Depression recognition method based on huang shuren painting and deep learning, computer device and storage medium

Abstract

The application discloses a depression recognition method based on Fang Shuren's painting and deep learning, a computer device and a storage medium. The depression recognition method based on Fang Shuren's painting and deep learning comprises the following steps: inputting Fang Shuren's painting information into a painting base feature encoder to process and acquire a plurality of group base feature information; inputting the group base feature information into a guiding base feature gate unit to process and acquire group gate information and a base vector; inputting the group gate information and the base vector into an adaptive channel feature fusion network to process and acquire a depression recognition result. The application provides a depression early self-recognition tool based on artificial intelligence deep learning. The tool realizes early recognition of depression without human interaction and strict conditions, and can reduce the risk of privacy information leakage of the recognized person. The application has a wide application condition and can be easily applied to people of different ages. The application is widely applied in the technical field of image processing.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to a method, computer device, and storage medium for depression identification based on house-tree-person drawing and deep learning. Background Technology

[0002] Depression is a mental illness, and its identification is crucial for treatment and medical research. With the development of artificial intelligence (AI) technology, several techniques are now being applied to depression detection. For example, AI can be used to identify various facial expressions mimicked by a subject, or to induce emotions in a subject through video, detecting dynamic changes in facial expressions or characteristics of physiological signals such as electrocardiograms (ECG), electrical skin activity (EDA), and respiratory activity (RSP) to determine if the subject has depression. Alternatively, AI can be used for brain mapping (e.g., magnetic resonance imaging, MRI) to make diagnoses. Some technologies also use AI to analyze a subject's social media content to determine if they have depression.

[0003] The aforementioned technologies generally suffer from drawbacks such as numerous limitations, poor flexibility, and susceptibility to other problems. For example, recognizing imitative facial expressions and dynamic facial expressions induced by video requires subjects to strictly adhere to specific data collection conditions, such as having their faces directly facing the camera. However, most people watch videos in a comfortable manner, and these specific conditions may cause discomfort to the subjects, thus affecting the quality of the data. Collecting physiological signals and recognizing mind maps requires specialized equipment, making it unsuitable for self-testing by the general population and hindering widespread adoption. Using social media content from people with depression to design deep learning methods infringes on privacy, and data breaches could lead to significant social problems. Summary of the Invention

[0004] In view of the technical problems existing in current artificial intelligence-based depression recognition technology, such as numerous usage restrictions, poor flexibility, and easy generation of other problems, the purpose of this invention is to provide a depression recognition method, computer device, and storage medium based on house-tree-person drawing and deep learning.

[0005] On one hand, embodiments of the present invention include a method for identifying depression based on house-tree-person drawing and deep learning, the method comprising the following steps:

[0006] Obtain information on house-tree-person drawing;

[0007] Acquire the painting-based feature encoder, the guidance-based feature gating unit, and the adaptive channel feature fusion network;

[0008] The house-tree-person drawing information is input into the drawing basis feature encoder for processing to obtain multiple group basis feature information output by the drawing basis feature encoder;

[0009] Each of the aforementioned basis feature information is input into the guiding basis feature gating unit for processing, to obtain the group gate information and multiple basis vectors output by the guiding basis feature gating unit; wherein, each of the basis vectors corresponds to one of the aforementioned basis feature information;

[0010] The group gate information and each of the basis vectors are input into the adaptive channel feature fusion network for processing, and the depression recognition result output by the adaptive channel feature fusion network is obtained.

[0011] Furthermore, the acquisition of the painting basis feature encoder, the guidance basis feature gating unit, and the adaptive channel feature fusion network includes:

[0012] Establish a first convolutional layer, at least two residual modules, and at least two grouped residual modules; wherein, the residual modules include multiple second convolutional layers, and the grouped residual modules include multiple grouped convolutional layers;

[0013] The first convolutional layer, each residual module, and each grouped residual module are sequentially connected to obtain the painting base feature encoder.

