Deep Learning-Based Green Space Service Module Identification and Optimization Method and System

Abstract

The present invention discloses a method and system for identifying and optimizing green space service modules based on deep learning, which relates to the technical fields of landscape architecture, urban planning and artificial intelligence, and includes the following steps: receiving target green space real-scene image data, wherein the target green space real-scene image data is obtained by determining the distribution range of the target green space; inputting the target green space real-scene image into a pre-established green space service module identification model, and identifying the green space service modules in each target green space real-scene image; counting the green space service modules, and optimizing the counted green space service modules from two aspects: the single green space level and the green space group level; the present invention introduces deep learning technology, and by inputting green space real-scene images, it can realize the rapid and large-scale identification of various types of green space service modules.

Description

Technical Field

[0001] The present invention relates to the technical fields of landscape architecture, urban planning and artificial intelligence, specifically a method and system for identifying and optimizing green space service modules based on deep learning. Background Art

[0002] To ensure the sustainable development of cities, urban sprawl is strictly restricted, and urban renewal oriented towards optimizing stock space has become the focus of development in many high-density cities. The rationality and suitability of the configuration of green space service functions in the built environment have become the current research and practice hotspots, as well as the focus of attention in urban renewal practice. However, the existing adjustment of green space service functions usually studies the service status of green space in the built environment through on-site investigations and other means. The entire process requires a large amount of manpower, material resources and financial resources, and has large spatio-temporal limitations, making it difficult to carry out synchronously on a large scale within the city.

[0003] The existing patent CN110288202A discloses a method for evaluating and optimizing the status of urban park green space facilities, including extracting basic data of geographical national conditions elements and traffic big data and integrating them. The basic data includes community data, park data and statistical unit data; calculating 6 basic statistical indicators, including the park green space supply quantity indicator at the statistical unit scale, the green space supply quality indicator at the park scale, the park green space service capacity indicator at the statistical unit scale, the accessibility and service capacity indicator at the park scale, the accessibility and supply equalization indicator at the community scale, and the park green space supply equalization indicator at the research area scale; based on the calculated basic statistical indicators, conducting cluster comprehensive analysis; based on traffic big data, identifying blind areas according to the distribution of parks and communities within the city; based on the blind areas and traffic big data, using kernel density analysis to obtain areas where blind areas are more concentrated, and using the particle swarm optimization algorithm to obtain the optimization result of the location selection of newly built parks;

[0004] The existing patent has the following problems: It does not explore the facilities inside the green space: The existing technology starts from the spatial distribution situation and only evaluates the planar morphological structure of the green space, without covering the green space service level situation; It does not meet the requirements of large-scale data collection: Most of the existing data for evaluating the inside of the green space starts from on-site investigations, making it difficult to conduct large-scale evaluation of the entire green space. Summary of the Invention

[0005] To solve the deficiencies mentioned in the above background art, the purpose of the present invention is to provide a method and system for identifying and optimizing green space service modules based on deep learning, which can automatically identify the service function modules borne by the current green space, evaluate the service level of individual green spaces and the overall service situation of green space groups, and provide direction guidance for subsequent supplementation and improvement of green space functions.

[0006] The object of the present invention can be achieved by the following technical solutions: a method for identifying and optimizing a green space service module based on deep learning, the method comprising the following steps:

[0007] Receiving target green space real-scene image data, wherein the target green space real-scene image data is obtained by determining the distribution range of the target green space;

[0008] Inputting the target green space real-scene image data into a pre-established green space service module recognition model, and identifying the green space service modules in each piece of target green space real-scene image data;

[0009] Statistical analysis is performed on the green space service modules, and the statistically analyzed green space service modules are optimized from two aspects: the single green space level and the green space group level.

[0010] Preferably, the distribution range of the target green space is obtained by interpreting remote sensing images or based on relevant planning floor plans.

[0011] Preferably, the target green space real-scene image data includes at least one of real-scene image data collected based on Internet street view data and reflecting the corresponding green space, real-scene image data taken by users on-site and spontaneously uploaded to social media and reflecting the corresponding green space, and real-scene image data taken during field research and reflecting the corresponding green space.

[0012] Preferably, the green space service module recognition model is built based on the YOLOV5 model and trained using training samples in the green space service module training set.

[0013] Preferably, the green space service module training set is made by constructing a green space service module system.

[0014] Preferably, according to the green space service module system, the Labelimg tool is used, and the YOLOV5 mode is used to manually annotate the target green space real-scene images to obtain annotation samples, and the annotation samples are summarized to construct the green space service module training set.

