State analysis and identification method and system

Abstract

The invention relates to the technical field of information, in particular to a state analysis and recognition method, which comprises the following steps: responding to a mood recognition instruction, and receiving a free creation map uploaded by an acquisition end, the free creation map comprising a graffiti drawing map, a tree drawing drawing map and a sand table creation map; obtaining a normalized mapping graph based on the free creation mapping graph, calling a feature recognition model to perform feature recognition on the normalized mapping graph, and obtaining creation mapping features; obtaining a plurality of mood analysis groups according to the creation mapping features, obtaining creation basic attributes of each mood analysis group, and positioning essential features and fluctuation features corresponding to each creation basic attribute based on a mood feature knowledge graph; the essential features and the fluctuation features are fused, a mood feature fusion result is obtained, a vertical index evaluation model is called to evaluate the mood feature fusion result, and a mood evaluation result is generated; and calling a text generation model to perform statement organization on the mood evaluation result, and generating an evaluation analysis report.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a state analysis and identification method and system. Background Art

[0002] In psychological research and practice, non-verbal expressions such as graffiti, tree painting, and sand painting, as a direct reflection of emotions and psychological states, have a value that cannot be ignored. Graffiti, tree painting, and sand painting are the creative results of the subjects without restrictions or guidance, and they can be analyzed to obtain the emotional and psychological state of the subjects.

[0003] Currently, psychological analysis of graffiti, tree drawing and sand painting is highly dependent on the analyst’s personal experience and subjective judgment, and is easily influenced by the analyst’s personal bias, knowledge background and experience level, resulting in inaccurate analysis of the subject’s emotional and psychological state. Summary of the invention

[0004] The embodiment of the present invention provides a state analysis and recognition method and system, which can automatically analyze graffiti drawings, tree drawings and sand paintings, assist analysts in determining the emotional and psychological state of the subjects, and improve the accuracy of the analysis of the emotional and psychological state of the subjects.

[0005] To solve the above technical problems, a technical solution adopted in an embodiment of the present invention is: providing a state analysis and identification method, including: responding to a mood identification instruction, receiving a free creation mapping map uploaded by a collection end, the free creation mapping map including a graffiti drawing map, a tree drawing map and a sandbox creation map, normalizing the free creation mapping map to obtain a normalized mapping map, calling a feature recognition model to perform feature recognition on the normalized mapping map to obtain a creation mapping feature, obtaining a plurality of mood analysis groups according to the creation mapping feature, obtaining a creation basic attribute of each mood analysis group, locating the essential features and fluctuation features corresponding to each creation basic attribute based on a mood feature knowledge graph, fusing the essential features and the fluctuation features to obtain a mood feature fusion result, calling a vertical indicator evaluation model to evaluate the mood feature fusion result to generate a mood evaluation result, calling a text generation model to organize the mood evaluation result into sentences, and generating an evaluation analysis report.

[0006] Optionally, before calling a feature recognition model to perform feature recognition on the normalized mapping image and obtaining the creative mapping feature, the method includes: receiving a creative image sample set, removing redundancy from the creative image sample set based on preset screening rules to obtain an image screening sample set, normalizing the image screening sample set to obtain a normalized sample set, receiving creative feature annotations of each image in the normalized sample set, effectively cropping the corresponding images according to each creative feature annotation to obtain a cropped image set, calling a data enhancement strategy to process the cropped image set to generate an extended image set, training a network building model according to the extended image set to obtain a network training model, testing and verifying the network training model, and obtaining feedback adjustment data based on the test and verification, the feedback adjustment data including model convergence data and recognition accuracy data, optimizing parameters of the network training model according to the model convergence data and the recognition accuracy data to obtain a feature recognition model.

[0007] Optionally, the feature recognition model, the vertical indicator evaluation model and the text generation model are distributedly deployed on several servers, and the method includes: receiving stage processing data, identifying the stage processing data according to preset category judgment rules to obtain stage data categories, the stage data categories including normalized mapping graphs, mood feature fusion results and mood evaluation results, determining the server corresponding to the feature recognition model as a feature recognition server according to the normalized mapping graph, calling the feature recognition server to perform feature recognition on the normalized mapping graph to obtain creation mapping features, determining the server corresponding to the vertical indicator evaluation model as an indicator evaluation server according to the mood feature fusion result, calling the indicator evaluation server to evaluate the mood feature fusion result to obtain a mood evaluation result, determining the server corresponding to the text generation model as a text generation server according to the mood evaluation result, calling the text generation server to organize the mood evaluation result into sentences, and generating an evaluation analysis report.

[0008] Optionally, the graffiti pixel occupancy sequence includes a color pixel occupancy sequence, a line pixel occupancy sequence and a shape pixel occupancy sequence, and the highest pixel occupancy value is obtained based on the graffiti pixel occupancy sequence, and the visual dominant parameter corresponding to each graffiti mood analysis group is determined based on the highest pixel occupancy value, including: obtaining the highest color occupancy value based on the color pixel occupancy sequence, and obtaining a color main tone based on the highest color occupancy value, and the visual main tone includes a bright tone, a neutral tone and a dim tone; determining the bright tone or the neutral tone or the dim tone as the visual dominant parameter corresponding to the color analysis group; obtaining the highest line occupancy value based on the line pixel occupancy sequence, and obtaining a line main form based on the highest line occupancy value, and the line main form includes a straight form and a curved form; determining the straight form or the curved form as the visual dominant parameter corresponding to the line analysis group; obtaining the highest shape occupancy value based on the shape pixel occupancy sequence, and obtaining a shape main characteristic based on the highest shape occupancy value, and the shape main characteristic includes a sharp characteristic and a smooth characteristic; determining the sharp characteristic or the smooth characteristic as the visual dominant parameter corresponding to the shape analysis group.

[0009] Optionally, the creative mapping feature includes a tree painting feature, and multiple mood analysis groups are obtained according to the creative mapping feature, and the creative basic attributes of each mood analysis group are obtained. The essential characteristics and fluctuation characteristics corresponding to each creative basic attribute are located based on the mood feature knowledge graph, including: grouping the tree painting features according to the feature clustering grouping strategy to obtain multiple tree painting mood analysis groups, and the tree painting mood analysis groups include a crown analysis group, a branch analysis group and a root analysis group; obtaining edge detection data corresponding to each tree painting mood analysis group, and extracting corresponding element contour pixels based on each edge detection data; analyzing the element contour pixels to obtain corresponding visual layout parameters, and determining the visual layout parameters as the creative basic attributes corresponding to the tree painting mood analysis group; locating the corresponding essential characteristics and fluctuation characteristics in the mood feature knowledge graph according to the creative basic attributes corresponding to the tree painting mood analysis group.

[0010] Optionally, the element contour pixels include leaf contour pixels, branch contour pixels and root contour pixels, and the element contour pixels are analyzed to obtain corresponding visual layout parameters, including: locating the crown segmentation area based on the leaf contour pixels, performing regional segmentation on the crown segmentation area, obtaining multiple crown segmentation blocks, counting the leaf contour pixels of each crown segmentation block, obtaining the number of contour pixels corresponding to each crown segmentation block, calculating the average of the number of contour pixels corresponding to each crown segmentation block, obtaining the leaf layout density, determining the leaf layout density as the visual layout parameter corresponding to the crown analysis group, determining the outer node pixels of each branch contour based on the branch contour pixels, obtaining the vertical distance between each outer node pixel and the adjacent branch contour, and averaging multiple vertical distances of adjacent branch pairs. The vertical distribution distance of each pair of adjacent branches is calculated, the vertical distribution distance of each pair of adjacent branches is compared with the preset distribution range, the branch distribution characteristics are determined, the branch distribution characteristics are determined as the visual layout parameters corresponding to the branch analysis group, the root contour pixels are identified, the upper root contour pixels and the lower root contour pixels corresponding to each root contour are obtained, a plurality of inner node pixels are determined from the upper root contour pixels and the lower root contour pixels, the vertical distance from each inner node pixel of the current root contour to the upper root contour or the lower root contour is obtained, the average value of the width of the current root contour is determined based on the plurality of vertical distances, the root thickness characteristics are determined according to the plurality of width average values, and the root thickness characteristics are determined as the visual layout parameters corresponding to the root analysis group. .

[0011] Optionally, the creation mapping feature includes a sandbox feature, and multiple mood analysis groups are obtained according to the creation mapping feature, and the creative basic attributes of each mood analysis group are obtained. The essential characteristics and fluctuation characteristics corresponding to each creative basic attribute are located based on the mood feature knowledge graph, including: grouping the sandbox features according to the feature clustering grouping strategy to obtain multiple sandbox mood analysis groups, and the sandbox mood analysis groups include a mold analysis group, a component analysis group and an ornament analysis group; obtaining the component decision category corresponding to each sandbox mood analysis group, and determining the creative basic attributes of the corresponding sandbox mood analysis group based on the component decision category; locating the essential characteristics and fluctuation characteristics corresponding to the creative basic attributes of each sandbox mood analysis group based on the mood feature knowledge graph.

[0012] Optionally, the component decision category corresponding to each sandbox mood analysis group is obtained, and the basic creative attributes of the corresponding sandbox mood analysis group are determined based on the component decision category, including: obtaining the component decision category corresponding to the mold analysis group, the component decision category including terrain molds and architectural molds; determining the terrain molds and / or the architectural molds as the basic creative attributes of the mold analysis group; obtaining the component decision category corresponding to the component analysis group, the component decision category including traffic components, vegetation components and water components; determining the traffic components and / or the vegetation components and / or the water components as the basic creative attributes of the component analysis group; obtaining the component decision category corresponding to the ornament analysis group, the component decision category including human ornaments, animal ornaments and travel tool ornaments; determining the human ornaments and / or the animal ornaments and / or the travel tool ornaments as the basic creative attributes of the ornament analysis group.

[0013] To solve the above technical problems, another technical solution adopted in the embodiment of the present invention is: to provide a state analysis and identification system, including: a data receiving module, used to respond to a mood identification instruction, receive a free creation mapping map uploaded by a collection end, the free creation mapping map includes a graffiti drawing map, a tree drawing drawing and a sandbox creation map, a feature recognition module, used to normalize the free creation mapping map, obtain a normalized mapping map, call a feature recognition model to perform feature recognition on the normalized mapping map, and obtain a creation mapping feature, an attribute acquisition module, used to obtain multiple mood analysis groups according to the creation mapping feature, obtain the creation basic attributes of each mood analysis group, locate the essential features and fluctuation features corresponding to each creation basic attribute based on the mood feature knowledge graph, a result generation module, used to fuse the essential features and the fluctuation features, obtain a mood feature fusion result, call a vertical indicator evaluation model to evaluate the mood feature fusion result, and generate a mood evaluation result, and a report generation module, used to call a text generation model to organize the mood evaluation result into sentences, and generate an evaluation analysis report.

