:

[0020] ;

[0021] In the formula, is the building density of the street view image, is the number of building pixels, is the total number of pixels in the image;

[0022] The green view rate is the ratio of the number of green plant element pixels to the total number of image pixels, and the specific calculation formula is as follows:

[0023] ;

[0024] In the formula, is the green view rate of the street view image, is the number of tree pixels, is the number of vegetation pixels, is the number of grassland pixels, is the number of palm tree pixels, is the number of flower pixels, is the total number of pixels in the image;

[0025] The openness is the ratio of the number of sky element pixels to the total number of image pixels, and the specific calculation formula is as follows:

[0026] ;

[0027] In the formula, OP is the openness of the street view image, is the number of sky pixels, is the total number of pixels in the image;

[0028] The calculation of the image color features includes: using the OpenCV and CuPy libraries to read the RGB values of the street view images and calculate the color entropy and color quantity;

[0029] The color entropy calculates the color entropy value based on the information entropy theory, and the specific calculation formula is as follows:

[0030] ;

[0031] In the formula, represents the color entropy, represents the color probability in the image, represents the logarithm to the base 2, represents the number of different colors;

[0032] The number of colors is calculated by calculating the RGB values of the colors and counting the number of unique colors, and the specific calculation formula is as follows:

[0033] ;

[0034] In the formula, represents the number of colors, represents the color value in the image, and E represents the color entropy.

[0035] Collecting the electroencephalogram signals and emotional evaluation response indicators of the subjects when viewing the street scene images includes:

[0036] Designing an electroencephalogram experiment using a Latin square balance scheme, setting four environmental stimulus types: building clusters surrounding, green space, open view, and color entropy settlement, and selecting street scene images with typical representativeness for each type to form a visual stimulus library;

[0037] Using a Latin square design to balance the presentation order, randomly sorting the images within each block, setting the visual stimulus presentation time to a preset time, and having a gray blank screen for a preset period before the stimulus presentation to eliminate visual residue and adjust emotions, while recording the electroencephalogram signals;

[0038] Simultaneously recording the electroencephalogram signals, and the recording frequency bands include five internationally common frequency bands: Delta, Theta, Alpha, Beta, and Gamma;

[0039] Using an electronic display to present visual stimuli, with no other visual interference objects within the sight of the subjects except the electronic display; after each type of street scene picture ends, collecting the discrete emotion evaluation form and SAM scale of the subjects.

[0040] Collecting the discrete emotion evaluation form and SAM scale of the subjects includes:

[0041] The discrete emotion evaluation form contains twelve emotion response indicators with a 5-level rating: excited, calm, peaceful, worried, happy, sad, relaxed, depressed, tired, tense, sad, glad; among them, 1 represents no such type of emotion response, 5 represents a strong such type of emotion response, and the numbers represent the increasing relationship of emotion responses;

[0042] The SAM scale uses 9-level facial icons for emotional evaluation, representing the response levels of valence and arousal;

[0043] Record all behavioral responses, and synchronize and save the EEG data through wireless or wired communication methods.

[0044] The EEG signals are preprocessed and segmented, and the power spectral density is calculated to obtain the relative power of the EEG signals in the single-image presentation segment and the multi-image sequence segment, including:

[0045] Use the standard 10-20 electrode positioning system to perform spatial registration on 14-channel signals;

[0046] Use an FIR band-pass filter to filter out frequency bands below 0.5Hz and above 60Hz, and use a notch filter to eliminate the interference of the 48-52Hz alternating current frequency band;

[0047] Check for abnormal signal segments manually through time-domain spectrograms, and manually remove artifact signal segments such as eye movements and electromyograms; and, use independent component analysis to decompose the signal components, and manually identify and remove the artifact components;

[0048] The segmentation cuts the EEG signals in the time domain according to the picture group and the visual stimulus time period presented by each street view image, and divides them into a single-image presentation segment and a multi-image sequence segment in terms of time-domain characteristics;

[0049] The single-image presentation segment intercepts the time window from 2000ms after the start of the image stimulus to the end of the image, and the multi-image sequence segment intercepts the continuous signal during the presentation of the entire group of images;

[0050] The power spectrum analysis transforms the time-domain characteristics of the EEG signals in the single-image presentation segment and the multi-image sequence segment into frequency-domain characteristics through the fast Fourier transform method, for fourteen electrodes in four brain regions, including: Frontal lobe: AF3, AF4, F3, F4, F7, F8, FC5, FC6, Temporal lobe: T7, T8; Parietal lobe: P7, P8, Occipital lobe: O1, O2;

[0051] Obtain five characteristic frequency bands: Delta, Theta, Alpha, Beta, and Gamma;

[0052] The relative power value is normalized according to the formula, and the expression is as follows: ;

[0053] In the formula, is the absolute power of each frequency band, is the absolute power of the entire 0.5 - 60Hz frequency band; the finally generated feature matrix contains the data structure of the EEG signals in each segment with 5 frequency bands × 14 channels.

[0054] Using the relative power of the multi-image sequence segment EEG signals and the emotional evaluation response index to construct an EEG signal emotion prediction model to obtain the emotional response of any street view image, including:

[0055] Taking the EEG power of the multi-image sequence segment as features and the emotional response index recorded in the experiment as labels; and, using methods of injecting Gaussian noise and feature scaling to enhance the strategy to improve the model robustness; at the same time, constructing combined features through second-order polynomial feature expansion, retaining 95% variance through PCA dimensionality reduction, then performing multi-objective feature selection based on mutual information, screening the top 50 principal component features with the most discriminative power, and using the random forest algorithm for regression modeling;

[0056] Performing hyperparameter search based on the TPE algorithm of Optuna, and the random forest adjusts the tree depth and the number of split samples; and, using the macro-average F1 score of five-fold cross-validation as the optimization objective, synchronously monitoring the accuracy of the validation set to obtain the finally retained model components; the finally retained model components include: a feature processor, a label encoder, and independent classifiers for each emotional target;

[0057] Applying the EEG emotion prediction model to the EEG signals of the single-image presentation segment, automatically loading the trained processing pipeline during deployment, and performing the following processes on the input data: the feature processor automatically applies the same polynomial expansion and PCA transformation; dynamically matching the feature subset selected during training; performing multi-emotion prediction in parallel, and the results are restored to the original scale values through inverse label encoding; finally outputting the original metadata and the predicted values of each emotional dimension.