[0015] Preferably, the process of building the green space service module recognition model based on the YOLOV5 model is as follows:

[0016] Build a green space service module recognition model based on the YOLOV5 model, train the built green space service module recognition model with training samples, input the test set data into the green space service module recognition model, and use the mean average precision (mAP) to evaluate whether the model accuracy meets the standard. If the mean average precision (mAP) meets the standard under the condition that the interaction ratio reaches a certain threshold requirement, the construction of the green space service module recognition model is completed; if not, by increasing the number of samples and adjusting the sample types, remake the training samples of the green space service module training set of the non-compliant types, so as to increase the recognition accuracy of the corresponding types.

[0017] Preferably, the accuracy evaluation index includes the mean average precision (mAP). The mean average precision is the area under the precision-recall curve at the intersection over union (IoU) threshold, and the average of the average precisions of all classes is calculated to obtain the mean average precision. When the mean average precision (mAP) reaches above 0.75, the construction of the green space service module recognition model is completed.

[0018] Preferably, the process of increasing the accuracy of the green space service module recognition model includes non-maximum suppression (NMS) and data augmentation techniques. Non-maximum suppression (NMS) means that for each category in all detection boxes, sort them in descending order according to the predicted scores, and then start from the box with the highest score, and successively judge whether the intersection over union (IoU) with the subsequent boxes is greater than the threshold. If it is greater than the threshold, delete the subsequent boxes. The data augmentation technique refers to increasing the number of samples by means of random cropping, random rotation, random flipping, random brightness transformation, etc., and adjusting the samples with lower classification accuracy by deleting and splitting samples. Among them, random cropping means randomly cropping a part of the original image as the training image to increase the training data volume and enrich the data distribution. Random rotation means randomly rotating the image by a certain angle to increase the model's learning ability of image rotation invariance. Random flipping means randomly flipping the image horizontally or vertically to increase the model's learning ability of image flipping invariance. Random brightness transformation means randomly adjusting the brightness, contrast and other attributes of the image, which can increase the model's adaptability to different lighting conditions.

[0019] In another aspect of the present invention, in order to achieve the above object, a green space service module recognition and optimization system based on deep learning is disclosed, including:

[0020] Data receiving module: used to receive the target green space real scene image data, where the target green space real scene image data is obtained by determining the distribution range of the target green space;

[0021] Recognition module: It is used to input the target green space real scene image data into the pre-established recognition model of the green space service module, and identify the green space service modules in each piece of the target green space real scene image data through recognition;

[0022] Optimization module: It is used to count the service modules included in the green space, and optimize the statistically obtained green space service modules from two aspects: the single green space level and the green space group level.

[0023] Advantages of the present invention:

[0024] The present invention introduces deep learning technology. By inputting the green space real scene image data, it can realize the rapid and large-scale recognition of the green space service modules, and on this basis, realize the optimization of the single green space and the group. Description of the drawings

[0025] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, for those of ordinary skill in the art, without creative efforts, other drawings can also be obtained according to these drawings;

[0026] Figure 1 It is a schematic diagram of the method flow of the present invention;

[0027] Figure 2 It is a schematic diagram of the overall process of the present invention;

[0028] Figure 3 It is a schematic diagram of the production process of the green space service module training set provided by the present invention;

[0029] Figure 4 It is a schematic diagram of the training evaluation index of the green space service module recognition model provided by the present invention

[0030] Figure 5 It is a schematic diagram of the output training result of the green space service module recognition model provided by the present invention

[0031] Figure 6 It is a schematic diagram of the system structure of the present invention. Detailed implementation manners

[0032] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0033] As Figure 1 shown, the method for identifying and optimizing green space service modules based on deep learning includes the following steps:

[0034] Receive the real-scene image data of the target green space, where the real-scene image data of the target green space is obtained by determining the distribution range of the target green space;

[0035] Input the real-scene image data of the target green space into the pre-established green space service module recognition model, and identify the green space service modules in each piece of real-scene image data of the target green space;

[0036] Statistically analyze the green space service modules, and optimize the statistically analyzed green space service modules from two aspects: the single green space level and the green space group level.

[0037] The distribution range of the target green space is obtained by interpreting remote sensing images or based on relevant planning floor plans;

[0038] In this example, collect the urban remote sensing images of the research area and perform a series of processing including geocalibration, image correction, and band synthesis to obtain standard false-color remote sensing images. On this basis, use the support vector machine model to identify the green space; and number the obtained green space and name it G1, G2... Gn.

