Tree painting projection data collection and analysis system and method

Abstract

The invention relates to a tree painting projection data collection and analysis system and method, and the system comprises a data collection module, a data preprocessing module, a feature extraction module, a quantitative analysis module, a classification and prediction module, and a result evaluation module. The modules cooperate with each other to realize standardized processing, feature extraction, quantitative analysis, intelligent classification and prediction and result evaluation of data. According to the method, the holographic theory and the Ama pothole theory are combined, the drawing features are finely analyzed through the cross-modal comparative learning and graph neural architecture search combined algorithm, a scientific and objective psychological data acquisition and analysis method is provided, the diagnosis sensitivity and specificity are improved, a powerful tool is provided for clinical psychological assessment, and the method is suitable for clinical psychological assessment. The method has important clinical and research values, and has wide application prospects in the fields of psychological problem screening, social psychological services, judicial, public security, marriage matching, talent selection and the like.

Description

Technical Field

[0001] The present invention relates to the technical fields of artificial intelligence technology and medical big data processing system, and in particular to a tree painting projection data collection and analysis system and method for individual psychological states. Background Art

[0002] Depression and anxiety are common and serious mental illnesses that have a profound impact on an individual's physical and mental health. Traditional methods for diagnosing depression and anxiety mainly rely on clinical interviews and questionnaires. Although these methods are effective to a certain extent, they have obvious limitations. First, these methods rely on individual self-reports and are easily influenced by the subject being assessed. Second, the assessment results are greatly affected by the experience and judgment of the assessor and lack objectivity and standardization.

[0003] The Tree Drawing Projective Test (TDT) is a non-verbal scientific psychological assessment tool that has been widely used in the fields of mental health, education, justice, and military in recent years. It has important clinical and research value. The test analyzes the psychological state of individuals by asking them to draw pictures of trees. However, the traditional tree drawing projective test analysis method mainly relies on the experience and subjective judgment of experts and lacks scientific quantitative standards, resulting in low reliability and consistency of the assessment results.

[0004] Although some studies have attempted to apply quantitative analysis methods to tree drawing projective tests, in practical applications, there is still a lack of a systematic and standardized collection and analysis system. In order to make up for this shortcoming, the present invention proposes a tree drawing projective data collection and analysis system and method based on a cross-modal contrastive learning combined with a graph neural architecture search algorithm. The system combines holographic theory and Armas's pit theory, and conducts a detailed analysis of tree drawing features through a cross-modal contrastive learning combined with a graph neural architecture search algorithm, providing strong data support for scientific and objective depression assessment. This improves the sensitivity and specificity of diagnosis and provides a powerful tool for clinical psychological assessment. Summary of the invention

[0005] The purpose of the present invention is to provide a tree painting projection data collection and analysis system and method to solve the problems of strong subjectivity and lack of quantitative standards in the background technology. Traditional tree painting tests often rely on subjective judgment and are difficult to conduct systematic and quantitative analysis and evaluation. The present invention realizes quantitative analysis of individual tree painting features and depression evaluation through the collaborative work of a data collection module, a data preprocessing module, a feature extraction module, a classification and prediction module, and a result evaluation module, thereby improving the objectivity and accuracy of the evaluation.

[0006] To achieve the above-mentioned purpose, the present invention provides the following technical solutions: a tree painting projection data collection and analysis system, the system comprises a data collection module, a data preprocessing module, a feature extraction module, a quantitative analysis module, and a classification and prediction module connected in sequence; the system analyzes the painting features by combining holographic theory and Armas's pit theory, and using cross-modal contrast learning and graph neural architecture search algorithm modules;

[0007] The data collection module collects the tree images drawn by individuals through a scanner and transmits them to the system storage unit; the module includes an image acquisition unit and a data transmission unit to ensure the clarity and integrity of the image and supports wired and wireless transmission methods;

[0008] The data preprocessing module performs standardization processing on the collected tree paintings, including image denoising, grayscale conversion and image enhancement; uses median filtering for denoising, uses brightness method for grayscale conversion, and improves image contrast and clarity through Laplace operator enhancement technology;

[0009] The feature extraction module extracts the quantitative features of trees, including overall height, width, trunk width, and crown area, from the processed tree drawings; these features are extracted through edge detection, morphological processing, and region segmentation algorithms, and normalized to ensure the accuracy and stability of the data;

[0010] The quantitative analysis module quantifies the extracted features and generates feature vectors for analysis to facilitate subsequent classification and prediction;

[0011] The classification and prediction module is based on quantized feature vectors, performs multi-category classification training through cross-modal contrast learning and graph neural architecture search algorithm, establishes a classification model, and performs model verification and optimization to improve classification performance.

[0012] Furthermore, it also includes a result evaluation module, which provides specific auxiliary diagnosis suggestions by generating an evaluation report, including an individual's psychological state analysis and depression evaluation results.

[0013] A method of a tree painting projection data collection and analysis system of the present invention comprises the following steps:

[0014] Step 1, collecting individual psychological state information, and the data collection module collecting tree drawings drawn by the individual, and collecting images of the tree drawings drawn by the individual through a scanner;

[0015] Step 2: The data preprocessing module performs standardized preprocessing on the collected tree paintings, including median filtering for image denoising, grayscale conversion, and image enhancement to eliminate image noise and improve image quality;

[0016] Step 3, the feature extraction module extracts multiple quantitative features from the processed tree drawings, including the overall height, width, trunk width, crown area, number of branches, and leaf density of the tree;

[0017] Step 4: The quantitative analysis module quantifies the extracted features and generates feature vectors for analysis;

[0018] Step 5: The classification and prediction module classifies and predicts the psychological state based on the quantized feature vectors by combining cross-modal contrastive learning with graph neural architecture search.

[0019] Furthermore, it also includes step 6, in which the result evaluation module evaluates the results of classification and prediction and outputs an analysis report, wherein the analysis report includes the individual's psychological state analysis and depression evaluation results, and generates a detailed evaluation report for clinical use by evaluating the classification and prediction results.

[0020] Further, step 1 includes:

[0021] Professional psychologists set up the following settings: using simple and clear language to explain the purpose and requirements of the drawing task to the individual in detail and assess their psychological state; providing standard A4 white paper, HB pencils and erasers, allowing the individual to make revisions, but recording the frequency of revisions to reflect their hesitation and decision-making process and obtain additional psychological state information; choosing a quiet and comfortable room to avoid noise and interference, ensuring soft and adequate lighting, and avoiding strong light and shadows that affect the quality of the drawing;

[0022] Observe and record psychological states: During the painting process, psychologists need to record the type of emotions, frequency and intensity of changes in detail, analyze the changes in facial expressions such as frowning, smiling and sighing; record the content of speech, emotional color, accuracy and speed of answers, use recording equipment to record conversations, and then transcribe and analyze; record hand movements, posture adjustments and micro-expressions, and use high-resolution cameras to record the entire painting process to ensure that no details are missed;

[0023] Data collection tools: High-resolution camera: records the painting process, including hand movements and facial expressions, to ensure clear video quality and capture subtle changes in facial expressions; heart rate monitor: monitors heart rate changes in real time during the painting process to reflect physiological stress levels; skin conductance meter: records skin electrical responses as an indicator of physiological stress response.

[0024] Further, in step 2:

[0025] The median filter module is used to effectively remove salt and pepper noise, retain edge details, and is suitable for discrete noise; let I be the original image, I′ be the filtered image, and N(i,j) be the neighborhood of pixel (i,j), then I′(i,j)=median{I(k,l)|(k,l)∈N(i,j)};

[0026] Where, I: original image; I′: filtered image; N(i,j): neighborhood of pixel (i,j); (k,l): pixel coordinates in the neighborhood; median: median operation; i: horizontal coordinate in the image; j: vertical coordinate in the image;

[0027] The noise detection module is used to accurately locate the noise area in the image, so as to perform targeted denoising to ensure that the details of the tree painting are preserved; the noise area in the image is identified and removed through the noise detection algorithm, thereby improving the overall quality of the image; let I be the original image, N(i,j) be the neighborhood of pixel (i,j), then the local variance is defined as

[0028] Where, I: original image; N(i,j): neighborhood of pixel (i,j); σ 2 (i, j): local variance of pixel (i, j); μ(i, j): mean value in the neighborhood; (k, l): pixel coordinates in the neighborhood; |N(i, j)|: number of pixels in the neighborhood;

[0029] A grayscale conversion module is used to ensure that the details and structure of the tree painting are preserved in the grayscale image. Grayscale conversion converts the RGB color image into a grayscale image, using the brightness method to ensure quality, simplify data, and retain brightness information. Let R, G, and B be the red, green, and blue components of the RGB image, respectively, and I be the grayscale image, then I = 0.299R + 0.587G + 0.114B.