[0058] Constructing a machine learning model based on the emotional response of the street view image and the street view environment index to obtain a street view image emotion prediction model based on EEG signal recognition, including:

[0059] Taking the street view environment index of the street view image as features and the corresponding emotional evaluation prediction value as labels; in the data preprocessing stage, adopting a robust standardization process, performing one-dimensional outlier processing on the environmental index, and using the 5%-95% quantiles for Winsorize truncation; calculating the mean and covariance matrix through the MinCovDet algorithm to achieve the standardization transformation based on the Mahalanobis distance; enhancing the training data by adding 5% Gaussian noise;

[0060] The machine learning model uses the GradientBoostingRegressor framework of Scikit-learn. Based on the Gradient Boosting Decision Tree algorithm, it automatically searches for key parameters, including n_estimators, learning_rate, and max_depth, through Bayesian optimization. The optimization objective is the R² score of cross-validation. A dedicated model is trained independently for each target variable, and the optimal parameter combination is saved. The training process is monitored through visualizing the learning curve.

[0061] In the evaluation stage of the machine learning model, K-fold cross-validation is used to record the fluctuations of the MAE and R² metrics. Finally, the prediction error distribution of each target variable is output on the test set. During deployment, the preprocessor and model parameters are automatically loaded, the same standardization process is performed on the input data, and batch prediction is carried out to output a result analysis report containing a 95% confidence interval. Moreover, the machine learning model is saved in pkl format, and each emotion is saved as a corresponding file.

[0062] A method for predicting visual emotions of urban street scenes, based on the signal system for predicting visual emotions of urban street scenes, includes:

[0063] Collect any street scene image and obtain the street scene environment indicators corresponding to the street scene image. The street scene environment indicators include building density, green view rate, openness, color quantity, and color entropy. Among them, the building density, green view rate, and openness are obtained through a semantic segmentation model, and the color quantity and color entropy are obtained through image color features.

[0064] Collect the electroencephalogram (EEG) signals and emotional evaluation response indicators of the subjects when watching the street scene image, including: The subjects are tested in four working conditions. Four groups of street scene pictures are played in each working condition. A fourteen-channel electroencephalograph is used to measure the EEG signals, and a discrete emotional evaluation form and the SAM scale are used to record the emotional responses of each street scene group. The street scene image groups are composed of street scene images of the same urban environment type.

[0065] Perform signal preprocessing and segmented cutting on the EEG signals, and calculate the power spectral density to obtain the relative power of the full-time EEG signals, including: Import EEG information, filter, remove artifact signals, and perform independent component analysis. The segmented cutting divides the EEG signals into a single-picture presentation segment and a multi-picture sequence segment in the time domain. The power spectral density obtains the relative power of the EEG signals in the single-picture presentation segment and the multi-picture sequence segment through fast Fourier transform.

[0066] Construct an EEG signal emotion prediction model using the relative power of the multi-image sequence segment EEG signals and the emotion evaluation response indicators to obtain the emotional response of any street view image, including: applying the EEG signal emotion prediction model to the EEG signals of the single-image presentation segment to obtain the emotional response of each street view image; and, the EEG signal emotion prediction model is constructed using the random forest algorithm;

[0067] Construct a machine learning model based on the emotional response of the street view image and the street view environment indicators to obtain a street view image emotion prediction model based on EEG signal recognition; and, the machine learning model is constructed using the gradient boosting tree algorithm;

[0068] According to the street view image to be recognized through the street view image emotion prediction model, use emotion prediction to obtain the scalar of different emotion response indicators corresponding to each street view image, and use ArcgisPro to draw it on the map to obtain the urban street view visual emotion map.

[0069] An application based on the above method includes:

[0070] When initializing the EEG signal emotion prediction model, automatically load the preprocessor and the trained gradient boosting tree model;

[0071] During prediction, perform numerical type verification and missing value processing on the input data, perform robust standardization based on MinCovDet, and then call the independent models of each target variable for parallel prediction;

[0072] Batch process the street view environment indicators of all street view images in the research area, calculate the mean of the street view environment indicators in four directions for each coordinate point as the feature data of this point, and obtain the predicted values of 14 emotion response indicators corresponding to each street view image;

[0073] Import the CSV file containing emotion indicators in ArcGIS Pro, set the longitude and latitude fields to generate a temporary point layer, and export it as a formal point feature class; use the spatial join tool to set the road network data as the target feature, and the emotion point data as the join feature, set the matching method, and retain all emotion indicator fields in the field mapping;

[0074] After completing the data connection, select the rendering method in the symbol system, specify the field as the emotion indicator; and generate an emotion raster surface, identify statistically significant emotion aggregation areas, and the results are represented by a gradient color to indicate high / low value clustering sections, to obtain the urban street view visual emotion map of the research area.