[0014] Different from the prior art, the present invention provides a state analysis and recognition method, including: responding to a mood recognition instruction, receiving a free creation mapping diagram uploaded by a collection end, wherein the free creation mapping diagram includes a graffiti drawing diagram, a tree drawing drawing diagram and a sand table creation diagram; normalizing the free creation mapping diagram to obtain a normalized mapping diagram, calling a feature recognition model to perform feature recognition on the normalized mapping diagram to obtain a creation mapping feature; obtaining multiple mood analysis groups according to the creation mapping feature, obtaining the creation basic attributes of each mood analysis group, locating the essential features and fluctuation features corresponding to each creation basic attribute based on the mood feature knowledge graph; fusing the essential features and the fluctuation features to obtain a mood feature fusion result, calling a vertical indicator evaluation model to evaluate the mood feature fusion result, generating a mood evaluation result, calling a text generation model to organize the mood evaluation result into sentences, and generating an evaluation analysis report. The present invention can automatically analyze graffiti drawings, tree drawing drawings and sand painting creation diagrams, assist analysts in determining the emotional and psychological state of the subjects, and improve the accuracy of the analysis of the emotional and psychological state of the subjects. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] One or more embodiments are exemplarily described by corresponding drawings, which do not constitute limitations on the embodiments. Elements with the same reference numerals in the drawings represent similar elements, and the figures in the drawings do not constitute proportional limitations unless otherwise stated.

[0016] Figure 1 It is a flow chart of a state analysis and identification method provided in an embodiment of the present application; Figure 2 is a schematic diagram of a graffiti normalization mapping diagram provided in an embodiment of the present application; Figure 3 It is a schematic diagram of a tree normalization mapping diagram provided in an embodiment of the present application; Figure 4 It is a structural diagram of a state analysis and identification system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0018] It should be noted that, if there is no conflict, the various features in the embodiments of the present invention can be combined with each other, and all are within the protection scope of the present invention. In addition, although the functional modules are divided in the device schematic diagram and the logical order is shown in the flow chart, in some cases, the steps shown or described can be performed in a different order from the module division in the device schematic diagram or the order in the flow chart.

[0019] The terms "first", "second", "third", "fourth", etc. (if any) in the description and claims of the present invention and the above drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in sequences other than those illustrated or described herein.

[0020] It should be understood that in the present invention, "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or apparatus comprising a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products or apparatuses.

[0021] Unless otherwise defined, all technical and scientific terms used in this specification have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention.

[0022] See also Figure 1 , Figure 1 It is a flow chart of a state analysis and identification method provided in an embodiment of the present application.

[0023] S1. Respond to the mood recognition instruction and receive the free creation mapping diagram uploaded by the acquisition end, wherein the free creation mapping diagram includes a graffiti drawing diagram, a tree drawing diagram and a sandbox creation diagram.

[0024] Among them, the mood recognition instruction represents the instruction that triggers the system to start the mood analysis process, which can be triggered by user operation or automatic trigger (time interval); the acquisition end represents a device that can collect the freely created works of the subjects, and the device is equipped with a camera, such as a mobile phone, computer, etc.; the free creation mapping diagram represents the visual works freely created by the subjects without restrictions and guidance; the graffiti drawing diagram represents the lines and shapes formed by the subjects freely choosing any color pen in an unrestricted and unguided state, and freely painting on white paper, and taking a picture; the tree drawing diagram represents the lines and shapes of the trees formed by the subjects freely painting on white paper in an unrestricted and unguided state, and taking a picture; the sand table creation diagram represents the pictures obtained by the subjects freely placing freely selected molds, components, ornaments, etc. on the sand table in an unrestricted and unguided state, and taking a picture.

[0025] Specifically, after receiving the mood recognition instruction initiated by the user or automatically triggered by the trigger, the system uses the network protocol to establish a communication channel with the collection end such as a tablet computer, mobile phone, etc., and receives the free creation map uploaded by the user through the collection end based on the communication channel. The free creation map uploaded by the collection end may include graffiti drawings, tree drawings, and sandbox creation drawings. The system needs to classify these images and store them using cloud storage, database, etc. to ensure that the image data can be quickly read and processed. It should be noted that the system can receive free creation maps in various image file formats such as JPEG, PNG, BMP, etc., and can distinguish the types of free creation maps through metadata such as the file name and tag of the image file or user selection, and verify the integrity and format correctness of each image by category. If the image is incomplete or the format is incorrect, the user can be prompted to re-upload to ensure the quality of the image uploaded by the user. The user can be a psychoanalyst.

[0026] It is worth mentioning that free-creation mapping provides a non-verbal and natural way for subjects to unconsciously express their innermost emotions and psychological states. Compared with direct questionnaires or text descriptions, the image creation method can better capture potential psychological fluctuations and emotional changes. By obtaining free-creation mapping and analyzing it, more accurate and in-depth emotion recognition can be achieved, and it can provide an important basis for mental health management.

[0027] S2. Normalize the free creation mapping image to obtain a normalized mapping image, call a feature recognition model to perform feature recognition on the normalized mapping image, and obtain a creation mapping feature.

[0028] Among them, the normalization processing means unifying data of different sources, formats, or scales into a standardized format or range for subsequent analysis; the normalized mapping diagram represents the data representation form after adjustment and processing under a unified standard. When the free creation mapping diagram is a graffiti painting diagram, a graffiti normalized mapping diagram is obtained after adjustment and processing; when the free creation mapping diagram is a tree painting diagram, a tree normalized mapping diagram is obtained after adjustment and processing; when the free creation mapping diagram is a sandbox creation diagram, a sandbox normalized mapping diagram is obtained after adjustment and processing; the feature recognition model represents an AI model (basic model) that has been trained and verified, which is specifically used to analyze and identify visual features related to emotional and psychological states in images; feature recognition represents the process of extracting visual features (useful features) related to emotional and psychological states from the input image, and these features can represent important information in the data; the creation mapping feature represents all visual features related to emotional and psychological states extracted from the image.

[0029] Specifically, after obtaining the normalized mapping map, the pre-trained feature recognition model is loaded, the normalized mapping map is input into the feature recognition model for feature recognition, the image features output by the feature recognition model, that is, the visual features, are obtained, all the output visual features are determined as the creative mapping features corresponding to the normalized mapping map and the creative mapping features are stored in the database for subsequent sentiment analysis and evaluation.

[0030] It should be noted that due to the collection of graffiti drawings, tree drawings and sand table creation Figure 1 Generally, the images are collected by taking photos with mobile phones, computers and other devices. Different images have different resolutions and sizes, as well as shooting angles, brightness, exposure, etc. To ensure the accuracy of recognition and eliminate the interference of other irrelevant factors, the collected images must first be converted into a standardized format or range for subsequent analysis. The specific normalization preprocessing means used is not limited in this embodiment, and can be adaptively adjusted according to the shooting environment and shooting equipment in the actual application scenario, such as straightening, color adjustment, denoising, resolution enhancement, alignment, perspective transformation, color normalization, color channel separation and other processing of the collected images.

[0031] It is worth noting that through feature recognition, complex image data can be converted into low-dimensional image features (creation mapping features). The extracted image features can be directly used for emotional state analysis without lengthy calculations of the entire image, which improves processing efficiency. In addition, each subject has different ways of expressing emotions, and obtaining the corresponding creation mapping features helps to achieve personalized emotional analysis.

[0032] In some embodiments, before calling a feature recognition model to perform feature recognition on the normalized map and obtaining the creation mapping features, the method includes steps A1-A5: A1. Receive a creative image sample set, remove redundancy from the creative image sample set based on a preset screening rule to obtain an image screening sample set, and normalize the image screening sample set to obtain a normalized sample set.

[0033] Among them, the creative image sample set represents a group of image samples collected for training the constructed basic model to obtain a feature recognition model, including a graffiti sample set, a tree painting sample set, or a sandbox sample set; the preset screening rule represents the standard or condition for screening the creative image sample set; redundancy removal represents deleting repeated and useless image data from the creative image sample set; the image screening sample set represents a set of images after screening, which is a streamlined data set for training the constructed basic model to obtain a feature recognition model, including a graffiti screening sample set, a tree painting screening sample set, and a sandbox screening sample set. The graffiti sample set is screened to obtain a graffiti screening sample set, the tree painting sample set is screened to obtain a tree painting screening sample set, and the sandbox sample set is screened to obtain a sandbox screening sample set; the normalized sample set is a data set obtained after normalizing the image screening sample set. The images in the normalized sample set have been standardized in terms of size, color, etc., that is, converted into a consistent format, which can be convenient for subsequent model processing.

[0034] Specifically, the collected creative images may contain images of different styles, qualities and sizes. After obtaining the creative images, these creative images are screened by preset screening conditions, such as setting a filtering benchmark to remove blurred images, overly similar images, etc., to obtain a concise and effective image data set, which is the image screening sample set. The images retained in the image screening sample set are standardized, including resizing, color space conversion, etc., to obtain a normalized image set, namely a normalized sample set. The normalized sample set includes a graffiti normalized sample set, a tree painting normalized sample set and a sand table normalized sample set. The graffiti screening sample set is normalized to obtain a graffiti normalized sample set, the tree painting screening sample set is normalized to obtain a tree painting normalized sample set, and the sand table screening sample set is normalized to obtain a sand table normalized sample set. The normalized sample set can be used for subsequent training of the constructed basic model.

[0035] For example, the system collects 1,000 graffiti images to form a graffiti sample set, of which 100 images are too blurry and 150 images are repeated. Through screening and redundancy removal, 750 graffiti images are retained as the graffiti screening sample set. These 750 graffiti images are straightened, toned, denoised, resolution enhanced, aligned, perspective transformed, color normalized, color channel separated, etc. according to a pre-set standardized format or range to obtain a normalized image set, namely the graffiti normalized sample set.

[0036] Among them, in order to improve the accuracy of graffiti recognition, the graffiti sample set is collected at the painting scene under a natural background as much as possible to lay the foundation for the actual promotion and application in the later stage. The graffiti sample set tries to select graffiti painting images of people with different personality environments who are mature in graffiti painting. For example, graffiti painting images of more than 30 branches in six categories, such as students over 12 years old, company employees, enterprise and institution staff, government workers, depression patients, and suicide notes, can be selected. There is no limit on the number of images. In order to improve the accuracy of subsequent model recognition, the number of sample sets is as large as possible. For example, 9.1 million samples can be collected as graffiti sample sets, and 80% of the data in each category can be used to build a training set, and the remaining 20% ​​of the data constitutes a test set.

[0037] It should be noted that screening and normalizing the received creative image samples can reduce invalid or redundant data and ensure that the data set used for training is representative, thereby improving the training efficiency and accuracy of subsequent training models.

[0038] A2. Receive the creative feature annotations of each image in the normalized sample set, and effectively crop the corresponding images according to each creative feature annotation to obtain a cropped image set.

[0039] The creative feature annotation refers to the annotation of specific parts or features of the image in the normalized sample set; the cropped image set refers to a cropped image set, which contains image parts related to specific emotions or psychological states.