[0014] Furthermore, the acquisition of the painting basis feature encoder, the guidance basis feature gating unit, and the adaptive channel feature fusion network includes:

[0015] Establish a global average pooling layer, a linear layer, and an activation layer;

[0016] The global average pooling layer, the linear layer, and the activation layer are connected sequentially to obtain the directional base feature gating unit.

[0017] Furthermore, the acquisition of the painting basis feature encoder, the guidance basis feature gating unit, and the adaptive channel feature fusion network includes:

[0018] A graph attention network is established as the adaptive channel feature fusion network.

[0019] Further, the step of inputting each of the group basis feature information into the guiding basis feature gating unit for processing, and obtaining the group gate information and multiple basis vectors output by the guiding basis feature gating unit, includes:

[0020] The global average pooling layer is used to process the basis feature information of each group to obtain the basis vector corresponding to each basis feature information;

[0021] The linear layer is used to process each of the basis vectors to obtain the query vector corresponding to each of the basis vectors.

[0022] Obtain the key vector corresponding to each of the aforementioned basis vectors;

[0023] A query matrix is ​​formed using each of the query vectors, and a key matrix is ​​formed using each of the key vectors;

[0024] Determine the group attention vector based on the query matrix and the key matrix;

[0025] The activation layer is used to process the group attention vector;

[0026] The output of the activation layer is processed according to a preset threshold to obtain the group gate information.

[0027] Further, the step of processing the output of the activation layer according to a preset threshold to obtain the group gate information includes:

[0028] Iterate through each element in the output of the activation layer;

[0029] For any element that has been traversed, compare the element's value with the preset threshold.

[0030] When the value of an element is greater than the preset threshold, the value of the element is assigned to 1; otherwise, the value of the element is assigned to 0.

[0031] Further, the step of inputting the group gate information and each of the basis vectors into the adaptive channel feature fusion network for processing, and obtaining the depression recognition result output by the adaptive channel feature fusion network, includes:

[0032] Based on the group gate information and each of the basis vectors, determine the adjacency matrix;

[0033] The graph attention network is used to perform inter-group feature fusion processing on the adjacency matrix to obtain the depression identification result.

[0034] Further, determining the adjacency matrix based on the group gate information and each of the basis vectors includes:

[0035] Based on the group gate information, each of the basis vectors is filtered to determine the filtered basis vectors and the retained basis vectors;

[0036] The adjacency matrix is ​​obtained by connecting the retained basis vectors.

[0037] On the other hand, embodiments of the present invention also include a computer device, including a memory and a processor, the memory for storing at least one program, and the processor for loading at least one program to execute the depression identification method based on house-tree-person drawing and deep learning in the embodiments.

[0038] On the other hand, embodiments of the present invention also include a computer-readable storage medium storing a processor-executable program, which, when executed by a processor, is used to perform the depression identification method based on house-tree-person drawing and deep learning in the embodiments.

[0039] The beneficial effects of this invention are as follows: The depression identification method based on house-tree-person drawing and deep learning in the embodiments provides an early self-identification tool for depression based on artificial intelligence deep learning. By identifying house-tree-person drawing information, it can make an objective assessment of one's mental state and achieve early identification of depression without the need for human-to-human interaction or strict adherence to rules. Since the house-tree-person drawing information is freely created by the person being identified, the risk of privacy information leakage can be reduced. The application conditions of the depression identification method based on house-tree-person drawing and deep learning in the embodiments are lenient, so it is easy to promote and apply to people of different ages. Attached Figure Description

[0040] Figure 1 This is a schematic diagram illustrating the steps of the depression identification method based on house-tree-person drawing and deep learning in the embodiment.

[0041] Figure 2 This is a schematic diagram illustrating the principle of the depression identification method based on house-tree-person drawing and deep learning in the embodiment.