[0075] The beneficial effects of the present disclosure are:

[0076] The signal system described in this disclosure collects street view images based on urban geographic information big data, calculates visual environment indicators such as building density, green view rate, and openness through a pixel-level semantic segmentation model, and obtains color quantity and color entropy parameters by analyzing the color characteristics of the images. Then, multi-band electroencephalogram (EEG) signals of the subjects are collected during the viewing of street view images through EEG experiments, and subjective emotional response data of the discrete emotion evaluation form and the SAM scale are synchronously recorded. Then, preprocessing such as FIR filtering, artifact removal, and ICA component analysis is performed on the EEG signals, which are segmented into single-image presentation segments and multi-image sequence segments. The power spectral density characteristics of each frequency band are calculated through fast Fourier transform. A random forest non-linear regression model is constructed from the multi-image sequence segments to realize the mapping from EEG signals to emotion scores, and it is applied to the single-image presentation segments for emotion prediction. Finally, a machine learning prediction model of street view environment indicators and emotional responses is established based on the gradient boosting tree algorithm to realize the visual emotion evaluation of large-scale street view images. Through the integration of EEG signals and artificial intelligence technology, this disclosure provides an objectively quantifiable visual emotion evaluation and prediction tool for urban landscape design, helps to create a human settlement environment that promotes mental health, promotes the practical application of environmental psychology in the construction of smart cities, and has the advantages of sensitive expression and accurate emotion capture. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] Figure 1 It is a system block diagram of the signal system described in this disclosure;

[0078] Figure 2 It is a flowchart for obtaining street view images and calculating street view environment indicators in this disclosure;

[0079] Figure 3 It is an experimental flowchart designed in this disclosure;

[0080] Figure 4 It is twelve discrete emotions and two-dimensional emotion evaluations designed in this disclosure;

[0081] Figure 5 It is the EEG signal measurement process and frequency domain analysis method in this disclosure;

[0082] Figure 6 It is a flowchart for training and predicting the machine learning model in this disclosure;

[0083] Figure 7 It is a flowchart of the signal processing method described in this disclosure. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0084] To better understand the above objects, features, and advantages of the present disclosure, the present disclosure will be further described below with reference to the accompanying drawings and embodiments. It should be noted that, without conflict, the embodiments of the present application and the features in the embodiments can be combined with each other. The present disclosure will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present disclosure and cannot be used to limit the protection scope of the present disclosure.

[0085] As is well known, electroencephalogram (EEG) technology provides a new technical path for the objective quantification of street view visual stimuli with its millisecond-level time resolution and cortical potential detection sensitivity. By analyzing the power spectral density of characteristic frequency bands, a mapping relationship between EEG signals and emotional states can be established, which brings a breakthrough in the scientific evaluation of street view visual environments.

[0086] In view of this, the present disclosure also discloses an embodiment: Specific Embodiment 1:

[0088] Such as Figure 1 , a signal system for predicting urban street view visual emotions, which predicts urban street view visual emotions based on brain wave signals and image recognition, including: an acquisition module 100, a monitoring module 200, a processing module 300, a first construction module 400, a second construction module 500, and an output module 600; wherein, the acquisition module 100 is used to acquire any street view image and obtain the street view environment index corresponding to the street view image; the monitoring module 200 performs data interaction with the acquisition module 100 and is used to acquire the EEG signal and emotional evaluation response index of the subject when viewing the street view image; the processing module 300 performs data interaction with the monitoring module 200 and is used to perform signal preprocessing and segmentation cutting on the EEG signal, and perform power spectral density calculation to obtain the relative power of the EEG signal in the single-image presentation segment and the multi-image sequence segment; the first construction module 400 performs data interaction with the processing module 300 and the monitoring module 200 and is used to construct an EEG signal emotion prediction model by using the relative power of the multi-image sequence segment EEG signal and the emotional evaluation response index to obtain the emotional response of any street view image; the second construction module 500 performs data interaction with the first construction module 400 and the acquisition module 100 and is used to construct a machine learning model according to the emotional response of the street view image and the street view environment index to obtain a street view image emotion prediction model based on EEG signal recognition; the output module 600 performs data interaction with the second construction module 500 and is used to output a urban street view visual emotion map according to the street view image to be recognized through the street view image emotion prediction model.

[0089] Figure 2This is a flowchart for obtaining street view images and calculating street view environment indicators in this embodiment. Based on the geographic information system platform, equidistant sampling method is used to set spatial sampling points at the road network nodes in the study area, and the sampling density is set at an interval of 100 meters. By calling the Baidu Street View Map API service, street view images at four azimuth angles of 0°, i.e., due north, 90°, i.e., due east, 180°, i.e., due south, and 270°, i.e., due west, are obtained at the coordinates of each sampling point. The image acquisition parameters are set as a 90° horizontal viewing angle and a resolution of 1024×800 pixels, as close as possible to the human eye's viewing angle. The shooting time and latitude and longitude data are automatically recorded during the image acquisition process.

[0090] The SegNext semantic segmentation model based on the neural network Transformer architecture is used. The mIoU parameter on the ADE20K dataset reaches 43.45, and image parsing training is carried out. This model accurately identifies the pixel regions of building, tree, vegetation, grassland, palm tree, flower, and sky elements for calculating street view environment indicators such as building density, green view rate, and openness. Trees can be arbors or shrubs, but do not include palm trees; flowers refer to ornamental flowering plants.

