Visual health assessment method, system and equipment for university campus landscape

Abstract

The invention provides a visual health assessment method, system and device for a university campus landscape, and the method specifically comprises the steps: obtaining image data of the university campus landscape, carrying out the scoring of the image data of the university campus landscape based on the assessment features of the university campus landscape, carrying out the consistency check of scores through employing a Cohen's Kappa statistical method, and obtaining the scoring data; performing association matching on the university campus landscape image data and the score data to form a visual health assessment data set; inputting the visual health assessment data set into a pre-constructed deep learning model for training, and generating a visual health assessment model; and performing thermodynamic diagram generation on an output result of the visual health assessment model, and determining a key assessment area of the visual health assessment model for the university campus landscape image. The method aims to realize automation of the CLM evaluation process, improves the evaluation efficiency and objectivity, and can be applied to large-scale campus landscape analysis.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a method, system and device for visual health assessment of university campus landscape. Background Art

[0002] With the vigorous development of higher education, university campuses are important places for students to study, live and socialize. The environmental quality, especially the visual health quality of campus landscape, has a profound impact on the mental health, emotional state and overall well-being of students and faculty. Traditional campus landscape assessment methods mostly rely on the subjective judgment of experts and on-site surveys. Although this method can reflect the quality of the landscape to a certain extent, it has many limitations: first, the assessment process is cumbersome and time-consuming, and it is difficult to achieve large-scale and efficient assessment; second, the assessment results are easily affected by the personal experience and preferences of experts, and lack unified and quantifiable assessment standards; finally, traditional methods are difficult to capture the dynamic impact and subtle differences of landscape on pedestrian visual health.

[0003] The theoretical framework of landscape visual quality measurement (CLM) provides a theoretical basis for the scientific assessment of landscape quality. However, its limitations in practical applications are becoming increasingly prominent. CLM relies on manual scoring by experts, which is not only inefficient, but also difficult to update the assessment results in a timely manner in a rapidly changing campus environment. In addition, with the expansion of campus scale and the diversification of landscape design, relying solely on manual assessment can no longer meet the needs of comprehensive and accurate assessment of campus landscape visual health.

[0004] At the same time, the rapid development of artificial intelligence technology, especially the breakthrough progress of deep learning in the field of image recognition and analysis, has provided new ideas for automated and intelligent campus landscape visual health assessment. However, how to effectively combine CLM theory with artificial intelligence technology to develop an automated assessment system that can reflect the visual quality of the landscape and adapt to changes in the complex campus environment is still a major challenge in current research.

[0005] In addition, for the visual health assessment of campus landscapes, simply obtaining the assessment results is far from enough. How to further analyze the relationship between pedestrian activity patterns and landscape visual quality through the assessment results, optimize campus layout, and enhance the experience of teachers and students is also an urgent problem to be solved. Summary of the invention

[0006] The purpose of the present invention is to provide a visual health assessment method, system and equipment for university campus landscape, aiming to realize the automation of the CLM assessment process, improve the assessment efficiency and objectivity, and be applicable to large-scale campus landscape analysis to solve at least one of the above-mentioned prior art problems.

[0007] In a first aspect, the present invention provides a method for visual health assessment of university campus landscape, the method specifically comprising: Acquire university campus landscape image data, score the university campus landscape image data based on campus landscape assessment characteristics, and use Cohen's Kappa statistical method to check the consistency of the scores to obtain score data; Associating and matching the university campus landscape image data and the scoring data to form a visual health assessment data set; Inputting the visual health assessment dataset into a pre-built deep learning model for training to generate a visual health assessment model; By generating a heat map of the output results of the visual health assessment model, a key assessment area of ​​the visual health assessment model for the university campus landscape image is determined.

[0008] In a second aspect, the present invention provides a visual health assessment system for university campus landscape, the system specifically comprising: The first evaluation module is used to obtain university campus landscape image data, score the university campus landscape image data based on campus landscape evaluation characteristics, and use Cohen's Kappa statistical method to check the consistency of the score to obtain score data; A second evaluation module is used to associate and match the university campus landscape image data with the scoring data to form a visual health evaluation data set; A third assessment module is used to input the visual health assessment data set into a pre-built deep learning model for training to generate a visual health assessment model; The fourth evaluation module is used to generate a heat map of the output results of the visual health evaluation model to determine the key evaluation area of ​​the visual health evaluation model for the university campus landscape image.

[0009] In a third aspect, the present invention provides a computer device, comprising: a memory and a processor and a computer program stored in the memory, and when the computer program is executed on the processor, a visual health assessment method for a university campus landscape as described in any one of the above methods is implemented.

[0010] In a fourth aspect, the present invention provides a computer-readable storage medium, characterized in that a computer program is stored thereon, and when the computer program is executed by a processor, the visual health assessment method of the university campus landscape as described in any of the above methods is implemented.