[0042] Figure 3 This is a schematic diagram illustrating the structure and principle of the painting base feature encoder in the embodiment;

[0043] Figure 4 This is a schematic diagram illustrating the structure and principle of the guiding base feature gating unit in the embodiment;

[0044] Figure 5 This is a schematic diagram illustrating the structure and principle of the adaptive channel feature fusion network in the embodiment. Detailed Implementation

[0045] In this embodiment, refer to Figure 1 The depression identification method based on house-tree-person drawing and deep learning includes the following steps:

[0046] S1. Obtain information on the house-tree-person drawing;

[0047] S2. Obtain the painting-based feature encoder, the guidance-based feature gating unit, and the adaptive channel feature fusion network;

[0048] S3. Input the house-tree-person drawing information into the drawing basis feature encoder for processing, and obtain multiple group basis feature information output by the drawing basis feature encoder;

[0049] S4. Input the basis feature information of each group into the guiding basis feature gating unit for processing, and obtain the group gate information and multiple basis vectors output by the guiding basis feature gating unit; wherein, each basis vector corresponds to a corresponding group basis feature information;

[0050] S5. Input the group gate information and each basis vector into the adaptive channel feature fusion network for processing, and obtain the depression recognition result output by the adaptive channel feature fusion network.

[0051] In this embodiment, steps S1-S5 can all be executed by a computer, that is, all steps in the depression identification method based on house-tree-person drawing and deep learning can be implemented by a computer or other device.

[0052] In step S1, the acquired house-tree-person drawing information can be a hand-drawn picture containing images of a house, tree, and person by the person being identified. Specifically, the person can draw on paper and then obtain the house-tree-person drawing information by scanning the paper, or the person can draw on electronic paper, thereby directly generating the house-tree-person drawing information by collecting the drawing action.

[0053] Specifically, during step S1, multiple house-tree-person (HTP) drawing information can be obtained. The content and size of each HTP drawing are presented by the artist (the person being identified) based on their own emotions. Therefore, in step S1, data augmentation methods such as cropping and scaling, which may damage the drawing itself, are not used for preprocessing. To augment the data without destroying the content of the drawing, only horizontal flipping, vertical flipping, and horizontal-vertical flipping methods can be used to preprocess the HTP drawing information during step S1. Before performing steps S2-S5, all HTP drawing information can be uniformly organized to the same size (e.g., 300x200) and normalized.

[0054] In this embodiment, when performing step S2, which is to obtain the painting base feature encoder, the guidance base feature gating unit, and the adaptive channel feature fusion network, the following steps can be performed:

[0055] S201. Establish a first convolutional layer, at least two residual modules, and at least two grouped residual modules; wherein, the residual modules include multiple second convolutional layers, and the grouped residual modules include multiple grouped convolutional layers;

[0056] S202. Connect the first convolutional layer, each residual module, and each grouped residual module in sequence to obtain the painting base feature encoder;

[0057] S203. Establish a global average pooling layer, a linear layer, and an activation layer;

[0058] S204. Connect the global average pooling layer, the linear layer, and the activation layer in sequence to obtain the guided base feature gating unit.

[0059] S205. Establish a graph attention network as an adaptive channel feature fusion network.

[0060] Steps S201-S205 can construct a depression recognition algorithm architecture (DeHTP, Deep Network for House-Tree-Person), which can identify depression samples from house-tree-person drawing information.

[0061] The structure and principle of DeHTP are as follows: Figure 2 As shown. (Refer to...) Figure 2 DeHTP comprises three modules: DrawingBasis Encoder, Guided Basis Gated Module, and Adaptive Channel Feature Fusion.

[0062] Steps S201-S202 are steps for establishing the pictorial base feature encoder. In this embodiment, the structure and principle of the established pictorial base feature encoder are as follows: Figure 3 As shown. (Refer to...) Figure 3 The painting-based feature encoder consists of a series of convolutional layers. Apart from the first convolutional layer, the remaining convolutional layers form two residual modules and two grouped residual modules. Specifically, the second convolutional layer is a component of the residual modules, and the grouped convolutional layers are components of the grouped residual modules.