[0030] Among them, R: red component; G: green component; B: blue component; I: grayscale image;

[0031] The contrast adjustment module is used to enhance the contrast by stretching the grayscale value range to ensure that the details of the grayscale image are not lost; by adjusting the contrast, the image recognizability and analysis accuracy are improved, which is crucial for the extraction and recognition of features in tree paintings; let I be the grayscale image, I′ be the image after contrast adjustment, and the minimum grayscale value be

[0032] Where, I: grayscale image; I′: image after contrast adjustment; I min : The minimum gray value in the image; I max : The maximum gray value in the image;

[0033] Image enhancement module: Use histogram equalization technology to improve image contrast and make image details clearer; improve image contrast, make feature extraction more accurate, and enhance image processing effect; suppose the gray value of the grayscale image is I, the gray value after histogram equalization is I′, and the cumulative distribution function is CDF, then

[0034] Where, I: grayscale value of grayscale image; I′: grayscale value after histogram equalization; CDF: cumulative distribution function; CDF min : The minimum non-zero value of the cumulative distribution function; i: the horizontal coordinate in the image; j: the vertical coordinate in the image;

[0035] Brightness adjustment module: adjust the overall brightness level of the image, ensure that the overall brightness of the image is moderate, improve the visual effect of the image, and adjust the overall brightness of the image through linear transformation; brightness adjustment makes the overall visual effect of the image more in line with the perception habits of the human eye; let I be the grayscale image, I′ be the image after brightness adjustment, then I′(i,j)=I(i,j)+ΔL, where I: grayscale image; I′: image after brightness adjustment; ΔL is the brightness adjustment amount;

[0036] Sharpening filter module: Sharpening filter improves the sharpness of the image and enhances the edge details. Sharpening filter ensures the edge of the tree painting image is clear and improves the accuracy of feature extraction; it improves the sharpness by enhancing the contrast of the image edge; Sharpening filter enhances the edge details of the image. Let I be the grayscale image, I′ be the sharpened image, and Laplace operator be L, then I′(i, j) = I(i, j) - αL(i, j), where I: grayscale image; I′: sharpened image; L: Laplace operator; α is the sharpening intensity coefficient; i: the horizontal coordinate in the image; j: the vertical coordinate in the image;

[0037] Psychological state data synchronization module: matches the timestamps of psychological state data and image data to ensure synchronization and merge them to generate a comprehensive data file for subsequent analysis and processing;

[0038] The timestamp matching module is used to ensure the consistency of data in time; let the timestamp of the image data be T img , the timestamp of the mental state data is T psy , then the matching condition is T img ≈T psy ;

[0039] Data fusion fuses the matched psychological state data with the image data to generate a comprehensive data file in a standardized JSON format.

[0040] Further, the step 3 comprises:

[0041] The edge detection module extracts edge information from tree paintings to help identify structural features and reflect the painter's concentration and hand control. The Canny algorithm is optimized to improve the accuracy of edge detection and noise reduction, ensuring detail clarity.

[0042] Canny edge detection is used to extract edges in tree paintings. It detects edges by calculating the intensity gradient of the image, ensuring high accuracy and low noise in edge detection. It includes five steps: noise removal, gradient calculation, non-maximum suppression, dual threshold detection, and edge tracking.

[0043]

[0044] Among them, I(x,y): original image, which means the pixel value of the original image at position (x,y); I′(x,y): filtered image, which means the pixel value of the filtered image at position (x,y); G(x,y): Gaussian kernel; x: horizontal coordinate in the image; y: vertical coordinate in the image; σ: standard deviation of Gaussian filter, which determines the degree of smoothing; the larger the value, the stronger the smoothing effect;

[0045] Image gradient: Calculate image gradient for edge detection;

[0046] Among them, G x and G y Represent the gradients in the x and y directions respectively; G: Gradient magnitude, which indicates the gradient magnitude of the image and reflects the edge strength;

[0047] Non-maximum suppression: retain the local maximum points of the gradient to enhance the accuracy of the edge;

[0048] Dual threshold detection: Use high and low thresholds to classify gradient magnitudes, connect strong edges, and retain potential edges;

[0049] Morphological processing: remove noise through dilation, erosion, opening and closing operations, highlight the main structure of the tree, and ensure that the extracted image features are accurate and reliable; further optimize the results of edge detection, remove noise, and highlight the main structure; used to ensure that the extracted image features are accurate and reliable, not affected by noise, and enhance the effect of subsequent analysis;

[0050] Expansion Operation I dilated (x,y)=max(i,j)∈KI(x+i,y+j)

[0051] Corrosion Operation I eroded (x,y)=min(i,j)∈KI(x+i,y+j)

[0052] Opening operation: first erosion and then expansion I opened (x,y)=dilate(erode(I,K),K)

[0053] Closing operation: dilation followed by corrosion I closed(x,y)=erode(dilate(I,K),K)

[0054] Among them, dilate(I,K) means dilating images I and K to obtain a new image; erode(I,K) means eroding images I and K to obtain a new image; I dilated (x, y): pixel value of the expanded image at position (x, y) K: structural element, used to define the range of the operation; max: take the maximum value in the local area to expand the object; I eroded (x, y): pixel value of the image after corrosion at position (x, y); min: take the minimum value in the local area to shrink the object; I opened : Image after opening operation; I closed : The image after closing operation;

[0055] Region segmentation: Use the watershed algorithm to segment the image and accurately extract the various regions of the tree; provide detailed structural information; analyze the overall morphological characteristics, evaluate the spatial perception and detail grasping ability; help evaluate the painter's spatial perception and detail grasping ability; and contribute to the comprehensive assessment of the psychological state;

[0056] Gradient calculation: Calculate the gradient of the image through morphological gradient; used to identify the area with strong edges in the image; G = morpho log yEx (I, MORPH_GRADIENT, K);

[0057] Among them, G: gradient map, an image used to represent the edge strength in an image; I: input image, that is, the original image that needs to be gradient calculated; K: structural element, a template used to define morphological operations; morphologyEx: OpenCV function, here, it performs the MORPH_GRADIENT operation, which generates a gradient map by calculating the difference between image dilation and erosion; this operation emphasizes the edge area in the image and is often used for edge detection;

[0058] Threshold segmentation: Use threshold segmentation to generate a label map for marking different areas in the image; _,marker s=threshold(G,255,THRESH_BINARY+THRESH_OTSU)

[0059] Among them, _: This symbol indicates that part of the return value is not used; markers: marker map, used to mark different areas in the image; G: gradient map, that is, the image generated in the previous part; the image after gradient calculation is used as input for threshold segmentation; 0: threshold lower limit, indicating that pixels less than this value will be processed as background; 255: threshold upper limit, indicating that pixels greater than this value will be processed as foreground; THRESH_BINARY: binary flag, indicating that the image pixels are converted to binary images, that is, all pixels are either 0 or 255; THRESH_OTSU: Otsu's method flag, indicating that the optimal threshold is automatically determined to segment the foreground and background;

[0060] Watershed algorithm: perform watershed segmentation on the marker map; generate marker values ​​for different regions; mark s = watershed(I, mark s);

[0061] Among them, markers: the marker map processed by the watershed algorithm; after the watershed algorithm is processed, markers will contain different integer values, which correspond to different segmented regions in the image; the watershed algorithm uses this marker map to segment the image, and the pixels in each region are assigned different marker values; I: input image, the original image used for the watershed algorithm; the watershed algorithm operates on this image and segments the regions in the image in combination with the marker map; watershed: the watershed function in OpenCV; it receives the input image and the marker map, and returns the updated marker map, where each marker corresponds to a segmented region;

[0062] Psychological feature extraction: Through time recording and image changes, the painting speed, pressure changes and emotional fluctuations during the painting process are analyzed to comprehensively evaluate the mental health status;

[0063] Drawing speed: by calculating the completed part in each time period during the drawing process; Among them, drawingspeed i : Drawing speed in the i-th time period; drawing_speed: drawing speed; S i : the part completed in the i-th time period; t i : i-th time point;

[0064] Pressure change: by calculating the pressure change at each time period during the painting process; pressure change i =P i+1 -P i Among them, pressure change i : Pressure change in the i-th time period; pressure_change: pressure change; Pi : Pressure data at the i-th time point;

[0065] Emotional fluctuations: by recording the emotional changes in each time period during the painting process; emotionchange i =(E i ,E i+1 ) Among them, emotion change i :emotional change in the i-th time period emotion_change: emotional change; E i : Emotion data at the i-th time point.

[0066] Further, the step 5 comprises:

[0067] The classification and prediction module is the key link to conduct in-depth analysis and make accurate judgments from the rich information extracted from tree painting characteristics and psychological state data; specifically:

[0068] Intelligent algorithm module selection: In order to deeply analyze the complex relationship between the structural features of tree paintings and psychological state data, an algorithm module combining cross-modal contrastive learning and graph neural architecture search was adopted; through the cross-modal contrastive learning method, the image features of tree paintings and psychological state data were mapped to a unified feature space; this step ensured the consistency between the features of different modal data, thus laying the foundation for subsequent multimodal data fusion; during the model training process, the image features of tree paintings and psychological state data were input into the deep neural network and graph neural network respectively, and multiple feature-level fusions were performed in the fusion layer; through the innovative hierarchical fusion strategy, the model can better understand the complex relationship of the data and optimize it through the cross-entropy loss function;

[0069] The formula of the contrast loss function is:

[0070] Where: L: total loss value, which is used to measure the effect of the model in aligning different modal features in the current training batch; the smaller the loss value, the better the model is in bringing similar samples closer and moving away from dissimilar samples; represents the sum of all sample pairs (i, j); the loss function will traverse all possible sample pairs, calculate their contrast loss, and then sum them up to get the total loss value; y ij : The label of the sample pair (i, j); D(z i ,z j ): The distance measure between sample i and sample j in the feature space; this distance can be the Euclidean distance or other forms of distance measure; D(z i ,z j )=||z i -z j || Among them, zi and z j are the feature vector representations of sample i and sample j respectively; 1-y ij : The opposite label of the sample pair (i, j); when the sample pair (i, j) belongs to different categories, y ij =0, then 1-y ij =1, indicating that the sample pair is not similar; max(0,mD(z i ,z j )): This is a penalty term used to ensure that pairs of samples of different classes maintain at least a distance m in the feature space; m: a predefined boundary threshold used to control the minimum distance between samples of different classes; if two samples belong to different categories, the distance between them should be at least m; otherwise, a penalty value is generated, thereby increasing the loss value;

[0071] Graph Neural Architecture Search Module GNAS: Graph Neural Architecture Search is a cutting-edge algorithm for automatically designing the optimal architecture of graph neural networks (GNNs). GNAS is used to automatically design and optimize GNN architectures suitable for tree painting features.