[0091] The above-mentioned building density is the ratio of the number of building element pixels to the total number of pixels in the image. The specific calculation formula is as follows: ;

[0092] In the formula, BD is the building density of the street view image, Nb is the number of building pixels, Nt is the total number of pixels in the image;

[0093] The green view rate is the ratio of the number of green plant element pixels to the total number of pixels in the image. The specific calculation formula is as follows:

[0094] ;

[0095] In the formula, GVR is the green view rate of the street view image, Nt is the number of tree pixels, Nv is the number of vegetation pixels, Ng is the number of grassland pixels, Np is the number of palm tree pixels, Nf is the number of flower pixels, Nt is the total number of pixels in the image;

[0096] The openness is the ratio of the number of sky element pixels to the total number of pixels in the image. The specific calculation formula is as follows: ;

[0097] In the formula, OP is the openness of the street view image, Ns is the number of sky pixels, is the total number of pixels of the image;

[0098] Through the RGB color space conversion of OpenCV and the accelerated calculation of CuPy, extract the color feature parameters of the image to calculate the number of colors and color entropy:

[0099] (1) The color entropy is to calculate the information entropy value based on the RGB color histogram, which reflects the complexity of the color distribution. The specific calculation formula is as follows: ;

[0100] In the formula, represents the color entropy, represents the color probability in the image, represents the logarithm with base 2, represents the number of different colors;

[0101] (2) The number of colors is the number of unique hues after discretizing the hue channel H value in the RGB space, with 8-bit quantization. The specific calculation formula is as follows: ;

[0102] In the formula, represents the number of colors, represents the color value in the image, and E represents the color entropy;

[0103] The calculation results of all indicators are stored in the spatial database together with the geographical coordinates.

[0104] In this embodiment, the electroencephalogram signals and emotional response indicators of the subjects when watching street view images are collected through electroencephalogram experiments, including: Figure 3 is the experimental flow chart of this embodiment. The experimental design adopts the Latin square balance scheme, and four types of environmental stimulation are set, namely, buildings surrounded by buildings, that is, building density > 60%, green space, that is, green view rate > 60%, open view, that is, openness > 60%, and color entropy settlement, that is, color entropy > 9.0. Each type selects 72 typical and representative street view images to form a stimulus library. The Latin square design is used to balance the presentation order, and the images within each block are randomly sorted. The visual stimulus presentation time is 10 s, and there is a 15 s gray blank screen before the stimulus presentation to eliminate visual afterimage and adjust the mood.

[0105] Use the head-mounted electroencephalograph Emotiv EPOC× wireless electroencephalogram acquisition system. For example, 14 channels are arranged according to the international 10-20 system, and the sampling rate is 128 Hz to synchronously record the electroencephalogram signals. The experimental environment is a blank room with suitable temperature. A 120 Hz refresh rate electronic display is used to present visual stimuli, and there are no other visual interference objects in the field of view of the subjects except the electronic display.

[0106] Figure 4For the twelve discrete emotions and two-dimensional emotion evaluation in this embodiment, after each type of street view picture ends, the subjects need to complete:

[0107] (1) Discrete emotion evaluation - Use a 1-5 level scalar scale to select the degree of conformity from 12 emotion dimensions, including: excited, calm, peaceful, worried, happy, sad, relaxed, depressed, tired, tense, sad, glad; 1 represents no such type of emotional reaction, 5 represents a strong emotional reaction of this type, and the numbers represent the increasing relationship of emotional reactions;

[0108] (2) SAM scale - Perform a 9-level rating on two dimensions of valence and arousal, and select the picture that best matches the current situation for rating. All behavioral responses are recorded through the E-Prime 2.0 system, and the EEG data is strictly synchronized through Bluetooth signals and saved on the computer.

[0109] The experimental process includes: start interface, visual stimulus, end interface. The start interface provides experimental operations and informed consent forms for the experiment. The visual stimulus includes four groups of picture processes designed by Latin square balance. After each group of pictures is played, the subject is required to fill out an emotion evaluation form on the electronic display interface. The end interface notifies the end of the experimental process. The final dataset is the original EEG signal with time-domain features, such as in.edf format.

[0110] In this embodiment, the preprocessing, segmentation, and power spectral density calculation of EEG signals include: based on the MATLAB R2023b operating environment, call the EEGLAB toolbox to import the original EEG data in.edf format, and use the standard 10-20 electrode positioning system to perform spatial registration on the 14-channel signals. Use filters to filter the EEG signals: (1) FIR band-pass filter 0.5-60Hz to filter low-pass noise and high-frequency noise; (2) 50Hz power frequency notch filter, bandwidth ±2Hz, to eliminate the interference in the alternating current frequency band, and generate a time-domain spectrogram through the Scroll channel activities of EEGLAB, and manually mark the abnormal signal segments. Use the ICA algorithm for independent component analysis, and automatically detect artifact components through the ADJUST plug-in. The segmentation is cut according to the picture stimulus presentation time recorded during the experiment. For a single picture presentation segment, the time window from 2000ms after the start of the image stimulus to the end of the image is intercepted, and for a multi-picture sequence segment, the continuous signal during the presentation of the entire group of images is intercepted; each group contains 18 street view images with a duration of 180s.

[0111] Figure 5For the EEG signal measurement process and frequency domain analysis method in this embodiment, the frequency domain feature extraction uses the FFT transform weighted by the Hanning window. For the fourteen electrodes in four brain regions, including: Frontal lobe: AF3, AF4, F3, F4, F7, F8, FC5, FC6; Temporal lobe: T7, T8; Parietal lobe: P7, P8; Occipital lobe: O1, O2; A total of five characteristic frequency bands are calculated for Delta, with a frequency of 0.5 - 4 Hz, Theta, with a frequency of 4 - 8 Hz, Alpha, with a frequency of 8 - 13 Hz, Beta, with a frequency of 13 - 30 Hz, and Gamma, with a frequency of 30 - 60 Hz.

[0112] The relative power value is normalized according to the formula, and the expression is as follows: ;

[0113] In the formula, is the absolute power of each frequency band, is the absolute power of the full frequency band from 0.5 - 60 Hz. The finally generated feature matrix contains the EEG signal data structure of each segment with 5 frequency bands × 14 channels.