[0039] The real-scene image data of the target green space includes at least one of the real-scene image data collected based on Internet street view data and reflecting the corresponding green space, or the real-scene image data taken on-site by users and spontaneously uploaded to social media and reflecting the corresponding green space, and the real-scene image data taken during field research and reflecting the corresponding green space;

[0040] In this example, use the Internet street view platform to obtain the real-scene image data of the green space at the corresponding location. Select the roads with street view images within a 10m buffer range of the green space G1, and take street view photos perpendicular to the driving direction and facing the green space G1 at intervals of 50m, and number the obtained green space photos and name them G1-1, G1-2... G1-n. Repeat this step to obtain the photos of the green spaces G2, G3... Gn and construct a real-scene image data dataset.

[0041] The green space service module recognition model is built based on the YOLOV5 model and trained using the training samples in the green space service module training set;

[0042] The training set of the green space service module is constructed according to the constructed green space service module system. Using the Labelimg tool and the YOLO mode, manual annotation is carried out on the green space real-scene image data. The number of annotation samples for each service module is not less than 300. The annotation samples are summarized to construct the training set of the green space service module; the green space service module is divided into a training set and a test set according to the ratio of 8:2.

[0043] In this example, the manual annotation method using the Labelimg tool is as Figure 3 shown. After collecting the real-scene image data of the green space, manual annotation is carried out on the green space real-scene image data using the YOLO mode. When manually annotating, all service modules appearing in the real scene should be framed as much as possible, and the selection box should fit the object as much as possible.

[0044] Construction of the green space service module system:

[0045] Refer to relevant design specifications and construction guidelines such as the "Park Design Specification", and construct the green space service module system in combination with the actual situation of the site, and define the service facilities required for the service module. The service facilities include at least one recognition object such as activity venues, rest seats, architectural pavilions, landscape waterscapes, and natural plants.

[0046] In this example, according to the "Construction Guide for Pocket Parks in a Certain City" and in combination with the actual situation of a certain city, a green space service module system covering 12 service contents such as children's activities, equipment fitness, rest and sitting, cultural display, professional sports, lawn activities, artificial water landscape appreciation, natural water landscape appreciation, pavilion shelter from rain, arbor shade, plant appreciation, and square activities is constructed, and the service facilities specifically borne by it are defined. For example, the service facilities corresponding to the children's activity service module are children's outdoor activity facilities, and the service facilities corresponding to the equipment fitness service module include common outdoor fitness facilities such as horizontal bars and parallel bars.

[0047] It should be further noted that in the specific implementation process, a green space service module recognition model is built based on the YOLOV5 model, the built green space service module recognition model is trained using training samples, the training results of the model are evaluated using training indicators, and the test set data is input into the model to evaluate the performance of the trained model. The mean Average Precision (mAP) is used to evaluate whether the model performance meets the standard. If the mean Average Precision meets the standard under the condition that the Intersection over Union (IoU) reaches a certain threshold requirement, the construction of the green space service module recognition model is completed. If not, return to S32 to re-make the green space service module training samples of the non-compliant type by increasing the number of samples and adjusting the sample type, and re-train the model until it meets the standard.

[0048] The evaluation indicators in the training session of this example are asFigure 5 As shown below, the meanings of each index are as follows:

[0049] Detection box (Box): During the training and detection process of YOLO, GIoU (Generalized Intersection over Union) is used as the loss function for the prediction box, and Box is the result of averaging the GIoU loss function. The smaller the Box value, the more accurate the target prediction. Among them, GIoU is used to comprehensively evaluate the overlap degree between the training box and the prediction box of the two bounding boxes, and the calculation formula is formula (1):

[0050] GIoU = IoU - (C - U) / C (1)

[0051] Among them, IoU represents the intersection over union, that is, the intersection area of the two bounding boxes divided by their union area, C represents the area of the smallest convex hull of the two bounding boxes, and U represents the area of the union of the two bounding boxes.

[0052] Objectness score: The mean value of the loss function (GIoU) when YOLO performs object detection. The smaller the Objectness value, the more accurate the target prediction.

[0053] Classification score: The mean value of the loss function (GIoU) when YOLO performs object classification. The smaller the Classification value, the more accurate the target prediction.

[0054] Precision: Precision = correctly found positive classes / all found positive classes, and the calculation formula is formula (1). TP represents the number of positive class samples in the true label, and FP represents the number of negative class samples in the true label. It measures the probability that the positive classes separated by a classifier are correct. The larger the precision, the higher the probability of being correct. The formula is as follows:

[0055]

[0056] Recall: Recall = correctly found positive classes / all positive classes that should have been correctly found, and the calculation formula is formula (3). TP represents the number of positive class samples in the true label, and FP represents the number of positive classes mispredicted as negative classes. It measures the ability of a classifier to find all positive classes. The larger the recall value, the stronger the ability to find all positive classes. The formula is as follows:

[0057]

[0058] mAP@0.5: It represents the value of mean average precision (mAP). The calculation formula is formula (4). For multi-class object detection tasks, calculate the average precision (AP) for each class, and then take the average value to obtain mAP, which can comprehensively evaluate the performance of the model in multi-class object detection. Among them, @0.5 means using the Intersection over Union (IoU) threshold of 0.5 as the standard for judging whether the target box is correct.