[0072] The basic update formula of GNN is:

[0073] in: is the feature representation of node v at the k+1th layer; is the feature representation of node μ in the kth layer; σ is the activation function, ReLU, used to introduce nonlinearity; W (k) is the weight matrix of the kth layer, which represents the relationship weight between nodes; N(v) is the neighborhood set of node v, which includes other nodes connected to node v; b(k) is the bias term, which is used to adjust the overall offset of the feature;

[0074] Model training: During the model training phase, an innovative multimodal fusion model was used for multi-category classification training. By combining the image features of tree paintings with psychological state data, the model was able to understand and classify different categories of data more comprehensively and accurately. In addition, the use of data augmentation technology greatly increased the diversity of training data, further improving the generalization ability and robustness of the model.

[0075] The formula for feature-level fusion is:

[0076] Among them: z: fused multimodal feature representation; this feature vector fuses data features from different modalities and is the final input of the model for classification, regression or other tasks; f image (x image ): feature representation extracted from image data; x image : Input image data; fimage (): Image feature extraction function; this function is usually a deep neural network used to extract useful feature representations from the input image data; f psych (x psych ): feature representation extracted from mental state data; x psych : Input psychological state data; f psych (): psychological state feature extraction function; this function is a neural network or other form of model used to extract features from the input psychological data; ⊕: feature connection operator; this operation concatenates feature vectors from different modalities to generate a longer feature vector containing multimodal information;

[0077] The formula for decision-level fusion is: final =α·y image +(1-α)·y psych

[0078] Where: y final is the final fusion prediction result; y image and psych are the independent prediction results of image and psychological state data respectively; α is the weighting coefficient, which is used to control the influence of each modality on the final decision;

[0079] The data augmentation technique is introduced, which generates more diverse training samples by randomly rotating the image data and adding noise.

[0080] For image data, the transformation formula for data augmentation is: x′=transform(x);

[0081] Where: x′ is the transformed image data; transform(x) is the random transformation operation used for image data, including rotation, translation, and flipping;

[0082] Model validation and optimization: In the model validation phase, an advanced k-fold cross-validation method is used to evaluate the performance of the model, and the model parameters are optimized by the improved whale optimization algorithm. This innovative method combination not only improves the accuracy of classification, but also ensures the stable performance of the model on different data sets. In the model validation phase, the data set is divided into k subsets D1, D2, …, D k , then the accuracy calculation formula for the i-th verification is:

[0083] Among them: A i is the accuracy of the i-th verification; D i is the model obtained from the i-th training is to use the D iThe prediction result of the model trained with the remaining data except for x for sample x; y is the true label of the sample; 1 is the indicator function, which takes 1 when the prediction is correct and 0 otherwise;

[0084] In order to further optimize the model, the improved WOA algorithm is used to dynamically adjust the learning rate and regularization coefficient hyperparameters so that the model can achieve the best performance in the classification task;

[0085] The formula for the circling phase is:

[0086] Where: X t+1 is the whale position of generation t+1; is the position of the current optimal solution; A and C are parameters that control the intensity and distance of the circle, ensuring that the whale gradually approaches the optimal solution in the global search.

[0087] Further, the step 6 is specifically as follows:

[0088] Result evaluation is a key step in the classification and prediction of mental state. The evaluation model predicts the results and generates a detailed evaluation report. The report combines a variety of data visualization technologies and analysis methods to accurately display the results of clinical significance. After the classification and prediction are completed, the system generates a comprehensive report, including individual mental state analysis, depression assessment and emotional fluctuations during the painting process. Through images, charts and text descriptions, the report intuitively presents the evaluation results, including mental state analysis: based on the classification and prediction results, comprehensively analyze the individual's mental state, identify anxiety and depression problems, and conduct quantitative assessments in combination with clinical scales; depression assessment: integrate the results of mental state analysis, apply standardized depression assessment models, provide scores, and provide reference for clinicians; emotional fluctuation analysis: by analyzing the emotional dynamics during the painting process, generate emotional fluctuation charts, identify emotional changes in different time periods, and provide real-time monitoring data for clinical intervention.

[0089] Compared with the prior art, the present invention has the following beneficial effects:

[0090] 1. Improve the objectivity and accuracy of the analysis: The present invention significantly improves the objectivity and accuracy of the tree drawing test through systematic data processing and cross-modal contrastive learning combined with graph neural architecture search algorithms. Traditional tree drawing tests often rely on subjective visual judgment and evaluation. This method is easily affected by the evaluator's personal experience and judgment criteria, resulting in inconsistency and inaccuracy in the results. Relatively speaking, the present invention uses advanced data processing technology and cross-modal contrastive learning combined with graph neural architecture search algorithms to automatically analyze painting images. These technologies can extract detailed and stable features from images, reduce human intervention and subjective errors, and make the evaluation results more objective and accurate. This objectivity not only improves the credibility of the test results, but also helps to make consistency comparisons between different individuals, thereby providing more reliable psychological state assessment data in clinical applications.

[0091] 2. Realize quantitative assessment: The present invention can provide a comprehensive understanding of the individual's psychological state through quantitative analysis of tree painting characteristics. Evaluations in traditional tree painting tests often rely on qualitative descriptions, such as subjective interpretations of tree shape, size, color and other characteristics. This method is difficult to provide specific quantitative data, so it is difficult to conduct accurate psychological state assessments. Through the feature extraction module of the present invention, it is possible to conduct a detailed analysis of each quantitative feature in the painting. This quantitative assessment not only helps to gain a deep understanding of the individual's psychological state, but also provides more accurate psychological assessment results in a data-driven manner. Quantitative assessment results can provide more specific data support for mental health professionals.

[0092] 3. Provide professional advice: In the system developed by the present invention, the result evaluation module not only provides evaluation results such as depression, but also provides specific suggestions and treatment plans based on the analysis results. Traditional tree drawing tests usually only provide qualitative evaluation results, but lack specific intervention suggestions and treatment plans. Through the present invention, the result evaluation module generates a detailed evaluation report, which includes an analysis of the individual's mental state and an assessment of the risk of depression. In addition, the result interpretation unit will make personalized suggestions and treatment plans based on the analysis results. These suggestions may include recommendations for further psychological counseling, behavioral therapy, drug therapy and other interventions to help individuals better understand and manage their mental health. Providing specific treatment plans not only helps to improve the practicality and effectiveness of the assessment, but also helps mental health professionals better support patients in actual work and develop personalized treatment plans.

[0093] 4. Cross-modal contrastive learning establishes effective associations between different modalities (such as image features and psychological state data) and aligns the data features of these modalities so that they can be comprehensively analyzed in a unified feature space. This process enhances the complementarity of the data and avoids information loss due to modal differences. Data features of different modalities can be aligned and compared in the same space, which greatly improves the accuracy of data fusion and makes the classification and prediction of psychological states more accurate. Cross-modal contrastive learning can effectively process and bridge the differences between different modalities, thereby reducing classification errors and prediction biases due to data heterogeneity.

[0094] 5. Graph Neural Architecture Search (GNAS) ensures that the designed model can adapt to and optimize specific psychological data state analysis tasks by automatically searching for the optimal graph neural network structure. This automated network architecture design method not only reduces the time and cost of manual debugging, but also improves the performance of the model. GNAS can automatically design the best network structure according to different task requirements, so that the model can better capture the complex relationships in tree painting features and psychological state data, thereby improving the accuracy of classification and prediction. Through automated search, GNAS can find a more efficient and more suitable network architecture for task requirements, thereby improving the overall performance of the model and ensuring its stability and robustness on different data sets.

[0095] 6. Improve the accuracy of mental state classification: By combining cross-modal contrastive learning with the GNAS algorithm, the system can achieve higher-precision mental state classification and prediction. This combination enables the system to not only effectively integrate multimodal data, but also improve the performance of the classifier through optimized network structure to ensure the reliability of the prediction results. The effective fusion of multimodal data and the automated optimized network architecture design enable the system to show higher accuracy in mental state classification and reduce the risk of misclassification and misdiagnosis. Combining multimodal data and optimized network structure, the system shows stronger robustness and consistency in processing mental state data, ensuring the reliability of the prediction results.

[0096] 7. The system that combines cross-modal contrastive learning with GNAS can continuously monitor the individual's mental state and provide long-term support through intelligent analysis. This intelligent continuous monitoring can not only detect changes in mental state in a timely manner, but also adjust intervention plans based on real-time data to provide long-term mental health management support. The system can monitor the individual's mental health status in real time through continuous tree painting tests and mental state records, and provide timely feedback to psychologists or clinicians to facilitate rapid adjustment of intervention strategies. Through intelligent monitoring and evaluation, the system can provide long-term mental health support services, such as regular mental state assessments, personalized counseling, and educational content recommendations, to help individuals maintain and improve their mental health in the long term. BRIEF DESCRIPTION OF THE DRAWINGS

[0097] Figure 1 :The present invention introduces a flowchart of the algorithm combining cross-modal contrastive learning and graph neural architecture search;

[0098] Figure 2 : The present invention introduces the Canny edge detection algorithm flow chart;

[0099] Figure 3 : The present invention introduces an improved whale optimization algorithm flow chart. DETAILED DESCRIPTION

[0100] Embodiment: The present invention provides a depression tree painting projection analysis system, including a data collection module, a data preprocessing module, a feature extraction module, a classification and prediction module and a result evaluation module. The general algorithm and each module involved in the present invention are implemented by software coding, or by calling the corresponding algorithm module and calling the M function code by MATLAB. The algorithm can be implemented in a variety of programming environments and run in a general computer operating system, including but not limited to C language, Python and SPSSAU and other tools.