[0011] Compared with the prior art, the present invention has at least one of the following technical effects: 1. This invention aims to propose an innovative visual health assessment method for university campus landscape by combining CLM theory with artificial intelligence technology to realize the automation and intelligence of the assessment process, and to deeply analyze the relationship between landscape and pedestrian activities with the help of thermal map data, so as to provide scientific basis and technical support for optimizing the campus environment and improving the welfare of teachers and students.

[0012] 2. The present invention develops a prediction model that combines the self-attention mechanism and the multi-task learning (MTL) framework, which can predict the potential impact of future campus landscape changes on pedestrian visual health based on existing landscape characteristics and planning schemes.

[0013] 3. The present invention analyzes the thermal map data to establish the relationship between the activity pattern of campus pedestrians and the visual quality of the landscape, providing a more accurate assessment of visual health impacts.

[0014] 4. The present invention can provide university campus planners and managers with a systematic and quantifiable tool for evaluating and optimizing the impact of campus landscape on pedestrian visual health.

[0015] 5. This invention provides a model for combining CLM with artificial intelligence technology and applying it to other types of urban spaces.

[0016] 6. Scenario innovation: The present invention has achieved significant scenario innovation in the field of visual health assessment on university campuses. First, it combines the theory of the Contemplative Landscape Model (CLM) with practical applications, providing a new perspective for campus environmental assessment. Secondly, by introducing heat map data analysis, the present invention successfully established the relationship between campus pedestrian activity patterns and landscape visual quality, deepening the understanding of factors affecting the campus environment. Finally, the present invention has developed an innovative assessment tool that can not only evaluate the existing landscape, but also predict the potential impact of future campus landscape changes on pedestrian visual health, providing forward-looking decision-making support for campus planning and management.

[0017] 7. Model innovation: In terms of model design, the present invention has demonstrated many innovations. The core innovation lies in the design of a deep learning model architecture that combines the EfficientNet-B0 backbone network and the self-attention mechanism, which improves the ability of feature extraction and important information identification. At the same time, the multi-task learning framework is adopted to achieve the simultaneous prediction of the total visual health score and the scores of seven key features, thereby enhancing the comprehensive performance of the model. In addition, the present invention innovatively uses Grad-CAM to generate a heat map of the CLM score, which realizes the intuitive visualization of the evaluation results and provides a more intuitive and easy-to-understand form of expression for campus environment analysis. These model innovations together constitute an efficient and accurate campus visual health assessment system. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0019] Figure 1 It is a flow chart of a visual health assessment method for university campus landscape provided by one embodiment of the present invention; Figure 2 is a structural schematic diagram of a visual health assessment model provided by an embodiment of the present invention; Figure 3 It is a structural schematic diagram of a visual health assessment system for university campus landscape provided by one embodiment of the present invention; Figure 4 It is a structural schematic diagram of a computer device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0020] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present application.

[0021] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, wholes, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or combinations thereof.

[0022] It should also be understood that the term “and / or” used in the specification and appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0023] As used in the specification and appended claims of this application, the term "if" can be interpreted as "when" or "uponce" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrase "if it is determined" or "if [described condition or event] is detected" can be interpreted as meaning "uponce it is determined" or "in response to determining" or "uponce [described condition or event] is detected" or "in response to detecting [described condition or event]", depending on the context.

[0024] In addition, in the description of the present application specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0025] References to "one embodiment" or "some embodiments" etc. described in the specification of this application mean that one or more embodiments of the present application include specific features, structures or characteristics described in conjunction with the embodiment. Therefore, the statements "in one embodiment", "in some embodiments", "in some other embodiments", "in some other embodiments", etc. that appear in different places in this specification do not necessarily refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways. The terms "including", "comprising", "having" and their variations all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0026] In the embodiment of the present application, the execution subject of the process includes a terminal device, which includes but is not limited to: a server, a computer, a smart phone, a tablet computer, and other devices capable of executing the method disclosed in the present application. Figure 1 A flow chart of a visual health assessment method for a university campus landscape disclosed in an embodiment of the present invention is shown, and is described in detail as follows: S101, obtaining university campus landscape image data, scoring the university campus landscape image data based on campus landscape assessment characteristics, and using Cohen's Kappa statistical method to perform consistency check on the scores to obtain scoring data.

[0027] In one embodiment of the present invention, satellite remote sensing technology is used to obtain high-resolution image data within the campus, and landscape image data of different areas and different time periods in the university campus are collected, including but not limited to landscape elements such as green spaces, buildings, water bodies, roads, etc., and landscape design experts trained by CLM are invited to score based on 7 key features, namely, campus landscape evaluation features (landscape hierarchy, terrain, vegetation, color and light, compatibility, archetypal elements, and tranquility features).