[0059]

[0060] Among them, the calculation method of the average precision (AP) for each class is to sort according to the confidence of the prediction results. Then, for each confidence threshold, calculate the precision and recall at this threshold. According to the precision-recall curve under different thresholds, calculate the area under the curve. Finally, average the areas under the curves under all confidence thresholds to obtain the average precision.

[0061] mAP@0.5:0.95: It means that when calculating mAP, use all values in the range of the Intersection over Union (IoU) threshold from 0.5 to 0.95, and take the average value of mAP for all thresholds.

[0062] In this example, after 300 rounds of training, the relevant indicators in the training session show that the constructed model has high accuracy and precision in the Xu Lian result. The service module in the green space real scene image can be recognized through the object detection recognition method, and the position and confidence of the service module in the real scene image in the image can be given.

[0063] In the model performance evaluation session of this example, use the mean average precision (mAP) to evaluate whether the model performance meets the standard. The calculation formula is formula (4). For multi-class object detection tasks, calculate the average precision (AP) for each class, and then take the average value to obtain mAP, which can comprehensively evaluate the performance of the model in multi-class object detection.

[0064] In this example, the mean average precision (mAP) will be calculated under the condition that the intersection over union (IoU) threshold is 0.75. 20% (about 100 images) of the already labeled but untrained images are randomly selected from the database as the test set, and this test set is used to test the performance of the model. The test results show that under the condition that the intersection over union (IoU) threshold is 0.75, the mean average precision (mAP) is 0.765, indicating that the performance of this model is good and can meet the recognition requirements of the green space service module.

[0065] Inputting the target real-scene image data into the constructed green space service module recognition model can obtain the green space service modules contained in the real-scene image data, such as Figure 5 shown.

[0066] On the other hand, as Figure 6 shown, the embodiment of the present invention also discloses a green space service module recognition and optimization system based on deep learning, including:

[0067] A data receiving module: used to receive target green space real-scene image data, where the target green space real-scene image data is obtained by determining the distribution range of the target green space;

[0068] A recognition module: used to input the target green space real-scene image data into a pre-established green space service module recognition model, and identify the green space service modules in each piece of target green space real-scene image data;

[0069] An optimization module: used to count the green space service modules, and optimize the counted green space service modules from two aspects: the single green space level and the green space group level.

[0070] Based on the same inventive concept, the present invention also provides a computer device, which includes: one or more processors, and a memory for storing one or more computer programs; the program includes program instructions, and the processor is configured to execute the program instructions stored in the memory. The processor may be a Central Processing Unit (CPU), or may also be other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, which is used to implement one or more instructions, specifically for loading and executing one or more instructions in the computer storage medium to implement the above method.

[0071] It should be further noted that, based on the same inventive concept, the present invention also provides a computer storage medium, on which a computer program is stored, and the computer program, when run by a processor, executes the above method. The storage medium may be any combination of one or more computer-readable media. The computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium may, for example, but not be limited to, an electrical, magnetic, optical, electro-magnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples (non-exhaustive list) of the computer-readable storage medium include: an electrical connection having one or more wires, a portable computer disk, a hard disk, a Random Access Memory (RAM), a Read-Only Memory (ROM), an Erasable Programmable Read-Only Memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present invention, the computer-readable storage medium may be any tangible medium that contains or stores a program, and the program may be used by or combined with an instruction execution system, apparatus, or device.

[0072] In the description of this specification, the description with reference to terms such as "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present disclosure. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0073] The above has shown and described the basic principles, main features and advantages of the present disclosure. Those skilled in the art should understand that the present disclosure is not limited by the above embodiments. What is described in the above embodiments and the specification only illustrates the principles of the present disclosure. Without departing from the spirit and scope of the present disclosure, the present disclosure will have various changes and improvements, and these changes and improvements all fall within the scope of the present disclosure claimed.