[0101] Data collection module

[0102] The data collection module includes an image acquisition unit, a data transmission unit, and a psychological state recording unit. The image acquisition unit uses a high-resolution scanner to collect high-resolution images of tree paintings drawn by individuals to ensure the integrity and accuracy of image details. The psychological state recording unit obtains comprehensive psychological state data by observing and recording the emotional changes, verbal reactions, and body language of individuals during the painting process. The collected images and psychological state data are encrypted and transmitted to the system's storage unit through the data transmission unit, providing a data basis for subsequent processing.

[0103] Data preprocessing module

[0104] The data preprocessing module includes an image denoising module, a grayscale conversion module, an image enhancement module and a mental state data synchronization processing module. The image denoising module uses algorithms such as mean filtering, median filtering and Gaussian filtering to remove image noise; the grayscale conversion module converts color images into grayscale images to simplify the calculation process; the image enhancement module improves the clarity of the image by adjusting parameters such as contrast, brightness and sharpness to prepare for feature extraction. The mental state data synchronization processing module preprocesses and synchronizes the mental state data to ensure the consistency of the image data and the mental state data.

[0105] Feature extraction module

[0106] The feature extraction module extracts key features from the preprocessed image and conducts a comprehensive analysis in combination with the psychological state data. Specifically, it includes an edge detection unit, a morphological processing unit, a region segmentation unit, and an emotional feature extraction unit. The edge detection unit identifies the outline of trees through edge detection methods such as the Canny algorithm; the morphological processing unit applies morphological operations such as dilation, erosion, opening and closing operations to extract the structural features of trees; the region segmentation unit uses the watershed algorithm and the k-means clustering algorithm to divide the image into different regions and calculate the quantitative features of the trees, such as the overall height, width, trunk width, crown area, number of branches, and leaf density. The emotional feature extraction unit combines the psychological state data to extract the anxiety, depression, and tension emotional characteristics of individuals during the painting process.

[0107] Classification and prediction module

[0108] The classification and prediction module uses a combination of cross-modal contrastive learning and graph neural architecture search algorithms to classify and predict the extracted features. The model training unit uses a multimodal fusion model to perform multi-category classification training on labeled tree painting data and psychological state data to ensure high classification accuracy and low noise; the model verification unit uses a cross-validation method to evaluate model performance and ensure classification accuracy; the model optimization unit uses an improved whale optimization method to optimize model parameters to improve the accuracy and stability of classification and prediction.

[0109] Results Evaluation Module

[0110] The result evaluation module generates a detailed evaluation report. Based on the results of the classification and prediction module, the report generation module generates a detailed report including individual psychological state analysis, depression evaluation results, and emotional fluctuation analysis during the painting process. The result interpretation module explains the evaluation results in detail, provides individualized explanations and suggestions, helps individuals understand their mental health status, and develops specific psychological counseling, drug treatment, and behavioral intervention treatment plans. The evaluation report also includes multimedia content such as images, charts, and text descriptions to intuitively display the evaluation results.

[0111] Results Application Module

[0112] The results application module conducts individualized psychological interventions based on the recommendations and treatment plans provided in the evaluation report. The intervention measures formulation unit formulates individualized cognitive behavioral therapy, art therapy, and group therapy psychological intervention measures based on the evaluation report. Intervention measures are implemented under the guidance of professional psychologists to help individuals manage their mental health. The results application module also includes continuous monitoring and evaluation, regular tree painting tests and psychological state records, and dynamic adjustment of intervention measures to ensure the continuity and effectiveness of treatment effects. Through psychological counseling services and mental health education, individuals' understanding and management capabilities of mental health are improved to prevent the occurrence of psychological problems.

[0113] Implementation steps

[0114] In the depression tree painting projection analysis system and method, based on the combination of cross-modal contrast learning and graph neural architecture search algorithm, this process is optimized, and the efficient fusion and alignment of multimodal data are achieved through the algorithm; the network structure design is automatically optimized to improve the accuracy of psychological state classification and prediction, provide precise guidance for personalized psychological intervention, conduct continuous monitoring and long-term support intelligence, and generate more intuitive visual evaluation reports; the specific process is:

[0115] Step 1: Data collection and psychological state analysis

[0116] To ensure the standardization and validity of data collection, professional psychologists need to make the following settings: use simple and clear language to explain the purpose and requirements of the drawing task to the individual in detail and assess their psychological state; provide standard A4 white paper, HB pencils and erasers, allow individuals to make revisions, but record the frequency of revisions to reflect their hesitation and decision-making process and obtain additional psychological state information; choose a quiet and comfortable room to avoid noise and interference, ensure soft and sufficient lighting, and avoid strong light and shadows that affect the quality of the drawing;

[0117] Observe and record psychological states: During the painting process, psychologists need to record the type of emotions, frequency and intensity of changes in detail, analyze the changes in facial expressions such as frowning, smiling and sighing; record the content of speech, emotional color, accuracy and speed of answers, use recording equipment to record conversations, and then transcribe and analyze; record hand movements, posture adjustments and micro-expressions, and use high-resolution cameras to record the entire painting process to ensure that no details are missed;

[0118] Data collection tools: High-resolution camera: records the painting process, including hand movements and facial expressions, to ensure clear video quality and capture subtle changes in facial expressions; Heart rate monitor: monitors heart rate changes in real time during the painting process to reflect physiological stress levels; Skin conductance meter: records skin electrical responses as an indicator of physiological stress response;

[0119] Step 2: Data preprocessing

[0120] Denoising ensures the accuracy of subsequent feature extraction and analysis in tree painting data analysis; image denoising refers to reducing noise in an image through various filtering methods to improve image clarity; in the present invention, three methods, namely, mean filtering, median filtering and Gaussian filtering, are used;

[0121] Mean filtering initially removes random noise from the image and smoothes the image, laying the foundation for subsequent filtering. Let I be the original image, I′ be the filtered image, and N(i,j) be the neighborhood of pixel (i,j). Then we have

[0122] Where I: original image; I′: filtered image; N(i,j): neighborhood of pixel (i,j); k: horizontal coordinate of pixel in neighborhood; l: vertical coordinate of pixel in neighborhood; ∑: sum operation of all pixel values ​​in neighborhood; |N(i,j)|: number of pixels in neighborhood; i: horizontal coordinate in image; j: vertical coordinate in image;

[0123] Median filtering effectively removes salt and pepper noise, retains edge details, and is suitable for discrete noise. Let I be the original image, I′ be the filtered image, and N(i,j) be the neighborhood of pixel (i,j). Then I′(i,j)=median{I(k,l)|(k,l)∈N(i,j)}

[0124] Where, I: original image; I′: filtered image; N(i,j): neighborhood of pixel (i,j); (k,l): pixel coordinates in the neighborhood; median: median operation; i: horizontal coordinate in the image; j: vertical coordinate in the image;

[0125] Gaussian filtering smoothes the image, retaining image details and structure; at the same time, it maintains the overall structure of the image, making subsequent feature extraction more accurate. Let I be the original image, I′ be the filtered image, and G be the Gaussian kernel, then

[0126] Where, I: pixel value matrix in the original image; I′: pixel value matrix in the filtered image; G: Gaussian kernel matrix; (i, j): coordinates of the current pixel; (k, l): pixel coordinates within the Gaussian kernel; ∑: summation operation of all pixels in the convolution window; i: horizontal coordinate in the image; j: vertical coordinate in the image;

[0127] The Gaussian kernel G is defined as

[0128] Among them, G(x,y): the value of the Gaussian kernel at the coordinate (x,y); σ: the standard deviation of the Gaussian kernel, which controls the degree of smoothness; x: the horizontal coordinate in the Gaussian kernel matrix; y: the vertical coordinate in the Gaussian kernel matrix; exp: exponential function;

[0129] Noise detection is used to accurately locate the noisy areas in the image, so as to perform targeted denoising to ensure that the details of the tree painting are preserved; through the noise detection algorithm, the noisy areas in the image are identified and removed, thereby improving the overall quality of the image; let I be the original image, N(i,j) be the neighborhood of pixel (i,j), then the local variance is defined as

[0130] Where, I: original image; N(i,j): neighborhood of pixel (i,j); σ 2(i, j): local variance of pixel (i, j); μ(i, j): mean value in the neighborhood; (k, l): pixel coordinates in the neighborhood; |N(i, j)|: number of pixels in the neighborhood; ∑: summation operation of all pixels in the neighborhood; i: horizontal coordinate in the image; j: vertical coordinate in the image;

[0131] Grayscale conversion ensures that the details and structure of the tree painting are preserved in the grayscale image; this helps to improve the accuracy and consistency of tree painting feature extraction; grayscale conversion converts RGB color images into grayscale images, using the brightness method to ensure quality, simplify data, and retain brightness information; let R, G, B be the red, green, and blue components of the RGB image, respectively, and I be the grayscale image, then I = 0.299R + 0.587G + 0.114B

[0132] Among them, R: red component; G: green component; B: blue component; I: grayscale image;

[0133] Contrast adjustment enhances contrast by stretching the grayscale value range to ensure that grayscale image details are not lost; by adjusting the contrast, the image recognizability and analysis accuracy are improved, which is crucial for feature extraction and recognition in tree paintings; let I be the grayscale image, I′ be the image after contrast adjustment, and the minimum grayscale value be

[0134] Where, I: grayscale image; I′: image after contrast adjustment; I min : The minimum gray value in the image; I max : The maximum gray value in the image;