[0063] Reference Figure 3 In the painting basis feature encoder established through steps S201-S202, the structure of each residual module is the same as that of the residual block used in the ResNet18 network. The output channels of the two residual modules are 128 and 256, respectively.

[0064] Reference Figure 3 The grouped residual module differs from the residual module in that each second convolutional layer in the residual module can be replaced with a grouped convolutional layer to obtain the grouped residual module.

[0065] Reference Figure 3In the grouped residual module, the channels are divided into k groups. Channels within a group can exchange information, but channels between groups are not allowed to interact. Here, k is a configurable number.

[0066] Reference Figure 3 The painting basis feature encoder established through steps S201-S202 can extract features from the house-tree-person painting information. The output of the painting basis feature encoder is multiple sets of feature maps. In this embodiment, the multiple sets of feature maps output by the painting basis feature encoder are called grouped basis features.

[0067] In this embodiment, by executing steps S201-S202, a painting basis feature encoder is formed by combining the classic residual module and the adjusted grouped residual module. This reduces the number of parameters when extracting grouped basis feature information, saves computational resources, and avoids the problem of overfitting.

[0068] Steps S203-S204 are steps for establishing the guiding base feature gating unit. In this embodiment, the structure and principle of the established guiding base feature gating unit are as follows: Figure 4 As shown.

[0069] To explore how many sub-rules are needed to evaluate information in a house-tree-person drawing, we can preset a sufficiently large value for k, such that... Figure 2 The model shown, during training, learns to decide which sub-rules to use when faced with different drawings. Based on the above idea, steps S203-S204 are designed to establish the guiding basic feature gating unit.

[0070] Reference Figure 4 The guiding basic feature gating unit ultimately outputs a grouped channel gate as a mask to filter out unwanted sub-rules. (See reference...) Figure 4 The guided base feature gating unit includes a global average pooling layer, a linear layer (also known as a fully connected layer, etc.), and an activation layer (Sigmoid).

[0071] Step S205 is the step of establishing an adaptive channel feature fusion network. In this embodiment, the structure and principle of the established adaptive channel feature fusion network are as follows: Figure 5 As shown. (Refer to...) Figure 5 The main structure of the adaptive channel feature fusion network is a graph attention network.

[0072] In this embodiment, when performing step S4, which involves inputting the basis feature information of each group into the guided basis feature gating unit for processing and obtaining the group gate information and multiple basis vectors output by the guided basis feature gating unit, the following steps can be specifically performed:

[0073] S401. Use a global average pooling layer to process the basis feature information of each group and obtain the basis vectors corresponding to each group of basis feature information;

[0074] S402. Use a linear layer to process each basis vector to obtain the query vector corresponding to each basis vector;

[0075] S403. Obtain the key vector corresponding to each basis vector;

[0076] S404. Form a query matrix using each query vector and a key matrix using each key vector;

[0077] S405. Determine the group attention vector based on the query matrix and the key matrix;

[0078] S406. Use activation layers to process the group attention vector;

[0079] S407. Process the output of the activation layer according to the preset threshold to obtain the group gate information.

[0080] In step S401, refer to Figure 4 The basis feature information output by the drawing basis feature encoder, which has a size of h×w×c, can be processed by global average pooling in the guided basis feature gating unit to obtain basis vectors representing each set of basis feature information. In this embodiment, each set of basis feature information is processed to obtain a corresponding basis vector.

[0081] In step S402, the basis vectors can be reshaped to form a matrix of size k×c / k. Then, the linear layer in the guided basic feature gating unit is used to process this matrix to obtain the query vector corresponding to each basis vector.

[0082] In step S403, the key vector corresponding to each basis vector can be obtained.