[0040] Specifically, the normalized image is annotated, such as marking key elements in the image, such as a certain shape, color area, etc., and the key parts of the image are cropped out according to the annotation information, and irrelevant background or redundant information is removed, so as to obtain a cropped image set, that is, a cropped image set.

[0041] For example, there is a sand table sample, in which the sand and toy car areas are marked as the key parts of sentiment analysis. These areas are cropped to keep the sand and toy car and remove irrelevant parts of the background, such as the edge of the sand table.

[0042] It should be noted that in order to train the constructed basic model to obtain a feature recognition model, labels can be added to the key features of the images in the normalized sample set manually (such as the LabelImg tool). For example, 39 categories of personality recognition, 237 graffiti features, and 13 categories of 39 graffiti features of emotion recognition can be labeled respectively to generate a set of labeled files to improve the effectiveness of image processing and ensure that the model only focuses on the parts of the image related to the emotional and psychological state, avoiding processing irrelevant information.

[0043] A3. Calling a data enhancement strategy to process the cropped image set to generate an extended image set, and training a network building model according to the extended image set to obtain a network training model.

[0044] Among them, the data enhancement strategy refers to a technology that performs multiple transformations on the cropped image set to generate more samples, such as rotation, flipping, scaling, adding noise, etc.; the extended image set refers to a collection of multiple variant data of the original images in the cropped image set; the network building model, that is, the basic model built in advance, represents a deep learning model that can perform feature recognition, such as a convolutional neural network; the network training model refers to a trained basic model that can recognize and classify emotional psychological characteristics in images.

[0045] Specifically, there is a painting image, which is rotated 90 degrees, 180 degrees and mirror-flipped to generate 4 variant images. A convolutional neural network (CNN) is trained based on these images to let the model learn how to recognize emotional psychological characteristics in images (such as color changes or shape features).

[0046] It should be noted that by training the basic model built in advance, it can be enabled to recognize and extract features in the image, and the cleaned images are subjected to data enhancement processing, which significantly increases the number of images of each type (reaching more than 100,000), expanding the diversity of training data to ensure the model training effect when sample images are limited.

[0047] A4. Test and verify the network training model, and obtain feedback adjustment data based on the test and verification, wherein the feedback adjustment data includes model convergence data and recognition accuracy data.

[0048] Among them, the feedback adjustment data is the feedback information for adjusting the model based on the test verification results; the model convergence data represents an important indicator for measuring whether the model has successfully converged, that is, whether the model has effectively learned the characteristics of the data during the training process; the recognition accuracy data represents the accuracy of the model in the feature recognition task.

[0049] Specifically, the test set can be used to evaluate the trained basic model. Specifically, the training set and test set of the experiment can be constructed according to the ratio of 4:1 to perform model recognition training. The construction ratio can be adjusted according to actual conditions, and the present invention does not make specific limitations here.

[0050] A5. Optimize parameters of the network training model according to the model convergence data and the recognition accuracy data to obtain a feature recognition model.

[0051] The feature recognition model represents a mature model obtained after training, testing and optimization steps, which can stably extract relevant features of emotional psychological states from images.

[0052] It should be noted that parameter optimization means adjusting the performance parameters of the network training model, which can improve its performance in feature recognition tasks and make it more robust and accurate.

[0053] Taking the network construction model as the VGG-16 network structure as an example, after obtaining the training set and the test set, the specific implementation method of the model recognition training is specifically introduced. The model processing and training process under other model structures can refer to the introduction of this embodiment, which will not be repeated here.

[0054] Convolutional neural network is a multi-level non-fully connected neural network that simulates the structure of the human brain. It has powerful feature extraction capabilities. Through supervised multi-layer network learning, it can directly recognize and understand the image content from the original image. VGG-16 is one of the classic convolutional neural network models. This network model has 13 convolutional layers, 4 pooling layers and 3 fully connected layers, a total of 16 layers. It has the characteristics of local connection, weight sharing, pooling, etc. The input image size changes from 224x224 to 112x112, etc., until it finally becomes 7x7. Therefore, VGG-16 has a simple and beautiful structure, clear layers, and is also very regular in implementation. The biggest feature of VGG-16 is that it can extract small features contained in the input image through the combination and stacking of 3×3 filters.

[0055] The input image size is 224×224 with 3 channels. VGG-16 is used to build the intelligent recognition model. The entire network model is divided into convolution layer, pooling layer, fully connected layer and classification layer. The convolution layer has 5 convolution segments, and each convolution segment uses 2 or 3 convolution base layers to form a convolution sequence. The "SAME" mode is used in the convolution process, which will not change the resolution of the feature map. A 2×2 pooling window with a step size of 2 is used in the pooling layer to reduce the size of the feature image after convolution, and also realize the translation invariance of the model. The fully connected layer is combined by 3 consecutive full connections, and the number of channels is 4096, 4096, and 1000 respectively. Finally, the classification layer is classified and output by a SoftMax classifier with 1000 labels.

[0056] The graffiti feature recognition model can be completed on a server with 4 NVIDIA Tesla P40 24 GB cards and 2 Intel Xeon E5-2697v4 processors with 18 cores and 36 threads, and the server is equipped with an Ubuntu 16.0 operating system and a PaddlePaddle 1.8.4 framework. This embodiment only takes the operating environment of the graffiti feature recognition model as an example for introduction. The configurations under other operating environments can refer to the introduction of this embodiment and will not be repeated here.

[0057] The graffiti feature recognition model starts the model training program on a system equipped with four NVIDIA Tesla Ps. The training process uses weakly supervised convolutional neural network technology and applies the VGG convolutional neural network model to realize the recognition of graffiti detail features. The convolutional neural network is a multi-level non-fully connected neural network that simulates the structure of the human brain. It has a strong feature extraction capability and can directly recognize and understand the image content from the graffiti image through supervised multi-layer network learning. VGG-16 is a classic convolutional neural network model, which undergoes 13 convolutional layers, 4 pooling layers, and 3 fully connected layers, for a total of 16 layers. Through local connections, weight sharing, pooling and other processes, the input graffiti image size is reduced from 224x224 to 112x112, until it finally becomes 7x7. The entire network model is divided into convolution layer, pooling layer, fully connected layer and classification layer. The convolution layer has 5 convolution segments. Each convolution segment uses 2 or 3 convolution layers to form a convolution sequence. The "SAME" mode is used in the convolution process, which will not change the resolution of the feature map. In the pooling layer, a 2×2 pooling window with a step size of 2 is used to reduce the size of the feature image after convolution, and also realize the translation invariance of the model. In the fully connected layer, 3 consecutive full connections are combined, and the number of channels is 4096, 4096, and 1000 respectively. Finally, in the classification layer, a SoftMax classifier with 1000 labels is used for classification output. The structure is simple, the hierarchy is clear, and the process is regular. VGG-16 extracts the fine features contained in the input graffiti image through the combination and stacking of 3×3 filters, thereby obtaining a graffiti feature recognition model.

[0058] It should be noted that, in this embodiment, only the training process of the graffiti feature recognition model is introduced. The training processes of other machine learning models, such as the vertical indicator evaluation model and the text generation model, can refer to the introduction of this embodiment. For example, when the vertical indicator evaluation model is specifically an emotional state evaluation model, the personality traits and corresponding graffiti features of a number of students (for example, 5,100) from the second grade of junior high school, the third grade of junior high school, and the first grade of senior high school to the third grade of senior high school with different levels can be selected, and the corresponding evaluation model can be obtained by training based on the degree of influence of the personality traits; for example, when the vertical indicator evaluation model is specifically a negative emotion recognition model, a number of graffiti features (for example, 1,600) related to negative emotions can be selected through case verification and captured Internet big data analysis, and combined with clinical experience and empirical data, the single-dimensional graffiti feature model can be integrated for testing and parameter adjustment training.

[0059] In some embodiments, the feature recognition model, the vertical indicator evaluation model and the text generation model are distributed and deployed on a plurality of servers, and the method includes steps B1-B4: B1. Receive stage processing data, identify the stage processing data according to preset category judgment rules, and obtain stage data categories, wherein the stage data categories include normalized mapping diagrams, mood feature fusion results, and mood assessment results.

[0060] Among them, the vertical indicator evaluation model represents a model for realizing the evaluation of specific indicators in a vertical field, and the specific indicators belong to the performance tendency or ability specialty range of the user; the text generation model represents a natural language generation model, which is responsible for converting the mood evaluation results into an easy-to-understand natural language report; the server represents a computing device used to perform specific tasks; the stage processing data represents the intermediate data generated in different stages; the preset category judgment rules represent the pre-set rules for classifying stage processing data, which help the system to determine which category the data belongs to, such as judging by the data format, etc.; the stage data category represents the specific data type after classification; the normalized mapping diagram represents the standardized image data input by the system, which is the basis of feature recognition; the mood feature fusion result represents the comprehensive emotion data generated by integrating multiple emotion features; the mood evaluation result represents the specific evaluation result of the user's psychological state, such as the user's anxiety, depression, pleasure and other states.

[0061] Specifically, distributed deployment means that different models are deployed on different servers, each responsible for processing different types of data, rather than concentrating all functions on one server. The data category in the acquisition stage can determine the server and corresponding model to be called subsequently, ensuring that the data is processed correctly.

[0062] B2. According to the normalized mapping graph, the server corresponding to the feature recognition model is determined as the feature recognition server, and the feature recognition server is called to perform feature recognition on the normalized mapping graph to obtain the creation mapping feature.

[0063] Among them, the feature recognition server refers to a server deployed with a feature recognition model, which is specially used to process image data and can extract key features in the image, such as color, shape, etc.; feature recognition refers to the process of extracting useful features from the image, such as color distribution, etc.; the creative mapping feature refers to all visual features extracted from the normalized mapping diagram, which contains the key information of the image and is the basic data for sentiment analysis.

[0064] Specifically, the system recognizes that the stage processing data is a normalized mapping map. According to this category, the normalized mapping map is sent to the server where the feature recognition model is deployed, that is, the feature recognition model is called to extract image features such as color and shape.

[0065] B3. According to the mood feature fusion result, the server corresponding to the vertical indicator evaluation model is determined as the indicator evaluation server, and the indicator evaluation server is called to evaluate the mood feature fusion result to obtain a mood evaluation result.

[0066] Among them, the indicator evaluation server represents a server on which a vertical indicator evaluation model is deployed; the evaluation represents a quantitative or qualitative analysis of emotional data; the mood evaluation result represents a specific evaluation result of the user's psychological state, such as the user's anxiety, depression, pleasure, etc.

[0067] Specifically, the system recognizes that the stage processing data is the result of mood feature fusion. According to this category, the mood feature fusion result is sent to the server where the vertical indicator evaluation model is deployed, that is, the call to the vertical indicator evaluation model is realized to analyze the mood feature fusion result and generate a specific evaluation result of the user's psychological state.