[0135] Image enhancement: Use histogram equalization technology to improve image contrast and make image details clearer; improve image contrast, make feature extraction more accurate, and enhance image processing effect; suppose the grayscale value of the grayscale image is I, the grayscale value after histogram equalization is I′, and the cumulative distribution function is CDF, then

[0136] Where, I: grayscale value of grayscale image; I′: grayscale value after histogram equalization; CDF: cumulative distribution function; CDF min : The minimum non-zero value of the cumulative distribution function; i: the horizontal coordinate in the image; j: the vertical coordinate in the image;

[0137] Brightness adjustment: adjust according to the overall brightness level of the image, ensure that the overall brightness of the image is moderate, improve the visual effect of the image, and adjust the overall brightness of the image through linear transformation; brightness adjustment makes the overall visual effect of the image more in line with the perception habits of the human eye; let I be the grayscale image, I′ be the image after adjusting the brightness, then I′(i,j)=I(i,j)+ΔL

[0138] Where, I: grayscale image; I′: image after brightness adjustment; ΔL is the brightness adjustment amount;

[0139] Sharpening filter: Sharpening filter improves the sharpness of the image and enhances edge details. Sharpening filter ensures the edge of the tree painting image is clear and improves the accuracy of feature extraction. It improves sharpness by enhancing the contrast of the image edge. Sharpening filter enhances the edge details of the image. Let I be the grayscale image, I′ be the sharpened image, and Laplace operator be L, then I′(i,j)=I(i,j)-αL(i,j)

[0140] Where, I: grayscale image; I′: sharpened image; L: Laplace operator; α is the sharpening intensity coefficient; i: horizontal coordinate in the image; j: vertical coordinate in the image;

[0141] Synchronization of psychological state data: Matching the timestamps of psychological state data and image data to ensure synchronization and fusion to generate a comprehensive data file for subsequent analysis and processing;

[0142] Timestamp matching is used to ensure the temporal consistency of data; let the timestamp of the image data be T img , the timestamp of the mental state data is T psy , then the matching condition is T img ≈T psy

[0143] Data fusion fuses the matched psychological state data with the image data to generate a comprehensive data file in a standardized JSON format;

[0144] Step 3: Feature extraction

[0145] Feature extraction is the core step to extract key information from pre-processed images and psychological state data. Meaningful feature information is systematically extracted from tree paintings and combined with psychological state data to provide a scientific basis for psychological assessment and improve assessment accuracy.

[0146] Edge detection extracts edge information from tree paintings, helps identify structural features, and reflects the painter's concentration and hand control. By optimizing the Canny algorithm, the accuracy of edge detection and noise reduction are improved to ensure detail clarity.

[0147] Canny edge detection is used to extract edges in tree paintings. It detects edges by calculating the intensity gradient of the image, ensuring high accuracy and low noise in edge detection. It includes five steps: noise removal, gradient calculation, non-maximum suppression, dual threshold detection, and edge tracking.

[0148] Non-maximum suppression: retain the local maximum points of the gradient; used to enhance the accuracy of the edge;

[0149] Dual threshold detection: Use high and low thresholds to classify gradient magnitudes, connect strong edges, and retain potential edges;

[0150] Morphological processing removes noise through dilation, erosion, opening and closing operations, highlights the main structure of the tree, and ensures that the extracted image features are accurate and reliable; further optimizes the results of edge detection, removes noise, and highlights the main structure; it is used to ensure that the extracted image features are accurate and reliable, not affected by noise, and enhances the effect of subsequent analysis;

[0151] Region segmentation: Use the watershed algorithm to segment the image and accurately extract the various regions of the tree; provide detailed structural information; analyze the overall morphological characteristics, evaluate the spatial perception and detail grasping ability; help evaluate the painter's spatial perception and detail grasping ability; and contribute to the comprehensive assessment of the psychological state;

[0152] Gradient calculation: Calculate the gradient of the image through morphological gradient; used to identify the area with strong edges in the image; G = morpho log yEx (I, MORPH_GRADIENT, K)

[0153] Among them, G: gradient map, an image used to represent the edge strength in an image; I: input image, that is, the original image that needs to be gradient calculated; K: structural element, a template used to define morphological operations; morphologyEx: OpenCV function, here, it performs the MORPH_GRADIENT operation, which generates a gradient map by calculating the difference between image dilation and erosion; this operation emphasizes the edge area in the image and is often used for edge detection;

[0154] Threshold segmentation: Use threshold segmentation to generate a label map for marking different areas in the image; _,marker s=threshold(G,255,THRESH_BINARY+THRESH_OTSU)

[0155] Among them, _: This symbol indicates that part of the return value is not used; markers: marker map, used to mark different areas in the image; G: gradient map, that is, the image generated in the previous part; the image after gradient calculation is used as input for threshold segmentation; 0: threshold lower limit, indicating that pixels less than this value will be processed as background; 255: threshold upper limit, indicating that pixels greater than this value will be processed as foreground; THRESH_BINARY: binary flag, indicating that the image pixels are converted to binary images, that is, all pixels are either 0 or 255; THRESH_OTSU: Otsu's method flag, indicating that the optimal threshold is automatically determined to segment the foreground and background;

[0156] Watershed algorithm: perform watershed segmentation on the marker map; generate marker values ​​for different areas; mark s = watershed(I, mark s)

[0157] Among them, markers: the marker map processed by the watershed algorithm; after the watershed algorithm is processed, markers will contain different integer values, which correspond to different segmented regions in the image; the watershed algorithm uses this marker map to segment the image, and the pixels in each region are assigned different marker values; I: input image, the original image used for the watershed algorithm; the watershed algorithm operates on this image and segments the regions in the image in combination with the marker map; watershed: the watershed function in OpenCV; it receives the input image and the marker map, and returns the updated marker map, where each marker corresponds to a segmented region;

[0158] Psychological feature extraction analyzes the painting speed, pressure changes and emotional fluctuations during the painting process through time records and image changes, and comprehensively evaluates the mental health status;

[0159] Step 4: Classification and prediction

[0160] Classification and prediction are the key links to deeply analyze and make accurate judgments from the rich information extracted from tree painting characteristics and psychological state data; in this process, the latest intelligent algorithms and advanced model training technologies are innovatively combined to achieve accurate classification and prediction of psychological states;

[0161] Intelligent algorithm selection: In order to deeply analyze the complex relationship between the structural features of tree paintings and psychological state data, a unique combination method is adopted, namely the algorithm combining cross-modal contrastive learning and graph neural architecture search; this innovative combination not only enhances the classification effect, but is also particularly suitable for processing complex multimodal data sets, greatly improving the accuracy of classification; through the cross-modal contrastive learning method, the image features of tree paintings and psychological state data are mapped into a unified feature space; this step ensures the consistency between the features of different modal data, thus laying the foundation for subsequent multimodal data fusion; during the model training process, the image features of tree paintings and psychological state data are input into the deep neural network and the graph neural network respectively, and multiple feature-level fusions are performed in the fusion layer; through the innovative hierarchical fusion strategy, the model can better understand the complex relationship of the data and optimize it through the cross-entropy loss function;

[0162] Graph Neural Architecture Search (GNAS): Graph Neural Architecture Search is a cutting-edge algorithm for automatically designing the optimal architecture of graph neural networks (GNNs). It uses the GNAS method to automatically design and optimize the GNN architecture suitable for tree painting features. By dynamically adjusting the network structure, it ensures that the model can fully capture the complex relationships in the data and ultimately achieve the best classification effect.

[0163] Model training: During the model training phase, an innovative multimodal fusion model was used for multi-category classification training. By combining the image features of tree paintings with psychological state data, the model was able to understand and classify different categories of data more comprehensively and accurately. In addition, the use of data augmentation technology greatly increased the diversity of training data, further improving the generalization ability and robustness of the model.

[0164] In order to further improve the generalization ability of the model, data augmentation technology was introduced; that is, more diverse training samples were generated by randomly rotating the image data and adding noise.

[0165] For image data, the transformation formula for data augmentation is: x′=transform(x)

[0166] Where: x′ is the transformed image data; transform(x) is the random transformation operation used for image data, including rotation, translation, and flipping;

[0167] Model validation and optimization: In the model validation phase, an advanced k-fold cross-validation method is used to evaluate the performance of the model, and the model parameters are optimized by the improved whale optimization algorithm. This innovative method combination not only improves the accuracy of classification, but also ensures the stable performance of the model on different data sets. In the model validation phase, the data set is divided into k subsets D1, D2, …, D k , then the accuracy calculation formula for the i-th verification is:

[0168] Among them: A i is the accuracy of the i-th verification; A i is the model obtained from the i-th training is to use the D i The prediction result of the model trained with the remaining data except for x for sample x; y is the true label of the sample; 1 is the indicator function, which takes 1 when the prediction is correct and 0 otherwise;

[0169] In order to further optimize the model, the improved WOA algorithm is used to dynamically adjust hyperparameters such as learning rate and regularization coefficient so that the model can achieve the best performance in classification tasks.