[0028] Scoring method: As an embodiment, the campus landscape image data is calculated and analyzed using landscape pattern index software (such as FragStats); or, as an embodiment, experts use a Likert scale of 1-6 points to score, ensuring that at least 3 experts score independently to improve the reliability of the scoring; the spatial structural characteristics and layout rationality of the campus landscape are evaluated, and the campus landscape is scored based on the calculation results of the index and the scoring criteria.

[0029] Cohen's Kappa statistical method was used to calculate the consistency between the scores.

[0030]

[0031] in, indicates the observed consistency, Indicates expected consistency.

[0032] S102, associating and matching the university campus landscape image data and the scoring data to form a visual health assessment data set.

[0033] In this embodiment, data preprocessing and matching are mainly performed on the university campus landscape image data and the scoring data. Specifically, the scoring data and image data are first cleaned and standardized to ensure data validity, and then the image data are matched with the scoring data to construct a visual health assessment data set. When selecting samples, for each route, the images with the highest and lowest total CLM scores are selected; for the seven key features, the images with the highest and lowest scores are selected respectively.

[0034] S103, inputting the visual health assessment data set into a pre-built deep learning model for training to generate a visual health assessment model.

[0035] In this embodiment, a deep learning model including an EfficientNet-B0 network module, a self-attention mechanism module, a feature fusion module and a multi-task learning framework module is constructed. Among them, EfficientNet is used as a feature extractor. The network optimizes the balance between network depth, width and resolution to achieve improved model performance while maintaining computational efficiency. This module is responsible for extracting rich feature maps from the input image; on the basis of the feature map, a self-attention mechanism is introduced to enhance the model's ability to focus on key areas by calculating the attention weights between each position in the feature map. This module helps the model to better understand the complex structure and detail information in the image; the output results of the EfficientNet-B0 network module and the self-attention mechanism module are feature fused to form fusion features. This module improves the representation and generalization capabilities of the model by integrating feature information from different sources; on the basis of the fusion features, a multi-task learning framework is constructed, including tasks such as visual health score prediction, classification output and loss adjustment. This module improves the evaluation accuracy and robustness of the model by optimizing multiple task objectives at the same time.

[0036] The visual health assessment dataset is divided into a training set, a validation set, and a test set, and the training set is used to train the deep learning model. During the training process, appropriate optimization algorithms and loss functions are used to update the model parameters through the back propagation algorithm, so that the model gradually learns the mapping relationship from images to visual health scores. At the same time, the validation set is used to evaluate the performance of the model, and the model structure and parameter settings are adjusted according to the evaluation results.

[0037] The trained visual health assessment model is finally evaluated using the test set to verify its accuracy and reliability in practical applications. Evaluation indicators can include accuracy, recall, F1 score, etc. After the evaluation is passed, the model is applied to the visual health assessment of university campus landscapes. By inputting new campus landscape image data, the model can automatically output the corresponding visual health score and key assessment area heat map, providing scientific basis and technical support for the optimal design of campus landscapes.

[0038] S104, generating a heat map of the output result of the visual health assessment model to determine the key assessment area of ​​the visual health assessment model for the university campus landscape image.

[0039] In this embodiment, first, the university campus landscape image to be evaluated is input into the trained visual health assessment model. After a series of forward propagation calculations, the model will output a feature map (or heat map base map) with the same size as the input image or a specific resolution. Each pixel value on the feature map represents the importance or score of the corresponding image area in the visual health assessment.

[0040] In order to convert the feature map into an easy-to-understand heat map, the feature map needs to be normalized and visualized. The specific steps include: normalization, scaling all pixel values ​​in the feature map to a specific range (such as 0 to 255) for subsequent visualization; color mapping, mapping the normalized pixel values ​​to different colors, usually using the color coding scheme of the heat map (such as blue for low scores and red for high scores); image generation, generating the final heat map image based on the color mapping results.

[0041] By observing the distribution and intensity of colors on the generated heat map, it is possible to determine which areas are considered key assessment areas by the visual health assessment model. These areas usually have higher scores or significant color differences, reflecting their importance in the visual health of the campus landscape.

[0042] In this embodiment, the heat map intuitively displays the distribution of key assessment areas in the campus landscape image through color coding, so that non-professionals can also quickly understand the assessment results. The determination of key assessment areas helps campus managers and planners identify areas that need to be focused on and improved, providing a scientific basis for subsequent campus landscape optimization and transformation.