[0083] In step S404, the query vectors corresponding to all basis vectors can form a query matrix W. query Meanwhile, the key vectors corresponding to all the basis vectors can form a key matrix W. key .

[0084] In step S405, based on the query matrix W query Bond matrix W key This involves determining the group attention vector. Specifically, refer to... Figure 4 It is possible to query matrix W query Bond matrix W key Perform matrix multiplication to obtain a new matrix. Then, take the elements on the diagonal of this new matrix (for example, take the elements on the diagonal from the top left corner to the bottom right corner) to obtain a new vector. Use this new vector as the group attention vector.

[0085] The i-th element in the attention vector is the dot product between the query value and the key value of the i-th basis vector, representing the attention score of this basis vector.

[0086] In step S406, the group attention vector is processed using the activation layer (Sigmoid) in the guided basic feature gating unit. Then, step S407 is executed, which specifically involves the following steps:

[0087] S40701. Iterate through each element in the output of the activation layer;

[0088] S40702. For any element that has been traversed, compare the element's value with a preset threshold.

[0089] S40703. When the value of an element is greater than a preset threshold, the value of the element is set to 1; otherwise, the value of the element is set to 0.

[0090] When executing steps S40701-S40703, the first element in the output of the activation layer can be read first and compared with a preset threshold λ. If the value of the first element is greater than the preset threshold λ, then the value of the first element is assigned to 1; if the value of the first element is less than or equal to the preset threshold λ, then the value of the first element is assigned to 0. Next, the second element in the output of the activation layer can be read and compared with the preset threshold λ. If the value of the second element is greater than the preset threshold λ, then the value of the second element is assigned to 1; if the value of the second element is less than or equal to the preset threshold λ, then the value of the second element is assigned to 0... Repeat the above operation until all elements in the output of the activation layer have been assigned values.

[0091] By executing steps S40701-S40703, a grouped channel gate can be obtained, which can also be called a grouped channel gate. In this embodiment, the grouped channel gate can be used as a mask to determine whether the basis vector will be input into the adaptive channel feature fusion network.

[0092] In this embodiment, when performing step S5, which involves inputting the group gate information and each basis vector into the adaptive channel feature fusion network for processing and obtaining the depression recognition result output by the adaptive channel feature fusion network, the following steps can be specifically performed:

[0093] S501. Determine the adjacency matrix based on the group gate information and each basis vector;

[0094] S502. Use a graph attention network to perform inter-group feature fusion processing on the adjacency matrix to obtain depression recognition results.

[0095] In this embodiment, when performing step S501, which is to determine the adjacency matrix based on the group gate information and each basis vector, the following steps can be performed:

[0096] S50101. Filter each basis vector according to the group gate information to determine the basis vectors to be filtered and the basis vectors to be retained;

[0097] S50102. Connect the retained basis vectors to obtain the adjacency matrix.

[0098] In this embodiment, the basis vectors of each group can be regarded as node features of the graph, and the connection relationship between each node is determined by the group gate information generated by the previous module (i.e., the guiding basis feature gating unit). Therefore, the adjacency matrix required for the graph can be constructed through the group gate information. Specifically, when executing step S50101, the group gate information can be used as a mask, and each basis vector can be used as the graph. A mask operation is performed on the group gate information and each basis vector. If the result corresponding to a basis vector after the mask operation is different from the basis vector itself, then the basis vector is a filtered basis vector; if the result corresponding to a basis vector after the mask operation is the same as the basis vector itself, then the basis vector is a retained basis vector. The filtered basis vector can be regarded as having no connection relationship with other basis vectors (whether retained basis vectors or other filtered basis vectors).

[0099] In step S50102, all the retained basis vectors obtained in step S50101 are concatenated to obtain the adjacency matrix.

[0100] In step S502, refer to Figure 5The adjacency matrix obtained from steps S50101-S50102 is input into the graph attention network for processing. The graph attention network performs inter-group feature fusion on the adjacency matrix, thereby realizing channel information interaction.