[0068] B4. According to the mood assessment result, the server corresponding to the text generation model is determined as the text generation server, and the text generation server is called to organize the mood assessment result into sentences to generate an assessment analysis report.

[0069] Among them, the text generation server refers to a server that deploys a text generation model to help users understand the mood assessment results; sentence organization refers to converting data descriptions into fluent and natural language descriptions; the assessment analysis report refers to the final report output by the system, which contains a detailed analysis and suggestions on the emotional and psychological state of the subject.

[0070] Specifically, the system recognizes that the stage processing data is a mood assessment result, and sends the mood assessment result to a server where a text generation model is deployed, that is, the text generation model is called to convert the assessment result into a natural language description.

[0071] It should be noted that the present invention needs to call more than one learning model, and needs to call at least three models including a feature recognition model, a vertical index evaluation model, and a text generation model. The learning model consumes a large amount of system resources. To ensure the rapid response of the system, each model can be deployed in a distributed manner and concurrently scheduled. At the same time, the system realizes flexible resource scheduling and system load balancing at the software level. For example, while the feature recognition server performs feature recognition on image c, the index evaluation server can perform a corresponding emotion evaluation on image b when image b has called the feature recognition server. The text generation server can generate a text report corresponding to image a when image a has called the feature recognition server and the index evaluation server respectively. Therefore, multiple tasks can be processed at the same time, waiting time can be reduced, and the system response speed can be improved. In addition, the distributed deployment means that if a server fails, other parts of the system can still continue to work. This redundant design enhances the fault tolerance and stability of the system.

[0072] S3. Obtain multiple mood analysis groups according to the creation mapping features, obtain the creation basic attributes of each mood analysis group, and locate the essential characteristics and fluctuation characteristics corresponding to each creation basic attribute based on the mood feature knowledge graph.

[0073] Among them, the mood analysis group is the result of grouping the same or similar visual elements into a group. For example, in a graffiti painting, all color features can be grouped into a group, all shape features can be grouped into a group, and all line features can be grouped into a group; the basic creation attributes represent the core attributes extracted from the subject's creative process, which can be used to describe the subject's emotional expression method and reflect certain specific expressions or intentions in the subject's creation; the mood feature knowledge graph represents a knowledge network that contains the mapping relationship between each basic creation attribute and the corresponding emotional feature; the essential feature reflects the individual's relatively stable personality and behavior pattern over a long period of time; the fluctuation feature represents an immediate response to changes in the external environment, reflecting the individual's emotional fluctuations and behavioral changes in a short period of time.

[0074] Specifically, after obtaining the basic creative attributes corresponding to each mood analysis group, the mood feature knowledge graph is called to match each basic creative attribute with its corresponding emotional feature (essential features and fluctuation features) to obtain the relevant essential features and fluctuation features. The mood feature knowledge graph records the association information between different basic attributes and emotional features.

[0075] It should be noted that essential characteristics, also known as personal "traits", refer to a relatively persistent, consistent and stable characteristic of thoughts, emotions and actions that can be expressed in many environments. It is manifested in the behavioral tendencies of a person's personality characteristics. Basic traits include calm personality, self-control, cheerfulness, introversion, smoothness, sense of purpose, rigorous thinking, carefulness, and desire for pursuit. In the process of work, each person's personal characteristics will affect their external performance to varying degrees, or cause them to have tendency behaviors. Fluctuation characteristics are a kind of personal performance under the influence of external environment and affairs. Unlike stable essential personality traits, fluctuation characteristics often change with the changes in external affairs. It is a real-time characteristic, and personal fluctuation characteristics will also affect their external performance to varying degrees, resulting in differences in creative mapping characteristics. The factors that affect a person's performance can be divided into internal factors and external factors. Internal privacy is the essential characteristics, and external factors are the fluctuation characteristics. The two together affect their external free creative mapping map, making the creative mapping characteristics under different essential characteristics and fluctuation characteristics also have differences.

[0076] In some embodiments, the creation mapping feature includes a graffiti feature, and multiple mood analysis groups are obtained according to the creation mapping feature, and the creation basic attributes of each mood analysis group are obtained. The essential features and fluctuation features corresponding to each creation basic attribute are located based on the mood feature knowledge graph, including steps C1-C4: C1. Calling a feature clustering grouping strategy to group the graffiti features to obtain a plurality of graffiti mood analysis groups, wherein the graffiti mood analysis groups include a coloring analysis group, a line analysis group, and a shape analysis group.

[0077] It is worth noting that the feature recognition model includes a graffiti feature recognition model, a painted tree feature recognition model and a sandbox feature recognition model. The graffiti feature recognition model is called to perform feature recognition on the graffiti normalized mapping image to obtain the graffiti features.

[0078] Among them, the graffiti features represent visual elements extracted from graffiti drawings, including but not limited to the colors, lines, shapes, etc. used by the subjects in graffiti; the feature clustering grouping strategy is an algorithm strategy that groups similar features together, usually based on the similarity or correlation of features, and can be specifically K-means clustering, hierarchical clustering, etc.; the graffiti mood analysis group represents a specific emotion analysis group formed according to the graffiti feature grouping, focusing on a certain type of feature in graffiti; the coloring analysis group represents a group specifically used to analyze color features in graffiti; the line analysis group represents a group specifically used to analyze line features in graffiti; the shape analysis group represents a group specifically used to analyze shape features in graffiti.

[0079] It should be noted that the feature clustering grouping strategy can group the graffiti features extracted from the graffiti normalization map by category or similarity, such as grouping all color-related features into one group, all line-related features into one group, and all shape-related features into one group. Clustering grouping helps the system to process and analyze different graffiti features more effectively.

[0080] Example: See Figure 2 , Figure 2 This is a schematic diagram of a graffiti normalization mapping diagram provided by an embodiment of the present application. The color of the house and the chimney is black, the color of the sun is red, and the color of the chimney smoke is light gray. The above color-related features are grouped into a group, which can be named a coloring analysis group; the house includes 7 straight lines, the chimney includes 3 straight lines, the sun includes 8 straight lines and a curve, and the chimney smoke includes 2 curves. Note that when performing line analysis, the circle is truncated and regarded as a curve. The above line-related features are grouped into a group, which can be named a line analysis group; the house and the chimney include 2 rectangles and a trapezoid, the chimney smoke includes two circles, and the sun includes one circle. The above line-related features are grouped into a group, which can be named a shape analysis group.

[0081] C2. Obtain pixel occupancy data corresponding to each graffiti mood analysis group, wherein the pixel occupancy data includes each preset graffiti dimension and a corresponding pixel occupancy value, and sort the multiple pixel occupancy values ​​in descending order to obtain a graffiti pixel occupancy sequence.

[0082] Among them, the pixel occupancy data represents the number of pixels occupied by a specific area or feature in the image, reflecting the distribution and proportion of different visual elements in the image; the preset graffiti analysis dimension represents a predefined angle for classifying and quantifying different visual features in graffiti, such as for the coloring analysis group, the preset graffiti analysis dimension can specifically be bright coloring, neutral coloring and dim coloring; the pixel occupancy value represents the number of pixels occupied corresponding to a specific preset graffiti analysis dimension; the descending order arrangement represents the process of arranging multiple pixel occupancy values ​​from large to small; the graffiti pixel occupancy sequence represents sorting the pixel occupancy values ​​corresponding to each preset graffiti analysis dimension in order from large to small to form a sequence, which can reflect the primary and secondary status or prominence of each visual element in the image.

[0083] For example, for the coloring analysis group, the preset bright coloring corresponds to red, yellow, orange, etc., the neutral coloring corresponds to white, light gray, light blue, etc., and the dark coloring corresponds to black, gray, dark blue, etc. Figure 2The color of the sun (red) is bright coloring, and the corresponding pixel occupancy is 500 pixels. The color of the chimney smoke (light gray) is neutral coloring, and the corresponding pixel occupancy is 100 pixels. The color of the house and chimney (black) is dark coloring, and the corresponding pixel occupancy is 1000 pixels. 500, 100, and 1000 are arranged in descending order, and the graffiti pixel occupancy sequence is 1000, 500, and 100. It should be noted that among all coloring features, different coloring types may contain multiple colors, and the total number of pixels corresponding to each coloring type is calculated separately. The line analysis group and the shape analysis group have similar processing ideas. In the line analysis group, the number of pixels corresponding to the straight line is 1300 pixels, and the number of pixels corresponding to the curve is 400 pixels. The obtained graffiti pixel occupancy sequence is 1300 and 400. It should be noted that among all the graffiti lines, there may be multiple straight lines and curved lines. The total number of pixels corresponding to the straight lines and curved lines are calculated respectively. In the shape analysis group, the number of pixels corresponding to the rectangle is 600 pixels, the number of pixels corresponding to the trapezoid is 400 pixels, and the number of pixels corresponding to the circle is 500 pixels. The obtained graffiti pixel occupancy sequence is 600, 500, and 400. It should be noted that there may be multiple elements corresponding to various shape types. The total number of pixels corresponding to various shape types are calculated respectively.

[0084] C3. Obtaining a maximum pixel occupancy value based on the graffiti pixel occupancy sequence, determining a corresponding visual dominant parameter according to the maximum pixel occupancy value, and determining the visual dominant parameter as a creation basic attribute corresponding to the graffiti mood analysis group.

[0085] Among them, the highest pixel occupancy value represents the value that occupies the most pixels in the graffiti pixel occupancy sequence, that is, the first pixel occupancy value in the graffiti pixel occupancy sequence; the visual dominant parameter represents the visual feature used to describe the dominant position in the graffiti; the creative basic attribute represents the core attribute extracted from the subject's creative process, which can be used to describe the subject's emotional expression method and reflect certain specific expressions or intentions in the subject's creation.

[0086] In some embodiments, the graffiti pixel occupancy sequence includes a colored pixel occupancy sequence, a line pixel occupancy sequence and a shape pixel occupancy sequence, and step C3 (obtaining the highest pixel occupancy value based on the graffiti pixel occupancy sequence, and determining the visual dominant parameter corresponding to each graffiti mood analysis group based on the highest pixel occupancy value) specifically includes steps C31-C36: C31. Obtain a maximum coloring occupancy value based on the coloring pixel occupancy sequence, and obtain a coloring main tone based on the maximum coloring occupancy value, wherein the visual main tone includes a bright tone, a neutral tone, and a dim tone.