[0043] Based on the above steps S101 to S104, in some embodiments, refer to Figure 2 , the visual health assessment model includes an EfficientNet-B0 network module, a self-attention mechanism module, a feature fusion module and a multi-task learning framework module; The EfficientNet-B0 network module is used to extract features from the input image of the visual health assessment data set to generate a feature map; The self-attention mechanism module is used to calculate the attention weight of the feature map; The feature fusion module is used to perform feature fusion on the output result of the EfficientNet-B0 network module and the output result of the self-attention mechanism module to form a fusion feature; The multi-task learning framework module is used to perform visual health score prediction, classification output and loss adjustment on the fused features.

[0044] Furthermore, the EfficientNet-B0 network module includes an input layer, a basic convolutional layer, a backbone structure and a feature map output layer; The input layer is used to obtain the input image of the visual health assessment dataset; The basic convolution layer is used to perform a convolution operation on an input image of the visual health assessment dataset; The backbone structure includes a plurality of composite blocks, each of which includes a depthwise convolution block and a pointwise convolution block, wherein the depthwise convolution block is used to perform independent convolution on each input channel using a single convolution kernel, and the pointwise convolution block is used to perform channel mixing through a 1x1 convolution kernel to combine the features of each input channel; The feature map output layer is used to output the feature map formed after being processed by the basic convolution layer and the backbone structure.

[0045] In this embodiment, EfficientNet is used as the backbone network, which can achieve higher performance with fewer parameters. EfficientNet optimizes depth, width and resolution simultaneously through a composite scaling method. Batch Normalization and ReLU activation functions are added after each convolutional layer of the network to accelerate training and improve performance.

[0046] Input layer: Input image size: 224×224×3 (RGB image).

[0047] Basic convolution layer: In the basic convolution layer, a 3x3 convolution kernel is used and the stride is set to 2 to achieve downsampling of the feature map. By using "same" padding, the width and height of the input and output feature maps are kept consistent to ensure that edge information is not lost.

[0048] Activation function: ReLU (rectified linear unit) as an activation function helps to introduce nonlinear features and improve the expressive power of the network.

[0049] Batch Normalization: Batch Normalization is added after each convolutional layer to enhance the stability and convergence speed of the model. This layer effectively reduces internal covariate shift and improves training efficiency by standardizing each batch of inputs.

[0050] The core of EfficientNet-B0 lies in its efficient composite block structure. Each composite block consists of DepthwiseSeparable Convolution, as follows: Depthwise convolution: Use a single convolution kernel to perform independent convolution on each input channel. This method can significantly reduce the number of parameters and computational complexity.

[0051] Pointwise convolution: Then, channels are mixed through a 1x1 convolution kernel to combine the features of each channel. This process helps to improve the expressiveness of the features.

[0052] Activation function: The ReLU activation function is also used to further promote the learning of nonlinear features.

[0053] Batch Normalization: Batch Normalization is added after each convolutional layer to enhance the training stability of the model and accelerate convergence.

[0054] At the feature map output layer, the output feature map size of each block is reduced to half. This design of reducing the feature map size layer by layer enables the network to extract rich features at multiple levels while effectively reducing the amount of calculation. The details are as follows: 1st Block: Output shape (112, 112, 32) 2nd Block: Output shape (56, 56, 16) 3rd Block: Output shape (28, 28, 24) Similarly, the final output feature map shape is (7, 7, 1280).

[0055] Furthermore, the attention weight calculation of the feature map specifically includes: Determine the shape N, H, W and C of the feature map, where N represents the batch size, H and W represent the height and width of the feature map, respectively, and C represents the number of channels; Based on the shape of the feature map, transform the feature map through a fully connected layer to calculate the query Q, key K and value V; Based on the query Q, key K and value V, the importance of the features of the feature graph is determined by the attention weight calculation formula to obtain the self-attention output. The attention weight calculation formula is:

[0056] Among them, Attention(Q,K,V) represents the self-attention output, represents the similarity between query Q and key K, represents the weight matrix.

[0057] In this embodiment, the feature map is first input: the shape is (N, H, W, C), where N is the batch size, H and W represent the height and width of the feature map respectively, and C is the number of channels.

[0058] Calculate query, key and value (Q, K, V): In the self-attention mechanism, the input feature map is first transformed through the fully connected layer to calculate Q, K, V:

[0059] Next, the importance between features is determined by calculating the attention weights:

[0060] here, The similarity between the query and the key is calculated. This is done by scaling it (dividing it by ) and applying the Softmax function, we get the weight matrix, which is then weighted by the sum value (V) to get the self-attention output.

[0061] Output feature map: The shape remains (N, H, W, C). This design ensures that the output of the self-attention layer is consistent with the shape of the input feature map, which facilitates subsequent feature fusion.

[0062] Furthermore, the feature fusion module includes a feature connection layer and a first fully connected layer; The feature connection layer is used to perform feature fusion on the output result of the EfficientNet-B0 network module and the output result of the self-attention mechanism module to form a preliminary fusion feature; The first fully connected layer is used to flatten the preliminary fusion features into a one-dimensional vector, and use a ReLU activation function to perform a nonlinear transformation on the one-dimensional vector to form a target fusion feature.