[0101] In step S502, refer to Figure 5 The output of the graph attention network is processed by averaging node features and linearizing the input to a linear layer, ultimately outputting a depression recognition result. The depression recognition result indicates whether the artist (the person being identified) of the house-tree-person drawing has depression or not.

[0102] In this embodiment, Figure 2 The DeHTP shown can be trained first, and then used for depression identification after it is trained.

[0103] Specifically, during training, steps S1-S5 can also be executed. In this case, the house-tree-person (HTP) drawing information obtained in step S1 can be hand-drawn by individuals known to have or not have depression. To train DeHTP, a HTP drawing dataset can be constructed, for example, by collecting data from healthy individuals and individuals diagnosed with depression. Specifically, over 100 HTP drawings hand-drawn by individuals diagnosed with depression and over 100 HTP drawings hand-drawn by healthy controls can be collected, along with the patients' overall descriptions of the drawings and the main elements of the house-tree-person. For example, in one practice, 123 HTP drawings hand-drawn by individuals diagnosed with depression and 105 HTP drawings hand-drawn by healthy controls can be collected, forming a HTP drawing dataset containing 228 HTP drawings. Each HTP drawing in the dataset is labeled with its corresponding actual depression status. For example, for HTP drawings hand-drawn by individuals diagnosed with depression, the actual depression status is labeled as "has depression"; for HTP drawings hand-drawn by healthy controls, the actual depression status is labeled as "does not have depression".

[0104] During training, steps S1-S5 are executed for each House-Tree-Person drawing. In step S5, the depression recognition result (actual output result) is obtained. The actual depression situation corresponding to the House-Tree-Person drawing information is used as the expected output. The loss function value is calculated based on the actual output result and the expected output. Based on the loss function value, the network parameters of at least one module in the drawing base feature encoder, the guidance base feature gating unit, and the adaptive channel feature fusion network are adjusted to achieve the training of DeHTP.

[0105] For example, the following are the relevant hyperparameters during a training process:

[0106] The training sample consisted of 105 healthy control samples and 123 case samples. Data augmentation was performed using horizontal rotation, vertical flipping, and both simultaneously. The number of groups (k) in the grouped convolutions used in DeHTP was set to 8, and the threshold for the guided basic feature gating unit was 0.5. The number of basic features per group output by the painting basic feature encoder was 128; similarly, the length of the node feature vectors in the constructed graph was also 128. The graph attention network had two layers, and each input layer underwent dropout with a probability of 0.2. 10-fold cross-validation was used to evaluate model performance. Adam was used as the optimizer with a learning rate of 0.01 and weight decay of 0.00001. The number of epochs was 400. Both training and testing of DeHTP were performed on two NVIDIA Tesla v100 processors.

[0107] In the process of depression identification, steps S1-S5 can be executed. At this time, the house-tree-person drawing information obtained in step S1 can be a drawing hand-drawn by the person being identified who is unaware of whether they have depression. The depression identification result obtained in step S5 represents a prediction of whether the person being identified has depression.

[0108] In the process of depression identification, the obtained depression identification results can be used not only to assist in the diagnosis of depression, but also for non-medical purposes. For example, for a house-tree-person drawing, steps S1-S5 can be performed to process it and obtain the corresponding depression identification result. This result can then be used as the emotional color classification information for the house-tree-person drawing, which can be labeled and shared. House-tree-person drawings can be searched using the emotional color classification information, thus enabling them to be used for artistic creations based on corresponding emotional color requirements.

[0109] A computer program can be written to execute the depression identification method based on house-tree-person drawing and deep learning in this embodiment. The computer program can be written into a computer device or storage medium. When the computer program is read out and run, the depression identification method based on house-tree-person drawing and deep learning in this embodiment can be executed, thereby achieving the same technical effect as the depression identification method based on house-tree-person drawing and deep learning in the embodiment.