[0170] Step 5: Result evaluation

[0171] Result evaluation is a key step in the classification and prediction of psychological states. The evaluation model predicts the results and generates a detailed evaluation report. The report combines a variety of data visualization technologies and analysis methods to accurately display the results of clinical significance. After the classification and prediction are completed, the system generates a comprehensive report, including individual psychological state analysis, depression assessment, and emotional fluctuations during the painting process. Through images, charts, and text descriptions, the report intuitively presents the evaluation results, including psychological state analysis: based on the classification and prediction results, comprehensively analyze the individual's psychological state, identify problems such as anxiety and depression, and conduct quantitative evaluation in combination with clinical scales; depression assessment: integrate the results of psychological state analysis, apply standardized depression assessment models, provide scores, and provide reference for clinicians; emotional fluctuation analysis: by analyzing the emotional dynamics during the painting process, generate emotional fluctuation charts, identify emotional changes in different time periods, and provide real-time monitoring data for clinical intervention;

[0172] Step 6: Apply the results

[0173] Result application is the final stage of psychological state classification and prediction, and the results of the assessment report are applied to psychological intervention and management; it includes personalized psychological intervention, continuous monitoring and evaluation, and mental health support. Through these measures, individuals can effectively manage and improve their mental health and improve their overall quality of life; Personalized psychological intervention: According to the assessment report, psychologists develop customized cognitive behavioral therapy (CBT), art therapy, and group therapy plans, and combine virtual reality (VR) technology and artificial intelligence-assisted tools to improve the accuracy and effectiveness of intervention; Continuous monitoring and evaluation: Regular tree painting tests and psychological state records are conducted, combined with digital health devices and big data analysis, and intervention measures are dynamically adjusted to ensure sustained treatment effects; Mental health support: Based on personalized intervention and continuous monitoring, continuous psychological counseling services and health education are provided, and through online platforms and mobile applications, individuals are helped to maintain daily mental health and prevent future psychological distress.

[0174] Specific numerical analysis

[0175] Step 1: Data collection and psychological state analysis

[0176] Input: a. Mental state assessment task: Provide drawing task instructions, tools and drawing environment. Individual physiological data: Real-time physiological data collected by heart rate monitor and skin conductance meter. b. Drawing process record: Including the drawing process, emotional changes, verbal reactions, body language, etc. recorded by the camera.

[0177] Output: a Standardized painting data: the painting image after denoising. b Comprehensive psychological status report: a report generated by combining physiological data, painting data and psychologist's observation records.

[0178] Implementation case study:

[0179] Data collection: a Explanation of the drawing task: The psychologist explained the purpose of the task: "Please draw a tree on A4 paper, as detailed as possible." b Material preparation: Use A4 paper, HB pencil, eraser, and record the number of revisions as 3 times. c Environmental setting: Conducted in a quiet, softly lit room. d Physiological data collection: Heart rate monitoring: Initial heart rate 72 beats / minute, rising to 85 beats / minute in the middle, and finally falling back to 74 beats / minute. Skin conductance: Initial value 2.5μS, peak value 3.8μS, e Psychological state recording: Emotional changes: Frowning 3 times, sighing 2 times. Verbal response: Talking to oneself: "This part is not easy to draw", answering questions slower. Body language: Hand tremor 2 times, adjusting sitting posture 4 times. f Data collection tools: High-resolution camera: Record the drawing process (1080p), capture hand tremors and facial expression changes. Heart rate monitor and skin conductance meter: Data is uploaded to the system in real time, synchronized with the video.

[0180] Step 2: Data preprocessing

[0181] a Denoising: Mean filter: preliminary denoising, neighborhood size 3x3. Median filter: further denoising, retaining edge details. Gaussian filter: final denoising, using a Gaussian kernel with a standard deviation of 1.0 to smooth the image. b Noise detection and processing: identifying and processing high-frequency noise areas. c Grayscale conversion and contrast adjustment: converting RGB images to grayscale images, adjusting contrast to improve visual recognition.

[0182] Step 3: Feature extraction

[0183] Input: a Preprocessed image data: tree painting image after denoising and grayscale conversion. b Synchronized psychological state data: heart rate, skin conductance, emotional changes, verbal response, etc.

[0184] Output: a Image features: edge information, region segmentation data, drawing speed, pressure changes, etc. b Psychological features: emotional fluctuations, changes in psychological state, etc.

[0185] Implementation case analysis: a Edge detection and morphological processing: Canny edge detection: low threshold 50, high threshold 150, detect the edges of structures such as trunks and branches. b Morphological processing: Dilation: fill the gaps between branches, using a 3x3 matrix. Erosion: enhance edge clarity, using a 3x3 matrix. c Region segmentation: Watershed algorithm: extract regions such as trunks, branches, and crowns. Gradient calculation: use morphological gradients to calculate image gradients. Marker generation and watershed segmentation: generate a marker map and perform segmentation. d Psychological feature extraction: Drawing speed: calculate the drawing speed for each time period. Pressure changes: calculate heart rate and skin conductance changes. Emotional fluctuations: analyze facial expression changes and frequency.

[0186] Step 4: Classification and prediction

[0187] Input: extracted images and psychological features: including edge information, region segmentation results, drawing speed, pressure changes, emotional fluctuations, etc. Training data set: annotated psychological states and corresponding feature data.

[0188] 1) The thickness of the line reflects the user's emotional intensity. A thicker line indicates tension or anxiety (the score range is 0-1 from thin to thick); in addition, a higher modification frequency indicates that the user may be hesitant or nervous when expressing himself (the score range is 0-1 from low to high frequency); the weight ratio is 0.1;

[0189] 2) During the painting process, the system will record the position of the pen tip in each time period and calculate the painting speed through the change of the pen tip position. Fast painting expresses passion, tension or excitement, while slow painting is more suitable for expressing contemplation, introspection or delicate emotions; weight ratio 0.1;

[0190] 3) The system will record the user's pen pressure at each time point and quantify the change in pen pressure through variance. The larger the variance, the greater the user's emotional fluctuations and the stronger the emotion (the score range is 0-1 according to the degree of pen pressure variance); the weight ratio is 0.1;

[0191] Classification is carried out according to the Big Five personality dimensions and the drawing projective test index theory or other drawing projective theories. Since the drawing projective theory is a well-known psychological measurement method, it is not the focus of the present invention. The present invention focuses on the classification method of constructing an algorithm. Taking drawing trees as an example, the psychological state of talents is analyzed and classified into the following types:

[0192] Important clues about mood and stress can be obtained by analyzing the content of the paintings (the tree's trunk, crown, leaves, fruit, roots and ground line, as well as non-subject information), line intensity, and symbols in the drawings.

[0193] Emotional level: The emotions expressed when drawing the tree's trunk, crown, leaves, fruit, roots, and ground line, as well as non-thematic information (closed crown, top of the crown cut, etc.).

[0194] 4) Anxiety and stress: The picture shows many black spots, messy and fragmented lines, and the main body of the picture is too small. The degree of the above painting characteristics is set on a scale of 0-1, and the more the degree, the greater the value; the weight ratio is 0.3;

[0195] 5) Depressed state: light brushstroke pressure, stiff brushstrokes, poor fluency, and a snowy scene; the degree of the above painting characteristics is set to a range of 0-1, with the greater the degree, the greater the value; the weight ratio is 0.4;

[0196] Output: Classification results: Individual psychological state classification, depression level is divided into intervals of 0-1, including severe depression (0.7-1), moderate depression (0.4-0.6), and mild depression (0-0.3). Prediction results: It is also necessary to combine heart rate, skin conductance, emotional changes, and verbal response data to comprehensively judge the future development trend of the individual's psychological state.

[0197] Implementation case analysis: a Intelligent algorithm selection: Cross-modal contrastive learning and GNAS algorithm: After multi-modal feature data is input, fusion features are output, alignment is optimized, and classification accuracy is improved. b Model training: Multi-modal fusion model training: Fusion of images and psychological features to generate comprehensive feature representation. c Model verification and optimization: k-fold cross validation: average accuracy of 88%. Whale Optimization Algorithm (WOA): After optimization, the accuracy rate is increased to 91%.

[0198] Step 5: Result evaluation

[0199] Input: Classification and prediction results: individual psychological state classification and prediction results. Extracted feature data: emotional fluctuations, stress changes, etc. during the painting process.

[0200] Output: Assessment report: Contains detailed information such as psychological state analysis, depression assessment, and mood swing analysis.

[0201] Implementation case analysis: a. Psychological state analysis: Generate radar charts to show dimensions such as anxiety, depression, and self-confidence. b. Depression assessment: Generate a bar chart of depression levels. c. Emotional fluctuation analysis during painting: Generate a time series chart of emotional changes.

[0202] Step 6: Apply the results

[0203] Input: Assessment report: Contains the individual's psychological state, depression assessment results, and mood swing analysis.

[0204] Output: Personalized psychological intervention program: cognitive behavioral therapy, art therapy, etc. Continuous monitoring and evaluation strategy: regular drawing tests and psychological state records. Mental health support plan: online psychological counseling and education.

[0205] Implementation case analysis: a Personalized psychological intervention: Develop CBT and other intervention programs based on the assessment report data, with the goal of reducing the number of anxiety attacks. b Continuous monitoring and evaluation: Regularly adjust the intervention strategy, with a monitoring interval of two weeks and an adjustment frequency of once a month. c Mental health support: Provide monthly online consultation and education, covering emotion regulation and mindfulness practice.

[0206] The present invention realizes quantitative analysis of individual tree painting characteristics through systematic data processing and intelligent algorithm analysis, thereby providing objective and accurate depression assessment and helping to improve mental health. Although the present invention is described in detail in the embodiments, those skilled in the art may modify or replace the embodiments without departing from the spirit and principles of the present invention, and all modifications and improvements within the scope of protection of the present invention should be included in the scope of protection of the present invention.

