A method, device and storage medium for monitoring the displacement of soil mass on the slope of a deep foundation pit

Abstract

This application proposes a method, device, and storage medium for monitoring the displacement of soil mass on the slope of a deep foundation pit. The method includes: obtaining real-time high-definition images covering the slope of the target deep foundation pit and environmental sensing data of the slope of the target deep foundation pit; preprocessing the real-time high-definition images to obtain the processed real-time high-definition images; calling the trained improved YOLO model to perform recognition and analysis on the processed real-time high-definition images to obtain the visual information of the slope of the target deep foundation pit; fusing and calculating the visual information and environmental sensing data to obtain the final displacement data of the slope of the target deep foundation pit. This application can achieve the effects of realizing the automation level of monitoring the slope of a deep foundation pit, expanding the monitoring range and accuracy of the slope of a deep foundation pit, and being able to timely identify safety risks such as soil cracks and soil displacement on the slope of a deep foundation pit.

Description

Technical Field

[0001] This application relates to the technical field of deep foundation pit engineering, and particularly to a method, device and storage medium for monitoring the displacement of soil mass on the slope of a deep foundation pit. Background Art

[0002] In the field of deep foundation pit engineering, monitoring the displacement and stability of the foundation pit is a crucial task to ensure construction safety and structural integrity.

[0003] Currently, most traditional monitoring methods are used to monitor the displacement of the foundation pit. The traditional monitoring methods use physical measurement tools (such as inclinometers, displacement meters, etc.) to monitor the displacement of the foundation pit. Although these physical measurement tools can provide accurate displacement data, there are some limitations, such as the following limitations:

[0004] (1) Traditional monitoring methods often require manual regular inspections, which are not only time-consuming but also may cause errors due to human factors.

[0005] (2) Physical measurement tools usually can only monitor at a fixed number of points, making it difficult to achieve full coverage. Moreover, due to the delay in the data collection and processing process, it is difficult to achieve real-time monitoring and early warning. Summary of the Invention

[0006] Embodiments of this application provide a method, device and storage medium for monitoring the displacement of soil mass on the slope of a deep foundation pit to solve the problems existing in the related technologies. The technical solutions are as follows:

[0007] In a first aspect, embodiments of this application provide a method for monitoring the displacement of soil mass on the slope of a deep foundation pit, including:

[0008] Obtain real-time high-definition images covering the slope of the target deep foundation pit and environmental sensing data of the target deep foundation pit;

[0009] Preprocess the real-time high-definition images to obtain processed real-time high-definition images;

[0010] Call the trained improved YOLO model to perform recognition and analysis on the processed real-time high-definition images to obtain visual information of the target deep foundation pit slope, where the visual information includes displacement recognition data;

[0011] Fuse and calculate the visual information and the environmental sensing data to obtain the final displacement data of the target deep foundation pit slope.

[0012] In an implementation manner, obtaining real-time high-definition images covering the slope of the target deep foundation pit and environmental sensing data of the target deep foundation pit includes:

[0013] Regularly capture the slope of the target deep foundation pit using a high-resolution camera to obtain the real-time high-definition image, and the shooting range of the high-resolution camera covers the slope of the target deep foundation pit;

[0014] Collect relevant data of the slope of the target deep foundation pit using an environmental sensor to obtain the environmental sensing data.

[0015] In one implementation, the preprocessing of the real-time high-definition image to obtain the processed real-time high-definition image includes:

[0016] Denoise, adjust brightness and contrast, and perform cropping and scaling on the real-time high-definition image to obtain a preliminarily processed real-time high-definition image;

[0017] Perform normalization on the preliminarily processed real-time high-definition image to obtain the processed real-time high-definition image.

[0018] In one implementation, the trained improved YOLO model is obtained through the following training process:

[0019] Obtain an image sample set, which includes multiple deep foundation pit slope images;

[0020] Preprocess each deep foundation pit slope image in the image sample set to obtain each processed deep foundation pit slope image;

[0021] Mark each processed deep foundation pit slope image, and label the key features in each deep foundation pit slope image to obtain a labeled image sample set, and the key features include soil cracks and soil displacements;

[0022] Train a preset YOLO model based on the labeled image sample set, and during the training process, use GPU to accelerate the training process, and continuously adjust the network structure and loss function of the preset YOLO model to obtain the trained improved YOLO model.

[0023] In one implementation, the adjustment of the network structure of the preset YOLO model includes: adjusting the aspect ratio and size of the anchor boxes of the preset YOLO model, the number of convolutional layers of the preset YOLO model, the filter size of the preset YOLO model, and the image input size of the preset YOLO model.

[0024] In one implementation, the adjustment of the loss function of the preset YOLO model includes: adjusting the position loss of the loss function, the size loss of the loss function, the confidence loss of the loss function, and the class probability loss of the loss function.

[0025] In one embodiment, fusing and calculating the visual information and the environmental sensing data to obtain the final displacement data of the target deep foundation pit slope includes:

[0026] Substituting the displacement recognition data in the visual information and the displacement sensing data in the environmental sensing data into a specified formula for fusion calculation to obtain the final displacement data.