[0110] It should be noted that, unless otherwise specified, when a feature is referred to as "fixed" or "connected" to another feature, it can be directly fixed or connected to the other feature, or indirectly fixed or connected to the other feature. Furthermore, the descriptions of "upper," "lower," "left," and "right" used in this disclosure are only relative to the relative positional relationships of the components of this disclosure in the accompanying drawings. The singular forms "a," "an," and "the" used in this disclosure are also intended to include the plural forms, unless the context clearly indicates otherwise. Moreover, unless otherwise defined, all technical and scientific terms used in this embodiment have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in this embodiment specification is only for describing particular embodiments and is not intended to limit the invention. The term "and / or" as used in this embodiment includes any combination of one or more of the associated listed items.

[0111] It should be understood that although the terms first, second, third, etc., may be used to describe various elements in this disclosure, these elements should not be limited to these terms. These terms are only used to distinguish elements of the same type from each other. For example, a first element may also be referred to as a second element without departing from the scope of this disclosure, and similarly, a second element may also be referred to as a first element. The use of any and all instances or exemplary language (“e.g.,” “such as,” etc.) provided in this embodiment is intended only to better illustrate embodiments of the invention and, unless otherwise required, does not impose a limitation on the scope of the invention.

[0112] It should be recognized that embodiments of the present invention can be implemented or carried out by computer hardware, a combination of hardware and software, or by computer instructions stored in a non-transitory computer-readable storage medium. The method can be implemented using standard programming techniques—including a non-transitory computer-readable storage medium configured with a computer program, wherein such a storage medium causes the computer to operate in a specific and predefined manner—according to the methods and drawings described in the specific embodiments. Each program can be implemented in a high-level procedural or object-oriented programming language to communicate with the computer system. However, if desired, the program can be implemented in assembly or machine language. In any case, the language can be a compiled or interpreted language. Furthermore, for this purpose, the program can run on a programmed application-specific integrated circuit (ASIC).

[0113] Furthermore, the procedures described in this embodiment can be performed in any suitable order unless otherwise indicated by this embodiment or clearly contradicted by the context. The procedures (or variations and / or combinations thereof) described in this embodiment can be executed under the control of one or more computer systems configured with executable instructions, and can be implemented by hardware or a combination thereof as code (e.g., executable instructions, one or more computer programs, or one or more applications) that commonly executes on one or more processors. A computer program includes multiple instructions executable by one or more processors.

[0114] Furthermore, the method can be implemented in any suitable type of computing platform, including but not limited to personal computers, minicomputers, mainframes, workstations, networked or distributed computing environments, standalone or integrated computer platforms, or in communication with charged particle tools or other imaging devices, etc. Aspects of the invention can be implemented as machine-readable code stored on a non-transitory storage medium or device, whether removable or integrated into a computing platform, such as a hard disk, optical read and / or write storage medium, RAM, ROM, etc., such that it is readable by a programmable computer, and when the storage medium or device is read by the computer, it can be used to configure and operate the computer to perform the processes described herein. Furthermore, the machine-readable code, or portions thereof, can be transmitted via wired or wireless networks. The invention of this embodiment includes these and other different types of non-transitory computer-readable storage media when such media comprises instructions or programs that implement the steps above in conjunction with a microprocessor or other data processor. When programmed according to the methods and techniques of the invention, the invention also includes the computer itself.

[0115] A computer program can be applied to input data to perform the functions of this embodiment, thereby transforming the input data to generate output data stored in non-volatile memory. The output information can also be applied to one or more output devices, such as a display. In a preferred embodiment of the invention, the transformed data represents physical and tangible objects, including specific visual depictions of physical and tangible objects generated on the display.

[0116] The above are merely preferred embodiments of the present invention. The present invention is not limited to the above-described embodiments. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention, as long as they achieve the technical effects of the present invention by the same means, should be included within the scope of protection of the present invention. Within the scope of protection of the present invention, the technical solutions and / or implementation methods can have various modifications and variations.