[0207] Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments, or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A tree painting projection data collection and analysis system, including a system body, characterized in that: The system includes a data collection module, a data preprocessing module, a feature extraction module, a quantitative analysis module, and a classification and prediction module, which are connected in sequence. The system analyzes the painting features by combining the holographic theory and Armas's pit theory, and using the cross-modal contrastive learning and graph neural architecture search algorithm modules. The data collection module collects the tree images drawn by individuals through a scanner and transmits them to the system storage unit; the module includes an image acquisition unit and a data transmission unit to ensure the clarity and integrity of the image and supports wired and wireless transmission methods; The data preprocessing module performs standardization processing on the collected tree painting projections, including image denoising, grayscale conversion and image enhancement; Median filtering is used for denoising, brightness method is used for grayscale conversion, and Laplace operator enhancement technology is used to improve image contrast and clarity; The feature extraction module extracts the quantitative features of trees, including overall height, width, trunk width, and crown area, from the processed tree drawings; these features are extracted through edge detection, morphological processing, and region segmentation algorithms, and normalized to ensure the accuracy and stability of the data; The quantitative analysis module quantifies the extracted features and generates feature vectors for analysis to facilitate subsequent classification and prediction; The classification and prediction module is based on quantized feature vectors, performs multi-category classification training through cross-modal contrast learning and graph neural architecture search algorithm, establishes a classification model, and performs model verification and optimization to improve classification performance.

2. A tree painting projection data collection and analysis system according to claim 1, characterized in that: It also includes an outcome assessment module, which provides specific auxiliary diagnostic suggestions by generating an assessment report, including an analysis of the individual's psychological state and depression assessment results.

3. A method for a tree painting projection data collection and analysis system, characterized in that: The following steps are involved: Step 1, collecting individual psychological state information, and the data collection module collecting tree drawings drawn by the individual, and collecting images of the tree drawings drawn by the individual through a scanner; Step 2, the data preprocessing module performs standardized preprocessing on the collected tree painting projections, including median filtering for image denoising, grayscale conversion and image enhancement to eliminate image noise and improve image quality; Step 3, the feature extraction module extracts multiple quantitative features from the processed tree drawings, including the overall height, width, trunk width, crown area, number of branches, and leaf density of the tree; Step 4: The quantitative analysis module quantifies the extracted features and generates feature vectors for analysis; Step 5: The classification and prediction module classifies and predicts the psychological state based on the quantized feature vectors by combining cross-modal contrastive learning with graph neural architecture search.

4. The method according to claim 3, characterized in that: It also includes step 6, in which the result evaluation module evaluates the results of classification and prediction and outputs an analysis report, which includes the individual's psychological state analysis and depression evaluation results. By evaluating the classification and prediction results, a detailed evaluation report is generated for clinical use.

5. The method according to claim 3, characterized in that: Step 1 includes: Professional psychologists set up the following settings: using simple and clear language to explain the purpose and requirements of the drawing task to the individual in detail and assess their psychological state; providing standard A4 white paper, HB pencils and erasers, allowing the individual to make revisions, but recording the frequency of revisions to reflect their hesitation and decision-making process and obtain additional psychological state information; choosing a quiet and comfortable room to avoid noise and interference, ensuring soft and adequate lighting, and avoiding strong light and shadows that affect the quality of the drawing; Observe and record psychological states: During the painting process, psychologists need to record the type of emotions, frequency and intensity of changes in detail, analyze the changes in facial expressions such as frowning, smiling and sighing; record the content of speech, emotional color, accuracy and speed of answers, use recording equipment to record conversations, and then transcribe and analyze; record hand movements, posture adjustments and micro-expressions, and use high-resolution cameras to record the entire painting process to ensure that no details are missed; Data collection tools: High-resolution camera: records the painting process, including hand movements and facial expressions, to ensure clear video quality and capture subtle changes in facial expressions; heart rate monitor: monitors heart rate changes in real time during the painting process to reflect physiological stress levels; skin conductance meter: records skin electrical responses as an indicator of physiological stress response.

6. The method according to claim 3, characterized in that: In step 2: The median filter module is used to effectively remove salt and pepper noise, retain edge details, and is suitable for discrete noise; let I be the original image, I′ be the filtered image, and N(i,j) be the neighborhood of pixel (i,j), then I′(i,j)=median{I(k,l)|(k,l)∈N(i,j)}; Where, I: original image; I′: filtered image; N(i,j): neighborhood of pixel (i,j); (k,l): pixel coordinates in the neighborhood; median: median operation; i: horizontal coordinate in the image; j: vertical coordinate in the image; The noise detection module is used to accurately locate the noise area in the image, so as to perform targeted denoising to ensure that the details of the tree painting are preserved; the noise area in the image is identified and removed through the noise detection algorithm, thereby improving the overall quality of the image; let I be the original image, N(i,j) be the neighborhood of pixel (i,j), then the local variance is defined as Where, I: original image; N(i,j): neighborhood of pixel (i,j); σ 2 (i, j): local variance of pixel (i, j); μ(i, j): mean value in the neighborhood; (k, l): pixel coordinates in the neighborhood; |N(i, j)|: number of pixels in the neighborhood; A grayscale conversion module is used to ensure that the details and structure of the tree painting are preserved in the grayscale image. Grayscale conversion converts the RGB color image into a grayscale image, using the brightness method to ensure quality, simplify data, and retain brightness information. Let R, G, and B be the red, green, and blue components of the RGB image, respectively, and I be the grayscale image, then I = 0.299R + 0.587G + 0.114B. Among them, R: red component; G: green component; B: blue component; I: grayscale image; The contrast adjustment module is used to enhance the contrast by stretching the grayscale value range to ensure that the details of the grayscale image are not lost; by adjusting the contrast, the image recognizability and analysis accuracy are improved, which is crucial for the extraction and recognition of features in tree paintings; let I be the grayscale image, I′ be the image after contrast adjustment, and the minimum grayscale value be Where, I: grayscale image; I′: image after contrast adjustment; I min : The minimum gray value in the image; I max : The maximum gray value in the image; Image enhancement module: Use histogram equalization technology to improve image contrast and make image details clearer; improve image contrast, make feature extraction more accurate, and enhance image processing effect; suppose the gray value of the grayscale image is I, the gray value after histogram equalization is I′, and the cumulative distribution function is CDF, then Where, I: grayscale value of grayscale image; I′: grayscale value after histogram equalization; CDF: cumulative distribution function; CDF min : The minimum non-zero value of the cumulative distribution function; i: the horizontal coordinate in the image; j: the vertical coordinate in the image; Brightness adjustment module: adjust the overall brightness level of the image, ensure that the overall brightness of the image is moderate, improve the visual effect of the image, and adjust the overall brightness of the image through linear transformation; brightness adjustment makes the overall visual effect of the image more in line with the perception habits of the human eye; let I be the grayscale image, I′ be the image after brightness adjustment, then I′(i,j)=I(i,j)+ΔL, where I: grayscale image; I′: image after brightness adjustment; ΔL is the brightness adjustment amount; Sharpening filter module: Sharpening filter improves the sharpness of the image and enhances the edge details. Sharpening filter ensures the edge of the tree painting image is clear and improves the accuracy of feature extraction. Sharpness is improved by enhancing the contrast of the image edge. Sharpening filter enhances the edge details of the image. Let I be the grayscale image, I′ be the sharpened image, and Laplace operator be L. Then I′(i,j)=I(i,j)-αL(i,j), where I: grayscale image; I′: sharpened image; L: Laplace operator; α is the sharpening intensity coefficient; i: horizontal coordinate in the image; j: vertical coordinate in the image. Psychological state data synchronization module: matches the timestamps of psychological state data and image data to ensure synchronization and merge them to generate a comprehensive data file for subsequent analysis and processing; The timestamp matching module is used to ensure the consistency of data in time; let the timestamp of the image data be T img , the timestamp of the mental state data is T psy , then the matching condition is T img ≈T psy ; Data fusion fuses the matched psychological state data with the image data to generate a comprehensive data file in a standardized JSON format.