[0114] In this embodiment, constructing a model to predict the emotional response of a single street view image includes: Figure 6 This is the flowchart of machine learning model training and prediction in this embodiment. This figure shows the process of constructing an EEG emotion prediction model from EEG data and emotional evaluation, and applying it to street view image street view environment indicators for machine learning training to construct a street view image emotion prediction model. The following are the steps to construct an EEG emotion prediction model:

[0115] 1. Data preprocessing and feature engineering: The EEG power of multi - image sequence segments is used as features, and the corresponding emotional response indicators recorded in the experiment are used as labels. Gaussian noise injection with μ = 0, σ = 0.05 and feature scaling with random coefficients of 0.95 - 1.05 are adopted to enhance the model robustness. Composite features are constructed through second - order polynomial feature expansion, and PCA dimensionality reduction is used to retain 95% of the variance. Multi - objective feature selection is performed based on mutual information MI to screen the top 50 principal component features with the most discriminative power for each emotional label, and the random forest algorithm is used for discrete emotional labels.

[0116] 2. Multi - objective model optimization: Hyperparameter search is performed based on the TPE algorithm of the model parameter optimization tool Optuna, such as 40 iterations, and the tree depth of the random forest is adjusted to 3 - 10 and the number of split samples to 2 - 10. The macro - average F1 score of cross - validation is used as the optimization objective, and the accuracy of the validation set is monitored synchronously. The finally retained model components include: feature processor, label encoder, and independent classifiers for each emotional target. The feature processor includes standardization + polynomial expansion + PCA; the label encoder supports floating - point safe encoding.

[0117] 3. Prediction System Implementation: When deployed, the trained processing pipeline is automatically loaded, and the following processes are performed on the input data, such as 70-dimensional features:

[0118] (1) The feature processor automatically applies the same polynomial expansion and PCA transformation;

[0119] (2) Dynamically match the feature subset selected during training, and fault-tolerantly skip invalid features;

[0120] (3) Perform multi-sentiment prediction in parallel, and the results are restored to the original scale values through inverse label encoding.

[0121] The system output includes the original metadata and the predicted values of each sentiment dimension, supporting batch prediction in CSV format.

[0122] The following are the steps to build a street view image sentiment prediction model:

[0123] Use the street view environment indicators of street view images as features, and the corresponding sentiment evaluation predicted values as labels. In the data preprocessing stage, a robust standardization process is adopted to perform single-dimensional outlier processing on five environmental indicators such as building density, and Winsorize truncation is performed using the 5%-95% quantiles. The mean and covariance matrix are calculated through the MinCovDet algorithm to achieve the standardization transformation based on the Mahalanobis distance. Gaussian noise with a standard deviation of 0.05 is injected in the data augmentation link.

[0124] The model training uses the GradientBoostingRegressor framework of machine learning Scikit-learn. Based on the Gradient Boosting Decision Tree algorithm, the key parameters are automatically searched through Bayesian optimization, including n_estimators 50 - 300, learning_rate 0.01 - 0.3, max_depth 3 - 10. The optimization objective is the R² score of cross-validation. An exclusive model is independently trained for each target variable and the optimal parameter combination is saved. The training process is monitored through visualizing the learning curve.

[0125] In the model evaluation stage, K-fold cross-validation is adopted to record the fluctuations of the MAE and R² metrics. Finally, the predicted error distribution of each label is output on the test set, and the optimal model is saved as the street view image sentiment prediction model. The model is saved in pkl format, and each sentiment is saved as a corresponding file. When deployed, the preprocessor and model parameters are automatically loaded, and the same standardization process is performed on the input data to batch predict the sentiment evaluation predicted values of the street view images in the research area.

[0126] The signal system described in this embodiment constructs a scientific emotional prediction model for street view images through the objective mapping relationship between EEG signals and emotional responses, combined with the analysis of street view environment characteristics, breaks through the subjective limitations of traditional questionnaires, and realizes cross-scale analysis from microscopic neural signals to macroscopic urban emotional maps. The generated emotional map can intuitively display the emotional distribution characteristics of urban spaces, providing a quantitative basis for urban planning and landscape design. Specific Embodiment 2:

[0128] This disclosure provides an embodiment:

[0129] As Figure 7 , a method for predicting visual emotions of urban street views, based on the signal system described in Specific Embodiment 1, for predicting the visual emotions of urban street views by collecting EEG signals, including:

[0130] S100, obtaining a street view image of a target area and calculating street view environment indicators;

[0131] S200, collecting EEG signals and emotional response indicators of the subject when viewing the street view image through an EEG experiment;

[0132] S300, preprocessing, segmenting, and calculating the power spectral density of the EEG signals;

[0133] S400, constructing a non-linear regression model to predict the emotional response of a single street view image;

[0134] S500, establishing a machine learning prediction model for street view environment indicators and emotional responses;

[0135] S600, applying the prediction model to draw an urban street view emotional map.

[0136] The specific steps are: collecting any street view image and obtaining the street view environment indicators corresponding to the street view image; collecting the EEG signals and emotional evaluation response indicators of the subject when viewing the street view image; performing signal preprocessing and segmenting the EEG signals, and calculating the power spectral density to obtain the relative power of the EEG signals in the single-image presentation segment and the multi-image sequence segment; constructing an EEG signal emotional prediction model using the relative power of the multi-image sequence segment EEG signals and the emotional evaluation response indicators to obtain the emotional response of any street view image; constructing a machine learning model based on the emotional response of the street view image and the street view environment indicators to obtain a street view image emotional prediction model based on EEG signal recognition; and outputting an urban street view visual emotional map according to the street view image emotional prediction model for the street view image to be recognized. Specific Embodiment 3:

[0138] This disclosure also provides an embodiment:

[0139] Drawing a visual emotional map of urban street scenes using the prediction model described in Specific Embodiment 1, including:

[0140] At initialization, the preprocessor and the trained gradient boosting tree model are automatically loaded. During prediction, strict numerical type checking and missing value processing are first performed on the input data. Robust standardization based on the anomaly detection algorithm MinCovDet consistent with the training stage is applied, using the saved mean and covariance matrices. Subsequently, the independent models for each emotion classification are called for parallel prediction.