[0063] In this embodiment, the output of the CNN (shape is (N, 7, 7, 1280)) is connected with the output of the self-attention layer (shape is (N, 7, 7, C)) to form a fusion feature, and the new shape is (N, 7, 7, 1280 + C).

[0064] Flatten Layer: Flattens the fused features into a 1D vector with a shape of (N, 7, 7, 1280+C). This operation helps convert multi-dimensional features into a form suitable for the fully connected layer.

[0065] Fully connected layer: The number of hidden layer nodes is set to 512, and the ReLU activation function is used. The output shape is (N, 512). Through the fully connected layer, the network can learn higher-order feature representations.

[0066] Dropout Layer: Set the Dropout ratio to 0.5 to reduce the risk of overfitting. Dropout effectively improves the generalization ability of the model by randomly discarding some neurons.

[0067] Furthermore, the multi-task learning framework module includes a regression output layer, a classification output layer and a loss function layer; The regression output layer is used to predict the overall visual health score of the fused features using a fully connected layer and a linear activation function; The classification output layer is used to determine the score of the fusion feature on each campus landscape assessment feature using a fully connected layer and a Sigmoid activation function; The loss function layer is used to balance the loss of the regression output layer and the loss of the classification output layer. The loss function layer includes a loss function L, and the loss function L satisfies

[0068]

[0069]

[0070] in, represents the regression loss function, which is used to evaluate the loss of the regression output layer. represents the classification loss function, which is used to evaluate the loss of the classification output layer. and Respectively represent the weight coefficients of the regression output layer and the classification output layer, which are used to balance the losses between the two. N represents the batch size. represents the true value, represents the model prediction value, represents the true label, Represents the model prediction probability.

[0071] In this embodiment, the multi-task learning framework includes output layer design and loss function design.

[0072] Among them, the output layer design is as follows: Visual health score prediction (regression output): This task is implemented through a fully connected layer with the number of nodes set to 1 and an output shape of (N,1). Here, N represents the batch size. The main purpose of this layer is to predict the overall visual health score, and a linear activation function is used to ensure that the output is a continuous value, which is suitable for regression tasks.

[0073] Key feature classification (classification output): This task includes 7 independent fully connected layers, the number of nodes in each layer is also set to 1, and the output shape is (N, 1). Each layer corresponds to the score of a key feature. In order to limit the output between 0 and 1, the Sigmoid activation function is used.

[0074] The loss function design is as follows: The loss function L is defined as:

[0075] Regression Loss : The mean square error (MSE) is used to measure the difference between the total visual health score predicted by the model and the true value. The MSE calculation formula is:

[0076] in is the true value, is the model prediction value. The advantage of MSE is that it penalizes larger errors more significantly, which helps to improve the prediction accuracy of the model.

[0077] Classification Loss : The cross-entropy loss function is used to evaluate the performance of the model in the key feature classification task. The cross-entropy loss calculation formula is:

[0078] here, is the true label (0 or 1), is the probability predicted by the model. The cross entropy loss effectively measures the difference between probability distributions and is particularly suitable for dealing with multi-class classification problems.

[0079] Weight coefficient and : The two coefficients are used to balance the loss of regression tasks and classification tasks. By adjusting these two coefficients, dynamic adjustments can be made during training according to the importance of the tasks, thus ensuring that the model can achieve good performance on both tasks.

[0080] In some embodiments, in the above step S103, the step of inputting the visual health assessment dataset into a pre-built deep learning model for training specifically includes: Determine the settings of hyperparameters of pre-built deep learning models, including the learning rate , batch size N, number of training epochs and dropout ratio; The Adam optimizer is used to optimize the deep learning model with set hyperparameters. The optimization calculation formula satisfies

[0081] in, represents the parameter value of iteration t+1, represents the parameter value of iteration t, and denote the estimates of the first and second moments of the gradient, respectively. represents a small constant to prevent division by zero; The Bayesian optimization algorithm is used to tune the hyperparameters of the deep learning model, and a learning rate scheduler is used to reduce the learning rate when the validation loss is no longer decreasing.

[0082] In this embodiment, the learning rate is first , batch size N, number of training rounds (epochs), dropout ratio, etc. are initialized; then the Adam optimizer is used. Based on its adaptive learning rate characteristics, it can quickly converge to the optimal solution, which is calculated as follows:

[0083] in, Indicates the parameter value of the iteration. and are estimates of the first and second moments of the gradient, respectively. is a small constant to prevent division by zero.