[0087] Among them, the colored pixel occupancy sequence represents the result of descending arrangement according to the pixel occupancy of different color tones in the image, reflecting the proportion of various color tones in the image; the line pixel occupancy sequence represents the result of descending arrangement according to the pixel occupancy of different line types in the image, reflecting the proportion of various line types in the image; the shape pixel occupancy sequence represents the result of descending arrangement according to the pixel occupancy of different shape types in the image, reflecting the proportion of various shape types in the image; the highest colored occupancy value represents the first colored occupancy value in the colored pixel occupancy sequence, that is, The maximum coloring occupancy value; the main coloring tone represents the overall color tone of the graffiti drawing, that is, the main color feature of the graffiti drawing, which is comprehensively determined by all color features in the coloring analysis group; the bright tone represents that the graffiti drawing is dominated by bright and vivid colors, such as red, yellow, orange, etc., and the corresponding coloring occupancy value is the highest; the neutral tone represents that the coloring in the graffiti drawing is moderate, not biased towards bright or dull, such as white, light gray, light blue, etc., and the corresponding coloring occupancy value is the highest; the dull tone represents that the graffiti drawing is dominated by dull and low-saturation colors, such as black, dark blue, etc., and the corresponding coloring occupancy value is the highest.

[0088] example: Figure 2 The graffiti pixel occupancy sequence 1000, 500, and 100 corresponding to the coloring analysis group is the coloring pixel occupancy sequence, with dark coloring being 1000 pixels, bright coloring being 500 pixels, and neutral coloring being 100 pixels. The highest coloring occupancy value is 1000, corresponding to dark coloring (black houses and chimneys), so the main coloring tone is a dark tone.

[0089] C32. Determine the bright tone, the neutral tone, or the dark tone as the visually dominant parameter corresponding to the coloring analysis group.

[0090] The visual dominance parameter is used to describe the dominant visual feature in the graffiti, and here it refers to the dominant color tone feature.

[0091] Example: For Figure 2 , the main tone of coloring is a dark tone, and the dark tone can be determined as the visual dominant parameter corresponding to the coloring analysis group.

[0092] C33. Obtain a maximum line occupancy value based on the line pixel occupancy sequence, and obtain a line main body shape based on the maximum line occupancy value, wherein the line main body shape includes a straight shape and a curved shape.

[0093] Among them, the highest line occupancy value represents the line occupancy value corresponding to the line type that occupies the largest number of pixels in the line pixel occupancy sequence; the line main body form represents the overall form of the graffiti line; the straight form represents that the line is mainly presented as a straight line; the curved form represents that the line is mainly presented as a curve.

[0094] example: Figure 2 The graffiti pixel occupancy sequence 1300 and 400 corresponding to the line analysis group is the line pixel occupancy sequence. The highest line occupancy value is 1300, and 1300 corresponds to a straight line, which is a straight shape. Therefore, the main shape of the line is a straight shape.

[0095] C34. Determine the straight shape or the curved shape as the visual dominant parameter corresponding to the line analysis group.

[0096] The visual dominant parameter is used to describe the dominant visual feature in the graffiti, and here refers to the dominant straightness and curvature feature of the line.

[0097] Example: For Figure 2 , the main shape of the line is straight, and the straight shape can be determined as the visual dominant parameter corresponding to the line analysis group.

[0098] C35. Obtain a highest shape occupancy value based on the shape pixel occupancy sequence, and obtain shape main body characteristics based on the highest shape occupancy value, wherein the shape main body characteristics include sharp characteristics and smooth characteristics.

[0099] Among them, the highest shape occupancy value represents the shape occupancy value corresponding to the shape occupying the largest number of pixels in the shape pixel occupancy sequence; the shape main feature represents the main characteristics of the shape in the graffiti drawing; the sharp characteristic represents that the main shape has sharp edges and angles; the smooth characteristic represents that the shape has rounded edges and curves.

[0100] example: Figure 2 The graffiti pixel occupancy sequence 600, 500, 400 corresponding to the shape analysis group is the shape pixel occupancy sequence, and the highest shape occupancy value is 600, which corresponds to a rectangle and is a sharp feature. Therefore, the shape main feature is a sharp feature.

[0101] C36. Determine the sharp characteristic or the smooth characteristic as the visual dominant parameter corresponding to the shape analysis group.

[0102] The visual dominant parameter is used to describe the dominant visual feature in the graffiti, and here refers to the dominant shape sharpness and bluntness characteristics.

[0103] Example: For Figure 2, the main characteristic of the shape is the sharp characteristic, which can be determined as the visual dominant parameter corresponding to the shape analysis group.

[0104] The visual dominant parameters are determined as the basic creation attributes corresponding to the graffiti mood analysis group, that is, for the coloring analysis group, the dark tone is the basic creation attribute, for the line analysis group, the straight shape is the basic creation attribute, and for the shape analysis group, the sharp feature is the basic creation attribute.

[0105] It should be noted that the specific graffiti mood analysis group can be adjusted according to actual conditions and can be set accordingly according to recognition needs. In this embodiment, only the above-mentioned graffiti mood analysis group is introduced as an example. The recognition based on other graffiti images can refer to the introduction of this embodiment and will not be repeated here.

[0106] C4. Calling the mood feature knowledge graph, and locating the corresponding essential features and fluctuation features in the mood feature knowledge graph according to the basic creation attributes corresponding to each graffiti mood analysis group.

[0107] Among them, the basic creation attributes, namely the basic graffiti attributes, represent the main visual features extracted from the graffiti drawing, which reflect the core performance elements in the creation process.

[0108] Example: According to the dark tone corresponding to the coloring analysis group, the straight shape corresponding to the line analysis group, and the sharp characteristics corresponding to the shape analysis group, the mood feature knowledge graph is traversed to find the essential characteristics and fluctuation characteristics corresponding to each graffiti mood analysis group.

[0109] In some embodiments, the creation mapping feature includes a drawing tree feature, and multiple mood analysis groups are obtained according to the creation mapping feature, and the creation basic attributes of each mood analysis group are obtained, including D1-D4: D1. Grouping the tree painting features according to the feature clustering grouping strategy to obtain a plurality of tree painting mood analysis groups, wherein the tree painting mood analysis groups include a crown analysis group, a branch analysis group and a root analysis group.

[0110] Among them, the tree drawing features represent visual elements extracted from the tree drawing pictures, including but not limited to the crown, branches, roots, etc. drawn by the subjects when drawing the trees; the tree drawing mood analysis group represents a specific emotion analysis group formed according to the grouping of tree drawing features, focusing on a certain type of features in the tree drawings; the crown analysis group represents a group specifically used to analyze the crown features in the tree drawings; the trunk analysis group represents a group specifically used to analyze the branch features in the tree drawings; the root analysis group represents a group specifically used to analyze the root features in the tree drawings.

[0111] It should be noted that the feature clustering grouping strategy can group the painting tree features extracted from the painting tree normalization mapping diagram according to category or similarity. By grouping, the painting tree features can be classified into different mood analysis groups, which can provide a basis for subsequent analysis and enable the system to more effectively process and analyze different painting tree features.

[0112] Example: See Figure 3 , Figure 3 is a schematic diagram of a tree normalization mapping diagram provided in an embodiment of the present application, Figure 3 Middle-a is the crown analysis group, Figure 3 Middle-b is the branch analysis group, Figure 3 Middle-c is the root analysis group, and the dividing line between the trunk and the root can be adjusted according to actual conditions, and this application does not make any specific limitations on this.

[0113] D2. Obtain edge detection data corresponding to each painting tree mood analysis group, and extract corresponding element contour pixels based on each edge detection data.

[0114] Among them, the edge detection data represents all boundary data of the painted tree, that is, the boundary line between the element and the background or other elements; the element contour pixels represent all edge pixels corresponding to the elements of different parts of the painted tree (such as leaves, branches, roots, etc.), that is, all pixel points contained in the object boundary data.

[0115] Specifically, edge detection algorithms (such as Sobel, Canny, etc.) can be used to identify the edges in the tree painting mood analysis group and represent them as a pixel set, which is the element contour pixel. The element contour pixel reflects the shape and structure of the tree painting. For the crown analysis group, the outer boundary of the leaves can be extracted, and the corresponding pixel set is represented as the leaf contour pixel; for the branch analysis group, the outer boundary of the trunk can be extracted, and the corresponding pixel set is represented as the branch contour pixel; for the root analysis group, the outer boundary of the root can be extracted, and the corresponding pixel set is represented as the root contour pixel.

[0116] It should be noted that the shapes and boundaries of different elements in the painting tree are defined through edge detection and contour extraction, which can provide data support for further visual layout analysis.

[0117] D3. Analyze the element outline pixels to obtain corresponding visual layout parameters, and determine the visual layout parameters as the creation basic attributes corresponding to the painting tree mood analysis group.

[0118] Among them, the visual layout parameters are parameters that describe the spatial distribution characteristics of elements in the image, and can be used to analyze the structure and layout of each part in the painting tree; the creative basic attributes represent the core attributes extracted from the subject's creative process, which can be used to describe the subject's emotional expression method and reflect certain specific expressions or intentions in the subject's creation.

[0119] It should be noted that by extracting visual layout parameters and analyzing the distribution and arrangement of different parts of the tree in the image, a foundation can be laid for identifying the creative intention and psychological state of the subjects.

[0120] In some embodiments, the element contour pixels include leaf contour pixels, branch contour pixels, and root contour pixels, and step D3 (analyzing the element contour pixels to obtain corresponding visual layout parameters) specifically includes steps D31-D36: D31. Locate the crown segmentation area based on the leaf contour pixels, perform regional segmentation on the crown segmentation area, obtain multiple crown segmentation blocks, count the leaf contour pixels of each crown segmentation block, and obtain the number of contour pixels corresponding to each crown segmentation block.

[0121] Among them, the leaf contour pixels represent the pixel points corresponding to the contour of the crown part; the branch contour pixels represent the pixel points corresponding to the contours of the branches and trunk parts; the root contour pixels represent the pixel points corresponding to the root contour of the drawn tree; the crown segmentation area is the minimum circumscribed rectangular area that completely covers all leaf contour pixels; the crown segmentation block represents the gridding processing of the crown segmentation area of ​​equal size, and each grid area is a crown segmentation block; the number of contour pixels represents the number of pixels occupied by the leaf contour in each crown segmentation block.

[0122] Specifically, in order to determine the crown segmentation area, the leaf contour pixels can be analyzed to determine the leftmost, rightmost, highest, and lowest pixel points. Based on the above four pixel points, a minimum circumscribed rectangular area containing all leaf contours is constructed, and then the minimum circumscribed rectangular area is evenly divided into multiple small rectangular blocks, and the number of leaf contour pixels contained in each small rectangular block is counted.

[0123] Example: See Figure 3 In 3-a, the minimum circumscribed rectangular area corresponding to the leaf contour pixels is divided into 6 crown segmentation blocks. The number of pixels occupied by the leaf contour in the first crown segmentation block is 100, the number of pixels occupied by the leaf contour in the second crown segmentation block is 200, the number of pixels occupied by the leaf contour in the third crown segmentation block is 100, the number of pixels occupied by the leaf contour in the fourth crown segmentation block is 150, the number of pixels occupied by the leaf contour in the fifth crown segmentation block is 50, and the number of pixels occupied by the leaf contour in the sixth crown segmentation block is 100.