[0141] Batch process the street scene environment indicators of all street scene images in the study area. Calculate the mean of the street scene environment indicators in four directions for each coordinate point as the feature data of that point. After correcting the outliers in the feature data, generate the prediction results. The output file retains the original data and appends a column of predicted values. At the same time, descriptive statistics and 95% confidence interval analysis are provided. All abnormal situations are accompanied by detailed error logs, and the predicted values of 14 emotional response indicators corresponding to each street scene image are obtained.

[0142] Import the CSV file containing emotional indicators into ArcGIS Pro 3.1.5, set the longitude and latitude fields to generate a temporary point layer, and export it as a formal point feature class. Subsequently, use the Spatial Join tool to set the road network data as the target feature and the emotional point data as the join feature, with the matching method being "closest", and all emotional indicator fields are retained in the field mapping.

[0143] After completing the data connection, select the "Graduated Colors" rendering method in the Symbology. The value field is specified as the emotional indicator, and a blue - red gradient color ramp is used. For example, red represents the high arousal value of the emotion, and green represents the low arousal value. Use the "Polyline to Raster" tool to generate the emotional raster surface. Finally, use the "Hot Spot Analysis" tool to identify statistically significant emotional aggregation areas, with p < 0.05. The results are represented by a red - blue gradient indicating high / low value clustering sections, and the emotional map of the street scene images in the study area is obtained.

[0144] The application described in this embodiment innovatively introduces electroencephalogram (EEG) signal analysis technology into the field of urban environmental emotion response evaluation, breaks through the limitations of traditional subjective questionnaire survey methods, establishes an EEG emotion prediction model and a street view image emotion prediction model based on physiological signals, and significantly improves the scientificity and reliability of evaluation results. In addition, through advanced machine learning algorithms, the automation and batch processing of urban street view emotion evaluation are realized, changing the inefficient mode of traditional manual point-by-point evaluation, and providing a scientific method for the drawing of emotion maps. This embodiment has strong repeatability through a standardized analysis process, and can achieve rapid processing and analysis of urban street view data at different scales through EEG experiments, helping to reduce the error of human subjective perception expression, realizing seamless expansion from individual streets to the entire urban area, and providing innovative technical support and decision-making basis for creating a humanized urban space.

[0145] Although the embodiments of the present disclosure have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present disclosure, and the scope of the present disclosure is defined by the appended claims and their equivalents.

Claims

1. A signal system for visual emotion prediction of urban street scenes, which predicts the visual emotion of urban street scenes based on electroencephalogram signals and image recognition, is characterized in that Including: A collection module, which is used to collect any street view image and obtain the street view environment indicators corresponding to the street view image; The street view environment indicators include: building density, green view rate, openness, color quantity and color entropy; among them, the building density, green view rate and openness are obtained through a semantic segmentation model, and the color quantity and color entropy are obtained through image color features; A monitoring module, which conducts data interaction with the collection module, and is used to collect the electroencephalogram signals and emotional evaluation reaction indicators of the subject when watching the street view image, including: testing the subject in at least 2 working conditions, playing four groups of street view pictures in each working condition, using a fourteen-channel electroencephalogram collector to measure the electroencephalogram signals, and using a discrete emotional evaluation form and a SAM scale as the emotional reaction records for each group of street view pictures; the street view image groups are composed of street view images of the same urban environment type; A processing module, which conducts data interaction with the monitoring module, and is used to perform signal preprocessing and segment cutting on the electroencephalogram signals, and obtain the power spectral density, and obtain the relative power of the full-time electroencephalogram signals, including: importing EEG information, filtering, removing artifact signals, and independent component analysis; the segment cutting divides the full-time electroencephalogram signals into a single-picture presentation segment and a multi-picture sequence segment in the time domain; the power spectral density obtains the relative power of the electroencephalogram signals in the single-picture presentation segment and the multi-picture sequence segment through fast Fourier transform; A first construction module, which conducts data interaction with the processing module and the monitoring module, and is used to construct an electroencephalogram signal emotion prediction model by using the relative power of the multi-picture sequence segment electroencephalogram signals and the emotional evaluation reaction indicators, and obtain the emotional reaction of any street view image, including: applying the electroencephalogram signal emotion prediction model to the electroencephalogram signals in the single-picture presentation segment to obtain the emotional reaction of each street view image; moreover, the electroencephalogram signal emotion prediction model is constructed by using the random forest algorithm; A second construction module, which conducts data interaction with the first construction module and the collection module, and is used to construct a machine learning model according to the emotional reaction of the street view image and the street view environment indicators, and obtain a street view image emotion prediction model based on electroencephalogram signal recognition; the machine learning model is constructed by using the gradient boosting tree algorithm; An output module, which conducts data interaction with the second construction module, and is used to obtain a scalar of different emotional reaction indicators corresponding to each street view image through the street view image emotion prediction model according to the street view image to be recognized, and use ArcgisPro to draw it on the map to obtain a visual emotional map of the urban street view.

2. The signal system for visual emotion prediction of urban street scenes according to claim 1, characterized in that The obtaining of the street view environment indicators corresponding to the street view image includes: Based on a geographic information system platform, equidistant sampling method is used to set spatial sampling points on the road network nodes in the research area, and the sampling density is set to a preset interval; At the coordinates of each sampling point, street view images of the four azimuths of due north, due east, due south and due west are obtained, the image acquisition parameters are set to a 90° horizontal viewing angle and a resolution of 1024×800 pixels, and the time and longitude and latitude data automatically recorded during the image acquisition process are recorded.