Claims

1. A method for identifying and optimizing a green space service module based on deep learning, characterized in that, The method includes the following steps: Receiving target green space real-scene image data, where the target green space real-scene image data is obtained by determining the distribution range of the target green space; Inputting the target green space real-scene image data into a pre-established green space service module recognition model, and identifying the green space service modules in each piece of the target green space real-scene image data through recognition; The green space service module recognition model is built based on the YOLOV5 model and trained using the training samples in the green space service module training set; The green space service module training set is made by constructing a green space service module system; According to the green space service module system, using the Labelimg tool, performing manual annotation on the target green space real-scene image data in the YOLO mode to obtain annotation samples, and summarizing the annotation samples to construct a green space service module training set; The process of building the green space service module recognition model based on the YOLOV5 model is as follows: Building a green space service module recognition model based on the YOLOV5 model, training the built green space service module recognition model using the training samples, inputting the test set data into the green space service module recognition model, and using the mean average precision mAP to check whether the recognition accuracy of the output green space service module recognition model meets the requirements. If it meets the standard, the construction of the green space service module recognition model is completed; if it does not meet the standard, by increasing the number of samples and adjusting the sample types, remaking the training samples of the green space service module training set of the non-compliant type, so as to increase the recognition accuracy; Statistical analysis is performed on the green space service modules, and the statistically analyzed green space service modules are optimized from two aspects: the single green space level and the green space group level.

2. The method for identifying and optimizing a green space service module based on deep learning according to claim 1, wherein The distribution range of the target green space is obtained by interpreting remote sensing images or based on relevant planning floor plans.

3. The method for identifying and optimizing a green space service module based on deep learning according to claim 1, wherein The target green space real-scene image data includes at least one of the real-scene image data collected based on Internet street view data and reflecting the corresponding green space, or the real-scene image data taken on-site by users and spontaneously uploaded to social media and reflecting the corresponding green space, or the real-scene image data taken on-site by users and spontaneously uploaded to social media and reflecting the corresponding green space, and the real-scene image data obtained by field research and reflecting the corresponding green space.

4. The method for identifying and optimizing the green space service module based on deep learning according to claim 1, wherein, The accuracy evaluation index includes the mean average precision mAP. The mean average precision is the area under the precision-recall curve at the intersection over union IoU threshold, and the average precision of all classes is averaged to obtain the mean average precision. When the mean average precision mAP reaches above 0.75, the construction of the green space service module recognition model is completed.

5. The method for identifying and optimizing a green space service module based on deep learning according to claim 1, wherein The process of increasing the accuracy of the green space service module recognition model includes non-maximum suppression (NMS) and data augmentation techniques. Non-maximum suppression (NMS) means that for each category in all detection boxes, they are sorted in descending order according to the predicted scores. Then, starting from the box with the highest score, it is successively determined whether the intersection over union (IoU) with the subsequent boxes is greater than the threshold. If it is greater than the threshold, the subsequent boxes are deleted. The data augmentation technique refers to increasing the number of samples through random cropping, random rotation, random flipping, and random brightness transformation, and adjusting the samples of the type with lower classification accuracy by deleting and splitting samples. Among them, random cropping means randomly cropping a part from the original image as the training image to increase the training data volume and enrich the data distribution. Random rotation means randomly rotating the image by a certain angle to increase the model's learning ability for image rotation invariance. Random flipping means randomly flipping the image horizontally or vertically to increase the model's learning ability for image flipping invariance. Random brightness transformation means randomly adjusting the brightness and contrast attributes of the image to increase the model's adaptability to different lighting conditions.

6. The green space service module recognition and optimization system based on deep learning is characterized in that, Including: Data reception module: used to receive the target green space real-scene image data, where the target green space real-scene image data is obtained by determining the distribution range of the target green space; Recognition module: used to input the target green space real-scene image data into the pre-established green space service module recognition model, and identify the green space service modules in each piece of target green space real-scene image data through recognition; The green space service module recognition model is built based on the YOLOV5 model and trained using the training samples in the green space service module training set; The green space service module training set is made by constructing a green space service module system; According to the green space service module system, using the Labelimg tool and the YOLO mode, the target green space real-scene image data is manually annotated to obtain annotated samples, and the annotated samples are summarized to construct the green space service module training set; The process of building the green space service module recognition model based on the YOLOV5 model is as follows: Build the green space service module recognition model based on the YOLOV5 model, train the built green space service module recognition model using the training samples, input the test set data into the green space service module recognition model, and use the mean average precision (mAP) to check whether the recognition accuracy of the output green space service module recognition model meets the requirements. If it meets the standard, the construction of the green space service module recognition model is completed; if it does not meet the standard, the training samples of the green space service module training set of the non-compliant type are remade by increasing the number of samples and adjusting the sample types, so as to increase the recognition accuracy; Optimization module: used to count the green space service modules, and optimize the counted green space service modules from two aspects: the single green space level and the green space group level.

Images