[0084] Next, we use Bayesian optimization to tune hyperparameters, using the validation set loss as the objective function. We select the point with the largest expected improvement based on the previous evaluation results:

[0085] in, represents the expected improvement value at point x, represents the objective function value at the hyperparameter point x, Indicates the current optimal point The objective function value on , E represents the expected calculation of possible improvements.

[0086] Use the learning rate scheduler (ReduceLROnPlateau) to reduce the learning rate when the validation loss is no longer decreasing to improve training efficiency. The calculation formula is as follows:

[0087] in, is the scaling factor, and defines 1, used to reduce the learning rate.

[0088] Based on the above, the early stopping method is used to monitor the validation set loss. If there is no improvement within N consecutive epochs, the training is stopped to prevent overfitting. The early stopping method and dynamic learning rate adjustment ensure the high efficiency of training and the generalization ability of the model.

[0089] In some embodiments, in the above step S104, generating a heat map of the output result of the visual health assessment model to determine the key assessment area of ​​the visual health assessment model for the university campus landscape image specifically includes: Acquire the landscape image data of the first university campus, input the landscape image data of the first university campus into the visual health assessment model, and obtain a visual health score; Based on the visual health score, the gradient of the category score of each campus landscape assessment feature to the last convolutional layer feature map of the visual health assessment model and the gradient average of the last convolutional layer are calculated, and the feature map is weighted according to the feature map weighting formula to obtain a weighted result. The feature map weighting formula satisfies

[0090] in, Represents the average gradient of category c, reflecting the influence of this category on the feature map. Represents the model's prediction score or output value for category c, Indicates the last convolutional layer Row, No. The activation value of a channel of the feature map of the column, Represents the spatial dimension of the feature map, used to normalize the sum of gradients, Represents the score feature map of category c The partial derivative of reflects the influence of each position in the feature map on the final classification result; A heat map is generated according to the weighted result, and the heat map is normalized. The normalized heat map and the original image of the first university campus landscape image data are then superimposed to determine the key assessment area of ​​the visual health assessment model for the university campus landscape image.

[0091] In this embodiment, the accuracy of the model is evaluated by evaluation indicators, including mean square error MSE, accuracy and F1-score, wherein mean square error MSE is used for regression output, and accuracy and F1-score are used for classification output.

[0092] After the model training is completed, the heat map is generated to analyze the model's attention focus area in visual health assessment. Randomly select image samples from the test set for heat map generation to intuitively display the model's decision basis. The present invention adopts the Grad-CAM heat map generation method to visualize the feature response of the convolutional neural network on a specific category.

[0093] Forward propagation: input the selected image into the trained model to obtain the visual health score; Backpropagation: Calculate the gradient of the target category score with respect to the feature map of the last convolutional layer to identify the influence of the feature map on the model's decision.

[0094] On this basis, the gradient average of the last convolutional layer is calculated and the feature map is weighted:

[0095] in, Represents the average gradient of category c, reflecting the influence of this category on the feature map; Represents the model's prediction score or output value for category c (the activation value of the last layer); Indicates the last convolutional layer Row, No. The activation value of a channel of the feature map of the column; Represents the spatial dimension of the feature map, used to normalize the sum of gradients; Score pair feature map representing the category The partial derivative reflects the influence of each position in the feature map on the final classification result.

[0096] Next, a heat map is generated based on the weights and normalized so that its value is between 0 and 1 for easy visualization.

[0097]

[0098] in, The feature map of the kth channel in the last convolutional layer reflects the responses of different feature channels. Through the above operations, Grad-CAM can effectively capture and visualize the key areas that the model focuses on when performing visual health assessment. Then, the heat map is superimposed on the original image, and the final generated image shows the areas that the model focuses on during the assessment process.

[0099] Finally, the trained model is used to predict new campus images and output the corresponding visual health scores and classification results of seven key features. A visual health status heat map is generated based on the model's prediction results to help identify areas with high and low visual health on campus, providing a basis for subsequent environmental optimization.

[0100] Reference Figure 3 An embodiment of the present invention provides a visual health assessment system 3 for university campus landscape, and the system 3 specifically includes: The first evaluation module 301 is used to obtain university campus landscape image data, score the university campus landscape image data based on campus landscape evaluation characteristics, and use Cohen's Kappa statistical method to check the consistency of the score to obtain score data; The second evaluation module 302 is used to associate and match the university campus landscape image data with the scoring data to form a visual health evaluation data set; The third assessment module 303 is used to input the visual health assessment data set into a pre-built deep learning model for training to generate a visual health assessment model; The fourth evaluation module 304 is used to generate a heat map of the output results of the visual health evaluation model to determine the key evaluation area of ​​the visual health evaluation model for the university campus landscape image.