3. The signal system for urban street view visual emotion prediction according to claim 1, characterized in that, The steps for obtaining the building density, green view rate, openness, number of colors, and color entropy include: The building density is the ratio of the number of building element pixels to the total number of pixels in the image. The specific calculation formula is as follows: ； In the formula, BD is the building density of the street view image, is the number of building pixels, is the total number of pixels in the image; The green view rate is the ratio of the number of green plant element pixels to the total number of pixels in the image. The specific calculation formula is as follows: ； Wherein, is the green view rate of the street view image, is the number of tree pixels, is the number of vegetation pixels, is the number of grassland pixels, is the number of palm tree pixels, is the number of flower pixels, is the total number of pixels of the image; The openness is the ratio of the number of sky element pixels to the total number of pixels in the image. The specific calculation formula is as follows: ； Where OP is the openness of the street view image, is the number of sky pixels, is the total number of pixels in the image; The calculation of the image color features includes: using the OpenCV and CuPy libraries to read the RGB values of the street view image and calculate the color entropy and the number of colors; The color entropy calculates the color entropy value based on the information entropy theory. The specific calculation formula is as follows: ； In the formula, represents the color entropy, represents the color probability in the image, represents the logarithm to the base 2, represents the number of different colors; The number of colors is to calculate the RGB values of the colors and count the number of unique colors. The specific calculation formula is as follows: ； In the formula, represents the number of colors, represents the color value in the image, and E represents the color entropy.

4. The signal system for visual emotion prediction of urban street scenes according to claim 1, characterized in that, The collection of the EEG signals and emotional evaluation response indicators of the subjects when viewing the street view image includes: Adopt a Latin square balance scheme to design an EEG experiment, set four environmental stimulus types: building cluster embrace, green space, open view, and color entropy settlement. Select typical representative street view images for each type to form a visual stimulus library; Adopt a Latin square design to balance the presentation order. The images within each block are randomly sorted. The visual stimulus presentation time is a preset time. There is a preset period of gray blank screen before the stimulus presentation to eliminate visual afterimages and adjust emotions, and at the same time record the EEG signals; Synchronously record the EEG signals. The recording frequency bands include five internationally common frequency bands: Delta, Theta, Alpha, Beta, and Gamma; Use an electronic display to present visual stimuli. There are no other visual interference objects in the field of view of the subjects except the electronic display; after each type of street view picture ends, collect the discrete emotion evaluation form and SAM scale of the subjects.

5. The signal system for urban street view visual emotion prediction according to claim 4, characterized in that The collection of the discrete emotion evaluation form and SAM scale of the subjects includes: The discrete emotion evaluation form contains twelve emotion response indicators with a 5-level scoring: excited, calm, peaceful, worried, happy, sad, relaxed, depressed, tired, tense, sad, glad; among them, 1 represents no such type of emotion response, 5 represents a strong emotion response of this type, and the numbers represent the increasing relationship of emotion responses; The SAM scale uses 9-level facial expression diagrams for emotion evaluation, representing the response degrees of valence and arousal; Record all behavioral responses, and synchronize and save the EEG data through wireless or wired communication methods.

6. The signal system for urban street view visual emotion prediction according to claim 1, characterized in that The signal preprocessing and segmentation cutting of the EEG signals, and obtaining the power spectral density, obtaining the relative power of the full-time EEG signals include: importing EEG information, filtering, removing artifact signals, and independent component analysis; the segmentation cutting divides the full-time EEG signals into single-picture presentation segments and multi-picture sequence segments in the time domain; the power spectral density obtains the relative power of the EEG signals in the single-picture presentation segments and multi-picture sequence segments through fast Fourier transform, including: Adopt a standard 10-20 electrode positioning system for spatial registration of 14-channel signals; Adopt a FIR band-pass filter to filter out frequency bands below 0.5Hz and above 60Hz, and use a notch filter to eliminate the interference of the 48-52Hz alternating current frequency band; Check the manually marked abnormal signal segments through the time-domain spectrogram, and manually remove the artifact signal segments such as eye movements and electromyograms; and, decompose the signal components using independent component analysis, and manually identify and remove the artifact components; The segmented cutting segments the EEG signals in the time domain according to the picture group and the visual stimulation time period presented by each street view image, and divides the EEG signals into single-picture presentation segments and multi-picture sequence segments in terms of time-domain characteristics; The single-picture presentation segment intercepts the time window from 2000 ms after the start of the image stimulation to the end of the image, and the multi-picture sequence segment intercepts the continuous signals during the presentation of the entire group of images; Power spectrum analysis converts the time-domain characteristics of the EEG signals in the single-picture presentation segment and the multi-picture sequence segment into frequency-domain characteristics through the fast Fourier transform method, for fourteen electrodes in four brain regions, including: frontal lobe: AF3, AF4, F3, F4, F7, F8, FC5, FC6, temporal lobe: T7, T8; parietal lobe: P7, P8, occipital lobe: O1, O2; Obtain five characteristic frequency bands of Delta, Theta, Alpha, Beta, and Gamma; The relative power value is normalized according to the formula, and the expression is as follows: ; Wherein, is the absolute power of each frequency band, is the absolute power of the full frequency band of 0.5 - 60 Hz; the finally generated feature matrix contains the EEG signal data structure of each segment with 5 frequency bands × 14 channels.