Claims

1. A method for identifying depression based on house-tree-person drawing and deep learning, characterized in that, The depression identification method based on house-tree-person drawing and deep learning includes: Obtain information on house-tree-person drawing; Establish a first convolutional layer, at least two residual modules, and at least two grouped residual modules; wherein, the residual modules include multiple second convolutional layers, and the grouped residual modules include multiple grouped convolutional layers; The first convolutional layer, each residual module, and each grouped residual module are connected sequentially to obtain the painting base feature encoder; Establish a global average pooling layer, a linear layer, and an activation layer; The global average pooling layer, the linear layer, and the activation layer are connected sequentially to obtain a directional base feature gating unit; Establish a graph attention network as an adaptive channel feature fusion network; The house-tree-person drawing information is input into the drawing basis feature encoder for processing to obtain multiple group basis feature information output by the drawing basis feature encoder; Each of the aforementioned basis feature information is input into the guiding basis feature gating unit for processing, to obtain the group gate information and multiple basis vectors output by the guiding basis feature gating unit; wherein, each of the basis vectors corresponds to one of the aforementioned basis feature information; The group gate information and each of the basis vectors are input into the adaptive channel feature fusion network for processing, and the depression recognition result output by the adaptive channel feature fusion network is obtained.

2. The depression identification method based on house-tree-person drawing and deep learning according to claim 1, characterized in that, The step of inputting each of the basis feature information groups into the guiding basis feature gating unit for processing, and obtaining the group gate information and multiple basis vectors output by the guiding basis feature gating unit, includes: The global average pooling layer is used to process the basis feature information of each group to obtain the basis vector corresponding to each basis feature information; The linear layer is used to process each of the basis vectors to obtain the query vector corresponding to each of the basis vectors. Obtain the key vector corresponding to each of the aforementioned basis vectors; A query matrix is ​​formed using each of the query vectors, and a key matrix is ​​formed using each of the key vectors; Determine the group attention vector based on the query matrix and the key matrix; The activation layer is used to process the group attention vector; The output of the activation layer is processed according to a preset threshold to obtain the group gate information.

3. The depression identification method based on house-tree-person drawing and deep learning according to claim 2, characterized in that, The step of processing the output of the activation layer according to a preset threshold to obtain the group gate information includes: Iterate through each element in the output of the activation layer; For any element that has been traversed, compare the element's value with the preset threshold. When the value of an element is greater than the preset threshold, the value of the element is assigned to 1; otherwise, the value of the element is assigned to 0.

4. The depression identification method based on house-tree-person drawing and deep learning according to claim 1, characterized in that, The step of inputting the group gate information and each of the basis vectors into the adaptive channel feature fusion network for processing, and obtaining the depression recognition result output by the adaptive channel feature fusion network, includes: Based on the group gate information and each of the basis vectors, determine the adjacency matrix; The graph attention network is used to perform inter-group feature fusion processing on the adjacency matrix to obtain the depression identification result.

5. The depression identification method based on house-tree-person drawing and deep learning according to claim 4, characterized in that, The step of determining the adjacency matrix based on the group gate information and each of the basis vectors includes: Based on the group gate information, each of the basis vectors is filtered to determine the filtered basis vectors and the retained basis vectors; The adjacency matrix is ​​obtained by connecting the retained basis vectors.

6. A computer device, characterized in that, The device includes a memory and a processor, wherein the memory is used to store at least one program and the processor is used to load at least one program to execute the depression identification method based on house-tree-person drawing and deep learning as described in any one of claims 1-5.

7. A computer-readable storage medium storing a processor-executable program, characterized in that, The processor-executable program, when executed by the processor, is used to perform the depression identification method based on house-tree-person drawing and deep learning as described in any one of claims 1-5.

Images