[0101] It is understandable that if Figure 1 The contents of the university campus landscape visual health assessment method embodiment shown in the figure are applicable to the university campus landscape visual health assessment system embodiment. The functions specifically implemented by the university campus landscape visual health assessment system embodiment are similar to those in the example shown in the figure. Figure 1 The visual health assessment method for the university campus landscape shown in the embodiment is the same as that in the embodiment shown in the embodiment of the present invention, and the beneficial effects achieved are the same as those achieved in the embodiment of the present invention. Figure 1 The beneficial effects achieved by the illustrated embodiment of the method for visual health assessment of university campus landscape are also the same.

[0102] It should be noted that the information interaction, execution process and other contents between the above-mentioned systems are based on the same concept as the embodiment of the method of the present invention. Their specific functions and technical effects can be found in the method embodiment part and will not be repeated here.

[0103] Those skilled in the art can clearly understand that, for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0104] Reference Figure 4 The embodiment of the present invention further provides a computer device 4, comprising: a memory 402 and a processor 401 and a computer program 403 stored in the memory 402. When the computer program 403 is executed on the processor 401, a visual health assessment method for a university campus landscape as described in any one of the above methods is implemented.

[0105] The computer device 4 may be a computing device such as a desktop computer, a notebook, a PDA, or a cloud server. The computer device 4 may include, but is not limited to, a processor 401 and a memory 402. Those skilled in the art will appreciate that Figure 4 It is only an example of the computer device 4 and does not constitute a limitation on the computer device 4. It may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, it may also include input and output devices, network access devices, etc.

[0106] The processor 401 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0107] In some embodiments, the memory 402 may be an internal storage unit of the computer device 4, such as a hard disk or memory of the computer device 4. In other embodiments, the memory 402 may also be an external storage device of the computer device 4, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the computer device 4. Further, the memory 402 may also include both an internal storage unit and an external storage device of the computer device 4. The memory 402 is used to store an operating system, an application program, a boot loader (BootLoader), data, and other programs, such as the program code of the computer program, etc. The memory 402 may also be used to temporarily store data that has been output or is to be output.

[0108] An embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the visual health assessment method for university campus landscape as described in any one of the above methods is implemented.

[0109] In this embodiment, if the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the processes in the above-mentioned embodiment method, which can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps of the above-mentioned various method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium can at least include: any entity or device that can carry the computer program code to the camera / terminal device, recording medium, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium. For example, USB flash drive, mobile hard disk, disk or optical disk. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electric carrier signals and telecommunication signals.

[0110] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0111] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0112] In the embodiments disclosed in the present application, it should be understood that the disclosed devices / terminal equipment and methods can be implemented in other ways. For example, the device / terminal equipment embodiments described above are only schematic. For example, the division of the modules or units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0113] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

Claims

1. A visual health assessment method for university campus landscape, characterized in that: The method specifically comprises: Acquire university campus landscape image data, score the university campus landscape image data based on campus landscape assessment characteristics, and use Cohen's Kappa statistical method to check the consistency of the scores to obtain score data; Associating and matching the university campus landscape image data and the scoring data to form a visual health assessment data set; Inputting the visual health assessment dataset into a pre-built deep learning model for training to generate a visual health assessment model; By generating a heat map of the output results of the visual health assessment model, a key assessment area of ​​the visual health assessment model for the university campus landscape image is determined.

2. The visual health assessment method of university campus landscape according to claim 1 is characterized in that: The visual health assessment model includes an EfficientNet-B0 network module, a self-attention mechanism module, a feature fusion module and a multi-task learning framework module; The EfficientNet-B0 network module is used to extract features from the input image of the visual health assessment data set to generate a feature map; The self-attention mechanism module is used to calculate the attention weight of the feature map; The feature fusion module is used to perform feature fusion on the output result of the EfficientNet-B0 network module and the output result of the self-attention mechanism module to form a fusion feature; The multi-task learning framework module is used to perform visual health score prediction, classification output and loss adjustment on the fused features.

3. The visual health assessment method of university campus landscape according to claim 2 is characterized in that: The EfficientNet-B0 network module includes an input layer, a basic convolutional layer, a backbone structure and a feature map output layer; The input layer is used to obtain the input image of the visual health assessment dataset; The basic convolution layer is used to perform a convolution operation on an input image of the visual health assessment dataset; The backbone structure includes a plurality of composite blocks, each of which includes a depthwise convolution block and a pointwise convolution block, wherein the depthwise convolution block is used to perform independent convolution on each input channel using a single convolution kernel, and the pointwise convolution block is used to perform channel mixing through a 1x1 convolution kernel to combine the features of each input channel; The feature map output layer is used to output the feature map formed after being processed by the basic convolution layer and the backbone structure.