[0124] It should be noted that the total number of pixels in each small rectangular block is a preset value, and an error is allowed within a certain range. The preset value and the error range can be set according to actual conditions, and the present invention does not make specific limitations on this. For example, the preset value of the total number of pixels in each small rectangular block is 300 pixels, and the total number of pixels allowed in each crown segment block is 290 to 310. The number of segmentations needs to be determined according to actual needs to ensure that the total number of pixels in each crown segment block is equal.

[0125] D32, calculating the mean value of the number of outline pixels corresponding to each crown segmentation block to obtain the leaf layout density, and determining the leaf layout density as the visual layout parameter corresponding to the crown analysis group.

[0126] Among them, mean calculation means statistical analysis of multiple contour pixel numbers by averaging the average value; the leaf layout density represents the distribution density of leaves in the segmented area, including low density, moderate density and high density. The leaf layout density is determined by the average value of the number of contour pixels in each crown segmentation block and the preset density range.

[0127] Specifically, the number of contour pixels in each crown segmentation block is averaged to obtain the average calculation result. The average calculation result is compared with the preset density range. If the average calculation result is within the preset density range, the leaf layout density is moderate density; if the average calculation result exceeds the upper limit of the preset density range, the leaf layout density is high density; if the average calculation result is lower than the lower limit of the preset density range, the leaf layout density is low density.

[0128] It should be noted that the preset density range is set based on the preset value of the total number of pixels in each crown segment block and can be adjusted according to actual conditions.

[0129] Example: See Figure 3 In 3-a, the number of contour pixels corresponding to each crown segmentation block is averaged, that is, (100+100+100+100+150+50+100)÷6=100, the preset density range is 40~80, and the mean calculation result 100 is compared with the preset density range 40~80. If 100 is higher than 80, the leaf layout density is judged to be high density, and high density is determined as the visual layout parameter corresponding to the crown analysis group.

[0130] D33, determining the outer node pixels of each branch outline based on the branch outline pixels, obtaining the vertical distance between each outer node pixel and the adjacent branch outline, calculating the average of multiple vertical distances of adjacent branch pairs, and obtaining the vertical distribution distance of each adjacent branch pair.

[0131] The branch outline represents the outer boundary of the branch shape; the outer node pixels represent the outer pixels at the corners, turning points or bifurcations in the branch outline, that is, the outermost key feature pixel points of the branch outline ( Figure 3 Not all are marked in 3-b, and the pixels marked at 1, 2, and 3 are outer node pixels); the adjacent branch contours represent the branch contours that are spatially adjacent to the current branch contour; the vertical distance represents the distance in the vertical direction between an outer node pixel of a branch contour and the adjacent branch contour; the adjacent branch pairs represent two branch contours that are adjacent to each other; the mean calculation represents averaging all vertical distances of a pair of adjacent branch pairs; the vertical distribution distance represents the average vertical spacing between each pair of adjacent branch pairs.

[0132] Specifically, the branch contour pixels include tree branch contour pixels and trunk contour pixels. Key inflection points, turning points and other external node pixels are located in the branch contour pixels. For each pair of adjacent branches, the vertical distance from each external node pixel to the adjacent branch contour is calculated. Each pair of branches may have multiple vertical distances. The average of all vertical distances of a pair of adjacent branches is calculated to obtain the average vertical distance of the pair of adjacent branches. The average vertical distance is the vertical distribution distance, until the vertical distribution distances corresponding to all adjacent branch pairs are obtained.

[0133] Example: See Figure 3 In 3-b, branch outline B and branch outline C are a pair of adjacent branches. Branch outline B has external node pixels 1 and 2 on the side close to branch outline C, and branch outline C has external node pixel 3 on the side close to branch outline B. The vertical distances from external node pixel 1 to branch outline C, external node pixel 2 to branch outline C, and external node pixel 3 to branch outline B are calculated respectively. The three vertical distances are averaged to obtain the vertical distribution distance corresponding to the pair of adjacent branches (branch outline B and branch outline C). The vertical distribution distance corresponding to branch outline A and branch outline B can be calculated with the same idea.

[0134] D34. Compare the vertical distribution distance of each adjacent branch pair with a preset distribution range to determine the branch distribution characteristics, and determine the branch distribution characteristics as the visual layout parameters corresponding to the branch analysis group.

[0135] Among them, the preset distribution range represents a standard reference range for judging the uniformity of branch distribution, which can be specifically composed of half of the width of the trunk as the reference point and a certain distance (such as 5 pixels) extending up and down. It can be set according to actual conditions, and the present invention does not impose specific restrictions on this; the branch distribution characteristics represent the uniform distribution state of the branches, including uniform distribution and uneven distribution.

[0136] Specifically, the vertical distribution distance of adjacent branch pairs is compared with a preset distribution range. If the vertical distribution distance is within the preset distribution range, the adjacent branch pairs are judged to be uniform adjacent pairs. If the vertical distribution distance is not within the preset distribution range, the adjacent branch pairs are judged to be non-uniform adjacent pairs. The uniformity of all adjacent branch pairs is judged, and the number of uniform adjacent pairs and the number of non-uniform adjacent pairs are counted. The distribution characteristics of the branches are judged by comparing the number of uniform adjacent pairs and the number of non-uniform adjacent pairs. If the number of uniform adjacent pairs is greater than or equal to the number of non-uniform adjacent pairs, the branch distribution characteristics are judged to be uniform distribution. If the number of uniform adjacent pairs is less than the number of non-uniform adjacent pairs, the branch distribution characteristics are judged to be non-uniform distribution.

[0137] Example: See Figure 3 In 3-b, if the vertical distribution distances of two pairs of adjacent branch pairs, branch outline A and branch outline B, and branch outline B and branch outline C, are both within the preset distribution range, then the number of uniform adjacent pairs is 2 pairs, and the number of uneven adjacent pairs is 0 pairs, and 2 pairs is greater than 0 pairs. Therefore, it is judged that the branch distribution characteristic is uniform distribution, that is, uniform distribution is determined as the visual layout parameter corresponding to the branch analysis group.

[0138] It should be noted that the internal pixels located at the corners, turning points or bifurcations can be selected from the tree trunk contour pixels as the internal feature pixels of the tree trunk contour ( Figure 3 Not all are marked in 3-b of FIG. 1 ), both sides (the first side and the second side) of the trunk contour may contain internal feature pixels. The vertical distance from each internal feature pixel to the trunk contour on the adjacent side is calculated, that is, the vertical distance from each internal feature pixel on the first side to the second side and the vertical distance from each internal feature pixel on the second side to the first side are calculated respectively. The average of the multiple vertical distances between the first side and the second side is calculated, and the obtained average distance is determined as the trunk width.

[0139] D35, identifying the root contour pixels, obtaining upper root contour pixels and lower root contour pixels corresponding to each root contour, and determining a plurality of internal node pixels in the upper root contour pixels and the lower root contour pixels.

[0140] The upper root contour pixels represent contour pixels located in the upper half of the tree root contour; the lower root contour pixels represent contour pixels located in the lower half of the tree root contour; the inner node pixels represent pixel points located inside the root contour, which may be inflection point pixels or turning point pixels inside the root contour, see Figure 3 3-c in (not all are marked).

[0141] D36. Obtain the vertical distance from each inner node pixel of the current root contour to the upper root contour or the lower root contour, determine the average width of the current root contour based on multiple vertical distances, determine the root thickness characteristics according to the multiple width averages, and determine the root thickness characteristics as the visual layout parameters corresponding to the root analysis group.

[0142] Among them, the current root contour represents the root contour currently being analyzed and processed; the vertical distance represents the vertical distance from the inner node pixel to the upper root contour or the lower root contour; the width average value represents the average value obtained by calculating the mean of all vertical distances of the current root contour; the root thickness characteristics can reflect the health and growth status of the tree roots, so as to analyze the psychological characteristics of the tree painter based on this.

[0143] Specifically, the root analysis group includes multiple root contours. For the current root contour, when the inner node pixel is located in the upper root contour, the vertical distance from the inner node pixel to the lower root contour is calculated. When the inner node pixel is located in the lower root contour, the vertical distance from the inner node pixel to the upper root contour is calculated. The average of all vertical distances corresponding to this root contour is obtained to obtain the average width. The average widths corresponding to all root contours are obtained in the above manner. The average of each width average is calculated again to obtain the overall root average. The overall root average is compared with the preset root width range. If the overall root average is within the preset root width range, the root thickness characteristic is judged to be a balanced root system. If the overall root average is greater than the upper limit of the preset root width range, the root thickness characteristic is judged to be a thick root system. If the overall root average is less than the lower limit of the preset root width range, the root thickness characteristic is judged to be a thin root system. Figure 3 In 3-c, if the average value of the width of the three root profiles is calculated, and the overall average value of the root is within the preset root width range, then the root thickness characteristic of the root analysis group is judged to be root system balanced. The preset root width range can be adjusted according to actual conditions such as the height and width of the tree, and the present invention does not impose specific restrictions on this.

[0144] It should be noted that the specific tree painting mood analysis group can be adjusted according to actual conditions and can be set accordingly according to recognition needs. In this embodiment, only the above-mentioned tree painting mood analysis group is used as an example for introduction. The recognition based on other tree painting images can refer to the introduction of this embodiment and will not be repeated here.

[0145] D4. According to the creative basic attributes corresponding to the painting tree mood analysis group, the corresponding essential features and fluctuation features are located in the mood feature knowledge graph.

[0146] For example: According to the high density corresponding to the crown analysis group, the uniform distribution corresponding to the branch analysis group, and the root system balance corresponding to the root analysis group, the mood feature knowledge graph is traversed to find the essential characteristics and fluctuation characteristics corresponding to each graffiti mood analysis group. For example, the high density corresponding to the crown analysis group may correspond to essential characteristics such as tension and attention to details, and fluctuation characteristics such as anxiety. Specifically, the mood feature knowledge graph is used as a benchmark, and the mood feature knowledge graph can be constructed according to existing empirical data, literature materials and other actual conditions, and the present invention does not make specific limitations on this.

[0147] It should be noted that by obtaining the basic creative attributes corresponding to each mood analysis group and traversing the mood feature knowledge graph based on the basic creative attributes, the emotional and psychological state of the subject is associated with his or her free creative style, which assists the psychoanalyst in judging the emotional and psychological state of the subject.

[0148] In some embodiments, the creation mapping feature includes a sandbox feature, and multiple mood analysis groups are obtained according to the creation mapping feature, and the creation basic attributes of each mood analysis group are obtained, including steps E1-E3: E1. Group the sandbox features according to the feature clustering grouping strategy to obtain multiple sandbox mood analysis groups, where the sandbox mood analysis groups include a mold analysis group, a component analysis group, and an ornament analysis group.