7. The method according to claim 3, characterized in that: The step 3 comprises: The edge detection module extracts edge information from tree paintings to help identify structural features and reflect the painter's concentration and hand control. The Canny algorithm is optimized to improve the accuracy of edge detection and noise reduction, ensuring detail clarity. Canny edge detection is used to extract edges in tree paintings. It detects edges by calculating the intensity gradient of the image, ensuring high accuracy and low noise in edge detection. It includes five steps: noise removal, gradient calculation, non-maximum suppression, dual threshold detection, and edge tracking. Among them, I(x,y): original image, which means the pixel value of the original image at position (x,y); I′(x,y): filtered image, which means the pixel value of the filtered image at position (x,y); G(x,y): Gaussian kernel; x: horizontal coordinate in the image; y: vertical coordinate in the image; σ: standard deviation of Gaussian filter, which determines the degree of smoothing; the larger the value, the stronger the smoothing effect; Image gradient: Calculate image gradient for edge detection; Among them, G x and G y Represent the gradients in the x and y directions respectively; G: Gradient magnitude, which indicates the gradient magnitude of the image and reflects the edge strength; Non-maximum suppression: retain the local maximum points of the gradient to enhance the accuracy of the edge; Dual threshold detection: Use high and low thresholds to classify gradient magnitudes, connect strong edges, and retain potential edges; Morphological processing: remove noise through dilation, erosion, opening and closing operations, highlight the main structure of the tree, and ensure that the extracted image features are accurate and reliable; further optimize the results of edge detection, remove noise, and highlight the main structure; used to ensure that the extracted image features are accurate and reliable, not affected by noise, and enhance the effect of subsequent analysis; Expansion Operation I dilated (x,y)=max (i,j)∈K I(x+i,y+j) Corrosion Operation I eroded (x,y)=min (i,j)∈K I(x+i,y+j) Opening operation: first erosion and then expansion I opened (x,y)=dilate(erode(I,K),K) Closing operation: dilation followed by corrosion I closed (x,y)=erode(dilate(I,K),K) Among them, dilate(I,K) means dilating images I and K to obtain a new image; erode(I,K) means eroding images I and K to obtain a new image; I dilated (x, y): pixel value of the expanded image at position (x, y) K: structural element, used to define the range of the operation; max: take the maximum value in the local area to expand the object; I eroded (x, y): pixel value of the image after corrosion at position (x, y); min: take the minimum value in the local area to shrink the object; I opened : Image after opening operation; I closed : The image after closing operation; Region segmentation: Use the watershed algorithm to segment the image and accurately extract the various regions of the tree; provide detailed structural information; analyze the overall morphological characteristics, evaluate the spatial perception and detail grasping ability; help evaluate the painter's spatial perception and detail grasping ability; and contribute to the comprehensive assessment of the psychological state; Gradient calculation: Calculate the gradient of the image through morphological gradient; used to identify the area with strong edges in the image; G = morphologyEx (I, MORPH_GRADIENT, K); Among them, G: gradient map, an image used to represent the edge strength in an image; I: input image, that is, the original image that needs to be gradient calculated; K: structural element, a template used to define morphological operations; morphologyEx: OpenCV function, here, it performs the MORPH_GRADIENT operation, which generates a gradient map by calculating the difference between image dilation and erosion; this operation emphasizes the edge area in the image and is often used for edge detection; Threshold segmentation: Use threshold segmentation to generate a marker map to mark different areas in the image; _, markers = threshold (G, 255, THRESH_BINARY + THRESH_OTSU) Among them, _: This symbol indicates that part of the return value is not used; markers: marker map, used to mark different areas in the image; G: gradient map, that is, the image generated in the previous part; the image after gradient calculation is used as input for threshold segmentation; 0: threshold lower limit, indicating that pixels less than this value will be processed as background; 255: threshold upper limit, indicating that pixels greater than this value will be processed as foreground; THRESH_BINARY: binary flag, indicating that the image pixels are converted to binary images, that is, all pixels are either 0 or 255; THRESH_OTSU: Otsu's method flag, indicating that the optimal threshold is automatically determined to segment the foreground and background; Watershed algorithm: perform watershed segmentation on the marker map; generate marker values ​​for different areas; markers = watershed(I, markers); Among them, markers: the marker map processed by the watershed algorithm; after the watershed algorithm is processed, markers will contain different integer values, which correspond to different segmented regions in the image; the watershed algorithm uses this marker map to segment the image, and the pixels in each region are assigned different marker values; I: input image, the original image used for the watershed algorithm; the watershed algorithm operates on this image and segments the regions in the image in combination with the marker map; watershed: the watershed function in OpenCV; it receives the input image and the marker map, and returns the updated marker map, where each marker corresponds to a segmented region; Psychological feature extraction: Through time records and image changes, the painting speed, pressure changes and emotional fluctuations during the painting process are analyzed to comprehensively evaluate the mental health status; Drawing speed: by calculating the completed part in each time period during the drawing process; Among them, drawingspeed i : Drawing speed in the i-th time period; drawing_speed: drawing speed; S i : the part completed in the i-th time period; t i : i-th time point; Pressure change: by calculating the pressure change at each time period during the painting process; pressure change i =P i+1 -P i Among them, pressure change i : Pressure change in the i-th time period; pressure_change: pressure change; P i : Pressure data at the i-th time point; Emotional fluctuations: by recording the emotional changes at each time period during the painting process; Among them, emotion change i :emotional change in the i-th time period emotion_change: emotional change; E i : Emotion data at the i-th time point.

8. The method according to claim 3, characterized in that: The step 5 comprises: The classification and prediction module is the key link to conduct in-depth analysis and make accurate judgments from the rich information extracted from tree painting characteristics and psychological state data; specifically: Intelligent algorithm module selection: In order to deeply analyze the complex relationship between the structural features of tree paintings and psychological state data, an algorithm module combining cross-modal contrastive learning and graph neural architecture search was adopted; through the cross-modal contrastive learning method, the image features of tree paintings and psychological state data were mapped to a unified feature space; this step ensured the consistency between the features of different modal data, thus laying the foundation for subsequent multimodal data fusion; during the model training process, the image features of tree paintings and psychological state data were input into the deep neural network and graph neural network respectively, and multiple feature-level fusions were performed in the fusion layer; through the innovative hierarchical fusion strategy, the model can better understand the complex relationship of the data and optimize it through the cross-entropy loss function; The formula of the contrast loss function is: Where: L: total loss value, which is used to measure the effect of the model in aligning different modal features in the current training batch; the smaller the loss value, the better the model is in bringing similar samples closer and moving away from dissimilar samples; represents the sum of all sample pairs (i, j); the loss function will traverse all possible sample pairs, calculate their contrast loss, and then sum them up to get the total loss value; y ij : The label of the sample pair (i, j); D(z i ,z j ): The distance measure between sample i and sample j in the feature space; this distance can be the Euclidean distance or other forms of distance measure; D(z i ,z j )=||z i -z j || Among them, z i and z j are the feature vector representations of sample i and sample j respectively; 1-y ij : The opposite label of the sample pair (i, j); when the sample pair (i, j) belongs to different categories, y ij =0, then 1-y ij =1, indicating that the sample pair is not similar; max(0,mD(z i ,z j )): This is a penalty term used to ensure that pairs of samples of different classes maintain at least a distance m in the feature space; m: a predefined boundary threshold used to control the minimum distance between samples of different classes; if two samples belong to different categories, the distance between them should be at least m; otherwise, a penalty value is generated, thereby increasing the loss value; Graph Neural Architecture Search Module GNAS: Graph Neural Architecture Search is a cutting-edge algorithm for automatically designing the optimal architecture of graph neural networks (GNNs). GNAS is used to automatically design and optimize GNN architectures suitable for tree painting features. The basic update formula of GNN is: in: is the feature representation of node v at the k+1th layer; is the feature representation of node μ in the kth layer; σ is the activation function, ReLU, used to introduce nonlinearity; W (k) is the weight matrix of the kth layer, which represents the relationship weight between nodes; N(v) is the neighborhood set of node v, which includes other nodes connected to node v; b(k) is the bias term, which is used to adjust the overall offset of the feature; Model training: During the model training phase, an innovative multimodal fusion model was used for multi-category classification training. By combining the image features of tree paintings with psychological state data, the model was able to understand and classify different categories of data more comprehensively and accurately. In addition, the use of data augmentation technology greatly increased the diversity of training data, further improving the generalization ability and robustness of the model. The formula for feature-level fusion is: Among them: z: fused multimodal feature representation; this feature vector fuses data features from different modalities and is the final input of the model for classification, regression or other tasks; f image (x image ): feature representation extracted from image data; x image : Input image data; f image (): Image feature extraction function; this function is usually a deep neural network used to extract useful feature representations from the input image data; f psych (x psych ): feature representation extracted from psychological state data; x psych : Input psychological state data; f psych (): psychological state feature extraction function; this function is a neural network or other form of model used to extract features from the input psychological data; ⊕: feature connection operator; this operation concatenates feature vectors from different modalities to generate a longer feature vector containing multimodal information; The formula for decision-level fusion is: final =α·y image +(1-α)·y psych Where: y final is the final fusion prediction result; y image and psych are the independent prediction results of image and psychological state data respectively; α is the weighting coefficient, which is used to control the influence of each modality on the final decision; The data augmentation technique is introduced, which generates more diverse training samples by randomly rotating the image data and adding noise. For image data, the transformation formula for data augmentation is: x′=transform(x); Where: x′ is the transformed image data; transform(x) is the random transformation operation used for image data, including rotation, translation, and flipping; Model validation and optimization: In the model validation phase, an advanced k-fold cross-validation method is used to evaluate the performance of the model, and the model parameters are optimized by the improved whale optimization algorithm. This innovative method combination not only improves the accuracy of classification, but also ensures the stable performance of the model on different data sets. In the model validation phase, the data set is divided into k subsets D1, D2, …, D k , then the accuracy calculation formula for the i-th verification is: Among them: A i is the accuracy of the i-th verification; D i is the model obtained from the i-th training is to use the D i The prediction result of the model trained with the remaining data except for x for sample x; y is the true label of the sample; 1 is the indicator function, which takes 1 when the prediction is correct and 0 otherwise; In order to further optimize the model, the improved WOA algorithm is used to dynamically adjust the learning rate and regularization coefficient hyperparameters so that the model can achieve the best performance in the classification task; The formula for the circling phase is: Where: X t+1 is the whale position of generation t+1; is the position of the current optimal solution; A and C are parameters that control the intensity and distance of the circle, ensuring that the whale gradually approaches the optimal solution in the global search.

9. The method according to claim 4, characterized in that: The step 6 is specifically as follows: Outcome evaluation is a key step in the classification and prediction of mental states. It evaluates the prediction results of the model and generates a detailed evaluation report. The report combines a variety of data visualization techniques and analysis methods to accurately display the results of clinical significance. After classification and prediction, the system generates a comprehensive report, including individual psychological state analysis, depression assessment, and emotional fluctuations during the painting process; the report intuitively presents the assessment results through images, charts, and text descriptions, including psychological state analysis: based on classification and prediction results, comprehensively analyze the individual's psychological state, identify anxiety and depression problems, and conduct quantitative assessments in combination with clinical scales; depression assessment: integrate psychological state analysis results, apply standardized depression assessment models, provide scores, and provide references for clinicians; Emotional fluctuation analysis: By analyzing the emotional dynamics during the painting process, emotional fluctuation charts are generated to identify emotional changes in different time periods and provide real-time monitoring data for clinical intervention.

Images