4. The visual health assessment method of university campus landscape according to claim 2 is characterized in that: The calculating of the attention weight of the feature map specifically includes: Determine the shape N, H, W and C of the feature map, where N represents the batch size, H and W represent the height and width of the feature map, respectively, and C represents the number of channels; Based on the shape of the feature map, transform the feature map through a fully connected layer to calculate the query Q, key K and value V; Based on the query Q, key K and value V, the importance of the features of the feature graph is determined by the attention weight calculation formula to obtain the self-attention output. The attention weight calculation formula is: Among them, Attention(Q,K,V) represents the self-attention output, represents the similarity between query Q and key K, represents the weight matrix.

5. The visual health assessment method of university campus landscape according to claim 2 is characterized in that: The feature fusion module includes a feature connection layer and a first fully connected layer; The feature connection layer is used to perform feature fusion on the output result of the EfficientNet-B0 network module and the output result of the self-attention mechanism module to form a preliminary fusion feature; The first fully connected layer is used to flatten the preliminary fusion features into a one-dimensional vector, and use a ReLU activation function to perform a nonlinear transformation on the one-dimensional vector to form a target fusion feature.

6. The visual health assessment method of university campus landscape according to claim 2 is characterized in that: The multi-task learning framework module includes a regression output layer, a classification output layer and a loss function layer; The regression output layer is used to predict the overall visual health score of the fused features using a fully connected layer and a linear activation function; The classification output layer is used to determine the score of the fusion feature on each campus landscape assessment feature using a fully connected layer and a Sigmoid activation function; The loss function layer is used to balance the loss of the regression output layer and the loss of the classification output layer. The loss function layer includes a loss function L, and the loss function L satisfies in, represents the regression loss function, which is used to evaluate the loss of the regression output layer. represents the classification loss function, which is used to evaluate the loss of the classification output layer. and Respectively represent the weight coefficients of the regression output layer and the classification output layer, which are used to balance the losses between the two. N represents the batch size. represents the true value, represents the model prediction value, represents the true label, Represents the model prediction probability.

7. The visual health assessment method of university campus landscape according to claim 1 is characterized in that: The step of inputting the visual health assessment dataset into a pre-built deep learning model for training specifically includes: Determine the settings of hyperparameters of pre-built deep learning models, including the learning rate , batch size N, number of training epochs and dropout ratio; The Adam optimizer is used to optimize the deep learning model with set hyperparameters. The optimization calculation formula satisfies in, represents the parameter value of iteration t+1, represents the parameter value of iteration t, and denote the estimates of the first and second moments of the gradient, respectively. Represents a small constant to prevent division by zero; The Bayesian optimization algorithm is used to tune the hyperparameters of the deep learning model, and a learning rate scheduler is used to reduce the learning rate when the validation loss is no longer decreasing.

8. The visual health assessment method of university campus landscape according to claim 1 is characterized in that: The step of generating a heat map of the output result of the visual health assessment model to determine the key assessment area of ​​the visual health assessment model for the university campus landscape image specifically includes: Acquire the landscape image data of the first university campus, input the landscape image data of the first university campus into the visual health assessment model, and obtain a visual health score; Based on the visual health score, the gradient of the category score of each campus landscape assessment feature to the last convolutional layer feature map of the visual health assessment model and the gradient average of the last convolutional layer are calculated, and the feature map is weighted according to the feature map weighting formula to obtain a weighted result. The feature map weighting formula satisfies in, Represents the average gradient of category c, reflecting the influence of this category on the feature map. Represents the model's prediction score or output value for category c, Indicates the last convolutional layer Row, No. The activation value of a channel of the feature map of the column, Represents the spatial dimension of the feature map, used to normalize the sum of gradients, Represents the score feature map of category c The partial derivative of reflects the influence of each position in the feature map on the final classification result; A heat map is generated according to the weighted result, and the heat map is normalized. The normalized heat map and the original image of the first university campus landscape image data are then superimposed to determine the key assessment area of ​​the visual health assessment model for the university campus landscape image.

9. An evaluation system for implementing the visual health evaluation method of a university campus landscape according to any one of claims 1 to 8, characterized in that: The system specifically comprises: The first evaluation module is used to obtain university campus landscape image data, score the university campus landscape image data based on campus landscape evaluation characteristics, and use Cohen's Kappa statistical method to check the consistency of the score to obtain score data; A second evaluation module is used to associate and match the university campus landscape image data with the scoring data to form a visual health evaluation data set; A third assessment module is used to input the visual health assessment data set into a pre-built deep learning model for training to generate a visual health assessment model; The fourth evaluation module is used to generate a heat map of the output results of the visual health evaluation model to determine the key evaluation area of ​​the visual health evaluation model for the university campus landscape image.

10. A computer device, characterized in that: include: A memory and a processor and a computer program stored in the memory, when the computer program is executed on the processor, implements a visual health assessment method for university campus landscape as described in any one of claims 1 to 8.

Images