7. The signal system for visual emotion prediction of urban street scenes according to claim 1, wherein The construction of the EEG signal emotion prediction model using the relative power of the multi-picture sequence segment EEG signals and the emotion evaluation response index to obtain the emotion response of any street view image, including: Taking the EEG power of the multi-picture sequence segment as the feature and the emotion response index recorded in the experiment as the label; and, adopting methods such as injecting Gaussian noise and feature scaling to enhance the strategy to improve the model robustness; at the same time, constructing combined features through second-order polynomial feature expansion, retaining 95% variance through PCA dimensionality reduction, and then performing multi-objective feature selection based on mutual information to screen the top 50 most discriminative principal component features, and using the random forest algorithm for regression modeling; Perform hyperparameter search based on the TPE algorithm of Optuna, and adjust the tree depth and split sample number of the random forest; and, use the macro-average F1 score of five-fold cross-validation as the optimization target, and synchronously monitor the accuracy of the validation set to obtain the finally retained model components; the finally retained model components include: feature processor, label encoder, and independent classifiers for each emotion target; Apply the EEG signal emotion prediction model to the EEG signals in the single-picture presentation segment. When deploying, automatically load the trained processing pipeline and perform the following processes on the input data: the feature processor automatically applies the same polynomial expansion and PCA transformation; dynamically matches the feature subset selected during training; performs multi-emotion prediction in parallel, and the results are restored to the original scale values through inverse label encoding; finally, the output contains the original metadata and the predicted values of each emotion dimension.

8. The signal system for visual emotion prediction of urban street scenes according to claim 1, characterized in that, The construction of a machine learning model based on the emotion response of the street view image and the street view environment index to obtain a street view image emotion prediction model based on EEG signal recognition, including: Use the street view environmental indicators of street view images as features, and use the corresponding predicted values of emotional evaluation as labels; in the data preprocessing stage, adopt a robust standardization process, perform one-dimensional outlier processing on the environmental indicators, and use the 5%-95% quantiles for Winsorize truncation; calculate the mean and covariance matrix through the MinCovDet algorithm to achieve the standardization transformation based on the Mahalanobis distance; enhance the training data by adding 5% Gaussian noise; The machine learning model uses the GradientBoostingRegressor framework of Scikit-learn. Based on the gradient boosting tree algorithm, automatically search for key parameters through Bayesian optimization, including n_estimators, learning_rate, max_depth; the optimization goal is the R² score of cross-validation. Train an exclusive model for each target variable independently and save the optimal parameter combination, and monitor the training process through visualizing the learning curve; In the evaluation stage of the machine learning model, use K-fold cross-validation, record the fluctuations of the MAE and R² indicators, and finally output the prediction error distribution of each target variable on the test set; when deploying, automatically load the preprocessor and model parameters, perform the same standardization process on the input data, batch predict and output a result analysis report containing a 95% confidence interval; moreover, the machine learning model is saved in pkl format, and each emotion is saved as a corresponding file.

9. A method for predicting visual emotions of urban street scenes, based on the signal system for predicting visual emotions of urban street scenes according to any one of claims 1-8, characterized in that, Including: Collect any street view image and obtain the street view environmental indicators corresponding to the street view image; the street view environmental indicators include building density, green view rate, openness, color quantity, and color entropy; among them, the building density, green view rate, and openness are obtained through a semantic segmentation model, and the color quantity and color entropy are obtained through image color features; Collect the electroencephalogram signals and emotional evaluation reaction indicators of the subjects when watching the street view images, including: the subjects are tested in four working conditions, each working condition plays four groups of street view pictures, use a fourteen-channel electroencephalogram collector to measure the electroencephalogram signals, and use a discrete emotional evaluation form and a SAM scale to record the emotional reactions of each street view group; the street view image groups are composed of street view images of the same urban environment type; Perform signal preprocessing and segmentation cutting on the electroencephalogram signals, and calculate the power spectral density to obtain the relative power of the full-time electroencephalogram signals, including: import EEG information, filter, remove artifact signals, and perform independent component analysis; the segmentation cutting divides the electroencephalogram signals into single-image presentation segments and multi-image sequence segments in the time domain; the power spectral density obtains the relative power of the electroencephalogram signals in the single-image presentation segment and the multi-image sequence segment through fast Fourier transform; Use the relative power of the multi-image sequence segment electroencephalogram signals and the emotional evaluation reaction indicators to construct an electroencephalogram signal emotion prediction model to obtain the emotional reaction of any street view image, including: apply the electroencephalogram signal emotion prediction model to the electroencephalogram signals in the single-image presentation segment to obtain the emotional reaction of each street view image; moreover, the electroencephalogram signal emotion prediction model is constructed using the random forest algorithm; Construct a machine learning model based on the emotional response of the street view image and the street view environment index to obtain a street view image emotion prediction model based on electroencephalogram (EEG) signal recognition; moreover, the machine learning model is constructed using the gradient boosting tree algorithm; According to the street view image to be recognized, pass it through the street view image emotion prediction model, use emotion prediction to obtain the scalar of different emotion response indexes corresponding to each street view image, and use ArcgisPro to plot it on the map to obtain the urban street view visual emotion map.

10. An application based on the method as claimed in claim 9, characterized in that, Including: When initializing the EEG signal emotion prediction model, automatically load the preprocessor and the trained gradient boosting tree model; During prediction, perform numerical type verification and missing value processing on the input data, perform robust standardization based on MinCovDet, and then call the independent models of each target variable for parallel prediction; Batch process the street view environment indexes of all street view images in the research area, calculate the mean value of the street view environment indexes in four directions for each coordinate point as the feature data of the coordinate point, and obtain the predicted values of 14 emotion response indexes corresponding to each street view image; Import the CSV file containing emotion indexes into ArcGIS Pro, set the longitude and latitude fields to generate a temporary point layer, and export it as a formal point feature class; Use the spatial join tool to set the road network data as the target feature, the emotion point data as the join feature, set the matching method, and retain all emotion index fields in the field mapping; After completing the data connection, select the rendering method in the symbol system, specify the field as the emotion index; and generate an emotion raster surface, identify statistically significant emotion aggregation areas, and the results are represented by a gradient color to indicate high / low value clustering sections, obtaining the urban street view visual emotion map of the research area.

Images