[0149] Among them, the sand table characteristics represent the key element features in sand table creation, such as terrain, buildings, transportation, etc., reflecting the decisions and tendencies of the subjects when constructing the sand table; the sand table mood analysis group represents the result of grouping the features in the sand table by category, focusing on a certain type of features in sand painting; the mold analysis group represents the focus on the grouping of mold features in the sand table; the component analysis group represents the focus on the grouping of component features in the sand table; the ornament analysis group represents the focus on the grouping of ornament features in the sand table.

[0150] It should be noted that the feature clustering grouping strategy can group the sand table features extracted from the sand table normalization map by category or similarity. By grouping, the sand table features can be classified into different mood analysis groups, which can provide a basis for subsequent analysis and enable the system to process and analyze different sand table features more effectively.

[0151] E2. Obtain the component decision category corresponding to each sandbox mood analysis group, and determine the creation basic attributes of the corresponding sandbox mood analysis group based on the component decision category.

[0152] The component decision category represents the decision-making pattern and tendency reflected by the subject when selecting, configuring or combining different components in the creative environment of the sandbox.

[0153] It should be noted that the selection or combination of each component reflects the subjects’ decisions on materials, layout, function, and aesthetics, and can lead to the subjects’ psychological characteristics and emotional expression patterns.

[0154] In some embodiments, step E2 (obtaining the component decision category corresponding to each sandbox mood analysis group, and determining the creation basic attributes of the corresponding sandbox mood analysis group based on the component decision category) specifically includes steps E21-E26: E21. Obtain a component decision category corresponding to the mold analysis group, where the component decision category includes a terrain mold and a building mold.

[0155] Among them, the terrain mold represents a basic model for creating terrain structures (such as hills, plains, canyons, etc.) in a sandbox; the building mold represents a basic model for creating building structures (such as houses, towers, city walls, etc.) in a sandbox.

[0156] E22. Determine the terrain mold and / or the architectural mold as the creation basic attributes of the mold analysis group.

[0157] Specifically, if the component decision category is terrain mold, the terrain mold is determined as the basic creation attribute of the mold analysis group; if the component decision category is architectural mold, the architectural mold is determined as the basic creation attribute of the mold analysis group; if the component decision category is terrain mold and architectural mold, the terrain mold and architectural mold are determined as the basic creation attributes of the mold analysis group.

[0158] It should be noted that the mold analysis group focuses on the mold elements used in the sandbox. The mold is usually a preset, standardized sandbox component, which reflects how the subjects use existing tools and standards to build the sandbox. It shows the subjects' reliance on standardized or traditional choices, and exhibits regularity and orderliness.

[0159] E23. Obtain component decision categories corresponding to the component analysis group, where the component decision categories include traffic components, vegetation components, and water components.

[0160] Among them, the traffic component represents traffic-related components, such as roads, bridges, etc., which reflect the layout and connectivity of the traffic network in the sandbox; the vegetation component represents elements related to the natural environment, such as trees, shrubs, lawns, etc., which may reflect the subjects' concern about the ecological environment; the water component represents water elements in the sandbox, such as lakes, rivers, etc.

[0161] It should be noted that the component analysis group focuses on the components customized by the subjects in the sandbox. These components are non-preset, or personalized sandbox components composed of multiple elements. They are often constructed by the subjects based on their understanding and intentions, and reflect the subjects' personalized expression, creative thinking, etc.

[0162] E24. Determine the traffic component and / or the vegetation component and / or the water component as the creation basic attributes of the component analysis group.

[0163] E25. Obtain the component decision category corresponding to the ornament analysis group, where the component decision category includes human ornaments, animal ornaments, and travel tool ornaments.

[0164] Among them, the character ornaments refer to small models used to simulate human characters in the sandbox, usually expressed as static character images; the animal ornaments refer to models used to represent animals in the sandbox, such as birds, horses, pets, etc.; the travel tool ornaments refer to small models of transportation tools in the sandbox, such as bicycles, carriages, ships, etc.

[0165] It should be noted that the ornament analysis group focuses on small decorative elements placed in the sand table. These ornaments usually do not have major structural functions, but play a decorative or embellishment role in the sand table. Their performance in aesthetics and details can reflect the subjects' attention to details and pursuit of beauty. The number, position, and combination of ornaments can convey the creator's emotional and psychological state.

[0166] E26. Determine the human figure ornament and / or the animal ornament and / or the travel tool ornament as the basic creation attributes of the ornament analysis group.

[0167] By analyzing each mood analysis group, we can derive the psychological characteristics and emotional expression patterns of the subjects. For example, a sand tray that uses a large number of molds may indicate that the creator tends to favor stability and order, while a sand tray filled with ornaments may reflect the creator's emphasis on details and aesthetics.

[0168] It should be noted that the specific sandbox mood analysis group can be adjusted according to actual conditions and can be set accordingly according to recognition needs. In this embodiment, only the above-mentioned sandbox mood analysis group is used as an example for introduction. The recognition based on other sandbox images can refer to the introduction of this embodiment and will not be repeated here.

[0169] E3. Based on the mood feature knowledge graph, the essential characteristics and fluctuation characteristics corresponding to the basic attributes of the creation of each sandbox mood analysis group are located respectively.

[0170] It should be noted that by dividing the elements in the sandbox into three categories: molds, components, and ornaments, the grouping process effectively improves the analysis efficiency, and each mood analysis group has a corresponding component decision category, reducing the possibility of confusion and cross-analysis. The creative basic attributes of each sandbox mood analysis group are closely related to the creator's inner mood. By obtaining the essential characteristics and fluctuation characteristics corresponding to the creative basic attributes of each sandbox mood analysis group, it can provide data basis for the subsequent overall mood assessment of the subjects.

[0171] S4. Fusing the essential feature and the fluctuation feature to obtain a mood feature fusion result, calling a vertical indicator evaluation model to evaluate the mood feature fusion result, and generating a mood evaluation result.

[0172] Among them, fusion represents the process of integrating the essential characteristics and fluctuation characteristics corresponding to each mood analysis group respectively. The essential characteristics and fluctuation characteristics can be fused by data analysis techniques such as weighted fusion and hierarchical analysis method. Data analysis techniques such as weighted fusion and hierarchical analysis method are well known to those skilled in the art and will not be repeated here. The mood feature fusion result represents the comprehensive emotional feature representation after integrating the personal characteristics and real-time characteristics of the subject. The vertical indicator evaluation model represents a model for realizing the evaluation of specific indicators in a vertical field, wherein the specific indicators belong to the user's performance tendency or ability specialty range; the mood evaluation result represents the specific analysis result of the emotional and psychological state of the subject.

[0173] Since the specific application scenario is not limited in this embodiment, the specific evaluation indicators in the vertical indicator evaluation model are not limited. Specifically, the vertical indicator evaluation model can be specifically set according to the specific application field. For example, the vertical indicator evaluation model can be a negative emotion evaluation model to achieve early screening of negative emotions (depression).

[0174] It should be noted that the objects input into the vertical indicator evaluation model for analysis may include other data in addition to essential characteristics and fluctuation characteristics, such as creation mapping characteristics, so as to achieve multi-faceted comprehensive evaluation.

[0175] Since the specific model evaluation indicators for calling the vertical indicator evaluation model are not limited in this embodiment, the evaluation method under different evaluation indicators, that is, the performance degree standard of the essential characteristics and fluctuation characteristics under the specified indicators is also not limited. It can be determined based on the actual application object and the measured indicators, and will not be elaborated here.

[0176] After calling the vertical indicator evaluation model to conduct a comprehensive evaluation based on the performance of the mood feature fusion results under the specified indicator, an evaluation result for the specified indicator will be generated. The evaluation result indicates the subject's performance under the specified indicator, such as depression level, emotional state, etc.

[0177] S5. Calling a text generation model to organize the mood assessment results into sentences and generate an assessment analysis report.

[0178] Among them, the text generation model represents a natural language generation model, which is responsible for converting the mood assessment results into an easy-to-understand natural language report; the sentence organization represents the conversion of data descriptions into fluent and natural language descriptions; the assessment analysis report represents the final report output by the system, which contains a detailed analysis and suggestions on the emotional and psychological state of the subject.

[0179] It should be noted that the mood assessment results are generally several separate keywords. In order to improve readability, after obtaining the mood assessment results, this embodiment further calls the text generation model to apply natural semantic processing and deep learning technology to independently write an assessment analysis report based on the assessment results for the character traits of the free creation mapping diagram. The assessment analysis report is a logically complete and appropriately connected text. Organizing keywords into an assessment analysis report can improve the professionalism and readability of the results, thereby improving the user experience.

[0180] The specific model type and model structure of the text generation model specifically called are not limited in this embodiment. One implementation method is as follows: based on paddlehub, the image classification model xception71_imagenet is applied, based on the previously established creation mapping features and mood feature knowledge graph and the extracted text information, the text generation model ernie_gen_couplet and the sentiment analysis model senta_bow are called to generate synonyms with emotions and form report sentences, and then an evaluation and analysis report is formed according to the designed template.

[0181] For the user side (collection end), the implementation process of the above-mentioned emotional and psychological state analysis only requires taking a photo and uploading a free-created mapping diagram, and waiting for the system to return an evaluation and analysis report. The process is simple and does not require excessive manual operation, which optimizes the user experience and reduces the user's professional requirements.

[0182] like Figure 4 FIG. 1 is a schematic diagram of a state analysis and identification system provided by an embodiment of the present invention. The state analysis and identification system includes: The data receiving module is used to respond to the mood recognition instruction and receive the free creation mapping diagram uploaded by the acquisition end, wherein the free creation mapping diagram includes graffiti drawing diagram, tree drawing diagram and sandbox creation diagram.

[0183] The feature recognition module is used to normalize the free creation mapping image to obtain a normalized mapping image, call a feature recognition model to perform feature recognition on the normalized mapping image, and obtain a creation mapping feature.

[0184] The attribute acquisition module is used to obtain multiple mood analysis groups according to the creation mapping characteristics, obtain the creation basic attributes of each mood analysis group, and locate the essential characteristics and fluctuation characteristics corresponding to each creation basic attribute based on the mood feature knowledge graph.

[0185] The result generation module is used to fuse the essential features and the fluctuation features to obtain a mood feature fusion result, call a vertical indicator evaluation model to evaluate the mood feature fusion result, and generate a mood evaluation result.

[0186] The report generation module is used to call the text generation model to organize the mood assessment results into sentences and generate an assessment analysis report.

[0187] It should be noted that the above-mentioned state analysis and identification system can execute the state analysis and identification method provided in the embodiment of the present invention. For technical details and beneficial effects that are not fully described in the state analysis and identification system embodiment, please refer to the state analysis and identification method provided in the embodiment of the present invention.

[0188] Through the description of the above embodiments, it can be clearly understood by those skilled in the art that each embodiment can be implemented by means of software plus a general hardware platform, or by hardware. It can be understood by those skilled in the art that all or part of the processes in the above embodiment methods can be completed by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes of the embodiments of the above methods. Among them, the storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc.

[0189] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them. Under the concept of the present invention, the technical features in the above embodiments or different embodiments may also be combined, the steps may be implemented in any order, and there are many other changes in different aspects of the present invention as described above, which are not provided in detail for the sake of simplicity. Although the present invention has been described in detail with reference to the aforementioned embodiments, a person of ordinary skill in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A state analysis and identification method, characterized in that: include: In response to the mood recognition instruction, receiving the free creation mapping diagram uploaded by the acquisition end, wherein the free creation mapping diagram includes a graffiti drawing diagram, a tree drawing diagram and a sandbox creation diagram; Normalizing the free creation mapping image to obtain a normalized mapping image, calling a feature recognition model to perform feature recognition on the normalized mapping image to obtain a creation mapping feature; According to the creation mapping features, multiple mood analysis groups are obtained, the creation basic attributes of each mood analysis group are obtained, and the essential characteristics and fluctuation characteristics corresponding to each creation basic attribute are located based on the mood feature knowledge graph; The essential feature and the fluctuation feature are fused to obtain a mood feature fusion result, and a vertical indicator evaluation model is called to evaluate the mood feature fusion result to generate a mood evaluation result; The text generation model is called to organize the mood assessment results into sentences and generate an assessment analysis report.

2. The state analysis and identification method according to claim 1, characterized in that: Before calling a feature recognition model to perform feature recognition on the normalized map and obtaining creation mapping features, the method includes: Receiving a creative image sample set, removing redundancy from the creative image sample set based on a preset screening rule to obtain an image screening sample set, and normalizing the image screening sample set to obtain a normalized sample set; Receiving the creative feature annotations of each image in the normalized sample set, and effectively cropping the corresponding images according to each creative feature annotation to obtain a cropped image set; Invoke a data enhancement strategy to process the cropped image set to generate an extended image set, train a network building model according to the extended image set, and obtain a network training model; Testing and verifying the network training model, and obtaining feedback adjustment data based on the testing and verification, wherein the feedback adjustment data includes model convergence data and recognition accuracy data; The parameters of the network training model are optimized according to the model convergence data and the recognition accuracy data to obtain a feature recognition model.

3. The state analysis and identification method according to claim 1, characterized in that: The feature recognition model, the vertical indicator evaluation model and the text generation model are distributed and deployed on a plurality of servers, and the method includes: Receiving stage processing data, identifying the stage processing data according to a preset category judgment rule, and obtaining a stage data category, wherein the stage data category includes a normalized mapping diagram, a mood feature fusion result, and a mood evaluation result; According to the normalized mapping graph, the server corresponding to the feature recognition model is determined as a feature recognition server, and the feature recognition server is called to perform feature recognition on the normalized mapping graph to obtain a creation mapping feature; According to the mood feature fusion result, the server corresponding to the vertical indicator evaluation model is determined as the indicator evaluation server, and the indicator evaluation server is called to evaluate the mood feature fusion result to obtain a mood evaluation result; According to the mood assessment result, the server corresponding to the text generation model is determined as the text generation server, and the text generation server is called to organize the mood assessment result into sentences to generate an assessment analysis report.

4. The state analysis and identification method according to claim 1, characterized in that: The creation mapping feature includes a graffiti feature, According to the creation mapping features, multiple mood analysis groups are obtained, the creation basic attributes of each mood analysis group are obtained, and the essential characteristics and fluctuation characteristics corresponding to each creation basic attribute are located based on the mood feature knowledge graph, including: Calling a feature clustering grouping strategy to group the graffiti features to obtain a plurality of graffiti mood analysis groups, wherein the graffiti mood analysis groups include a coloring analysis group, a line analysis group, and a shape analysis group; Obtain pixel occupancy data corresponding to each graffiti mood analysis group, wherein the pixel occupancy data includes each preset graffiti dimension and a corresponding pixel occupancy value, and sort the multiple pixel occupancy values ​​in descending order to obtain a graffiti pixel occupancy sequence; Acquire a maximum pixel occupancy value based on the graffiti pixel occupancy sequence, determine a corresponding visual dominant parameter according to the maximum pixel occupancy value, and determine the visual dominant parameter as a creation basic attribute corresponding to the graffiti mood analysis group; The mood feature knowledge graph is called, and the corresponding essential features and fluctuation features are located in the mood feature knowledge graph according to the basic creation attributes corresponding to each graffiti mood analysis group.

5. The state analysis and identification method according to claim 4, characterized in that: The graffiti pixel occupancy sequence includes a color pixel occupancy sequence, a line pixel occupancy sequence and a shape pixel occupancy sequence, obtaining a maximum pixel occupancy value based on the graffiti pixel occupancy sequence, and determining a visual dominant parameter corresponding to each graffiti mood analysis group based on the maximum pixel occupancy value, including: Obtaining a maximum coloring occupancy value based on the coloring pixel occupancy sequence, and obtaining a coloring main tone based on the maximum coloring occupancy value, wherein the visual main tone includes a bright tone, a neutral tone, and a dim tone; Determine the bright tone, the neutral tone, or the dark tone as the visual dominant parameter corresponding to the coloring analysis group; Obtaining a maximum line occupancy value based on the line pixel occupancy sequence, and obtaining a line main body shape based on the maximum line occupancy value, wherein the line main body shape includes a straight shape and a curved shape; Determining the straight form or the curved form as the visual dominant parameter corresponding to the line analysis group; Acquire a highest shape occupancy value based on the shape pixel occupancy sequence, and obtain a shape main body characteristic based on the highest shape occupancy value, wherein the shape main body characteristic includes a sharp characteristic and a smooth characteristic; The sharp characteristic or the smooth characteristic is determined as the visual dominant parameter corresponding to the shape analysis group.

6. The state analysis and identification method according to claim 5, characterized in that: The creation mapping feature includes a drawing tree feature, and multiple mood analysis groups are obtained according to the creation mapping feature, and the creation basic attributes of each mood analysis group are obtained. The essential features and fluctuation features corresponding to each creation basic attribute are located based on the mood feature knowledge graph, including: The tree drawing features are grouped according to the feature clustering grouping strategy to obtain a plurality of tree drawing mood analysis groups, wherein the tree drawing mood analysis groups include a crown analysis group, a branch analysis group and a root analysis group; Obtain edge detection data corresponding to each painting tree mood analysis group, and extract corresponding element contour pixels based on each edge detection data; Analyze the element outline pixels to obtain corresponding visual layout parameters, and determine the visual layout parameters as the creation basic attributes corresponding to the painting tree mood analysis group; According to the creative basic attributes corresponding to the painting tree mood analysis group, the corresponding essential characteristics and fluctuation characteristics are located in the mood feature knowledge graph.

7. The state analysis and identification method according to claim 6, characterized in that: The element contour pixels include leaf contour pixels, branch contour pixels, and root contour pixels. The element contour pixels are analyzed to obtain corresponding visual layout parameters, including: Locating the crown segmentation area based on leaf contour pixels, performing regional segmentation on the crown segmentation area to obtain multiple crown segmentation blocks, counting the leaf contour pixels of each crown segmentation block, and obtaining the number of contour pixels corresponding to each crown segmentation block; Calculate the mean value of the number of outline pixels corresponding to each crown segmentation block to obtain the leaf layout density, and determine the leaf layout density as the visual layout parameter corresponding to the crown analysis group; Determine the outer node pixels of each branch outline based on the branch outline pixels, obtain the vertical distance between each outer node pixel and the adjacent branch outline, calculate the average of multiple vertical distances of adjacent branch pairs, and obtain the vertical distribution distance of each adjacent branch pair; Comparing the vertical distribution distance of each adjacent branch pair with a preset distribution range to determine the branch distribution characteristics, and determining the branch distribution characteristics as the visual layout parameters corresponding to the branch analysis group; Identify the root contour pixels, obtain upper root contour pixels and lower root contour pixels corresponding to each root contour, and determine a plurality of internal node pixels in the upper root contour pixels and the lower root contour pixels; Obtain the vertical distance from each inner node pixel of the current root contour to the upper root contour or the lower root contour, determine the average width of the current root contour based on multiple vertical distances, determine the root thickness characteristics according to the multiple width averages, and determine the root thickness characteristics as the visual layout parameters corresponding to the root analysis group.

8. The state analysis and identification method according to claim 7, characterized in that: The creation mapping feature includes a sandbox feature. According to the creation mapping feature, multiple mood analysis groups are obtained, the creation basic attributes of each mood analysis group are obtained, and the essential characteristics and fluctuation characteristics corresponding to each creation basic attribute are located based on the mood feature knowledge graph, including: The sand table features are grouped according to the feature clustering grouping strategy to obtain a plurality of sand table mood analysis groups, wherein the sand table mood analysis groups include a mold analysis group, a component analysis group, and an ornament analysis group; Obtaining component decision categories corresponding to each sandbox mood analysis group, and determining creation basic attributes of the corresponding sandbox mood analysis group based on the component decision categories; Based on the mood feature knowledge graph, the essential characteristics and fluctuation characteristics corresponding to the basic attributes of the creation of each sandbox mood analysis group are located respectively.

9. The method for analyzing and identifying the driving emotional psychological state according to claim 8, characterized in that: Obtaining the component decision category corresponding to each sandbox mood analysis group, and determining the creation basic attributes of the corresponding sandbox mood analysis group based on the component decision category, including: Acquire a component decision category corresponding to the mold analysis group, wherein the component decision category includes a terrain mold and a building mold; determining the terrain mold and / or the building mold as a creation base attribute of the mold analysis group; Obtaining a component decision category corresponding to the component analysis group, wherein the component decision category includes a traffic component, a vegetation component, and a water component; Determine the traffic component and / or the vegetation component and / or the water component as the creation basic attributes of the component analysis group; Obtaining a component decision category corresponding to the ornament analysis group, wherein the component decision category includes human ornaments, animal ornaments, and travel tool ornaments; The character ornaments and / or the animal ornaments and / or the travel tool ornaments are determined as the basic creation attributes of the ornament analysis group.

10. A state analysis and identification system, characterized in that: include: A data receiving module, used to respond to the mood recognition instruction and receive the free creation mapping diagram uploaded by the acquisition end, wherein the free creation mapping diagram includes a graffiti drawing diagram, a tree drawing diagram and a sandbox creation diagram; A feature recognition module is used to normalize the free creation mapping map to obtain a normalized mapping map, call a feature recognition model to perform feature recognition on the normalized mapping map, and obtain a creation mapping feature; An attribute acquisition module, used to obtain multiple mood analysis groups according to the creation mapping features, obtain the creation basic attributes of each mood analysis group, and locate the essential characteristics and fluctuation characteristics corresponding to each creation basic attribute based on the mood feature knowledge graph; A result generation module is used to fuse the essential feature and the fluctuation feature to obtain a mood feature fusion result, call a vertical indicator evaluation model to evaluate the mood feature fusion result, and generate a mood evaluation result; The report generation module is used to call the text generation model to organize the mood assessment results into sentences and generate an assessment analysis report.

Images