[0027] In a second aspect, an embodiment of the present application further provides a device for monitoring the displacement of the soil body of a deep foundation pit slope, including:

[0028] A processing unit, configured to obtain real-time high-definition images covering the target deep foundation pit slope and the environmental sensing data of the target deep foundation pit slope; preprocess the real-time high-definition images to obtain preprocessed real-time high-definition images;

[0029] A monitoring unit, configured to call the trained improved YOLO model to perform recognition and analysis on the preprocessed real-time high-definition images to obtain the visual information of the target deep foundation pit slope, where the visual information includes displacement recognition data; fuse and calculate the visual information and the environmental sensing data to obtain the final displacement data of the target deep foundation pit slope.

[0030] In one embodiment, when the processing unit is used to obtain the real-time high-definition images covering the target deep foundation pit slope and the environmental sensing data of the target deep foundation pit slope, it is specifically configured to:

[0031] Regularly photograph the target deep foundation pit slope using a high-resolution camera to obtain the real-time high-definition images, where the shooting range of the high-resolution camera covers the target deep foundation pit slope;

[0032] Collect relevant data of the target deep foundation pit slope using an environmental sensor to obtain the environmental sensing data.

[0033] In one embodiment, when the processing unit is used to preprocess the real-time high-definition images to obtain preprocessed real-time high-definition images, it is specifically configured to:

[0034] Perform denoising, brightness and contrast adjustment, cropping and scaling processing on the real-time high-definition images to obtain preliminarily processed real-time high-definition images;

[0035] Perform normalization processing on the preliminarily processed real-time high-definition images to obtain the preprocessed real-time high-definition images.

[0036] In one embodiment, the trained improved YOLO model is obtained by the monitoring unit through the following training process:

[0037] Obtain an image sample set, where the image sample set includes multiple deep foundation pit slope images;

[0038] Preprocess each deep foundation pit slope image in the image sample set to obtain each preprocessed deep foundation pit slope image;

[0039] Mark each preprocessed deep foundation pit slope image, and label the key features in each deep foundation pit slope image to obtain a labeled image sample set, where the key features include soil cracks and soil displacements;

[0040] Train a preset YOLO model based on the labeled image sample set, and during the training process, use GPU to accelerate the training process, and continuously adjust the network structure and loss function of the preset YOLO model to obtain the trained and improved YOLO model.

[0041] In one implementation, the adjustment of the network structure of the preset YOLO model includes: adjusting the aspect ratio and size of the anchor boxes of the preset YOLO model, the number of convolutional layers of the preset YOLO model, the filter size of the preset YOLO model, and the image input size of the preset YOLO model.

[0042] In one implementation, the adjustment of the loss function of the preset YOLO model includes: adjusting the position loss of the loss function, the size loss of the loss function, the confidence loss of the loss function, and the class probability loss of the loss function.

[0043] In one implementation, when the monitoring unit is used to perform fusion calculation on the visual information and the environmental sensing data to obtain the final displacement data of the target deep foundation pit slope, it is specifically used for:

[0044] Substitute the displacement recognition data in the visual information and the displacement sensing data in the environmental sensing data into a specified formula for fusion calculation to obtain the final displacement data.

[0045] In a third aspect, an embodiment of the present application further provides a computer device, which includes: a memory and a processor, instructions are stored in the memory, and the instructions are loaded and executed by the processor to implement the method in any one of the above aspects, wherein the memory and the processor communicate with each other through an internal connection path.

[0046] In a fourth aspect, an embodiment of the present application further provides a computer-readable storage medium, in which a computer program is stored, and when the computer program runs on a computer, the method in any one of the above aspects is implemented.

[0047] The advantages or beneficial effects in the above technical solutions at least include:

[0048] I. Improving the level of monitoring automation: By using a trained deep learning model (such as the improved YOLO model that has been trained), automatically analyze the real-time high-definition images of the real-time target deep foundation pit slope, reduce manual intervention, and improve the automation level of the monitoring process.

[0049] II. Expanding the monitoring range and accuracy: Obtain real-time high-definition images covering the target deep foundation pit slope. For example, high-resolution cameras can be used to cover a wider area, and combined with the high-precision recognition function of the improved YOLO model that has been trained, achieve fine-grained monitoring of the entire deep foundation pit area.

[0050] III. Implementing real-time data processing and early warning: The trained improved YOLO model can immediately process image data, timely identify safety risks such as soil cracks and soil displacements on the deep foundation pit slope, quickly issue an early warning, and greatly improve the response speed and timeliness.

[0051] IV. Enhancing safety: Real-time and comprehensive monitoring can timely detect potential risks, thereby taking corresponding safety measures, reducing the probability of accidents, and protecting the safety of workers and equipment.

[0052] V. Reducing costs: Automated monitoring reduces the dependence on manpower. At the same time, quickly identifying and responding to potential problems reduces the costs of long-term maintenance and emergency handling.

[0053] VI. Enhancing the accuracy and reliability of monitoring data: By combining multi-source data through data fusion technology, the comprehensiveness and accuracy of monitoring data are improved, making engineering management decisions more scientific and effective.

[0054] The above summary is only for the purpose of the specification and is not intended to be limiting in any way. In addition to the illustrative aspects, embodiments, and features described above, further aspects, embodiments, and features of the present application will be readily apparent by reference to the drawings and the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In the drawings, unless otherwise specified, the same reference numerals throughout the several views denote the same or similar components or elements. These drawings are not necessarily drawn to scale. It should be understood that these drawings only depict some embodiments disclosed in the present application and should not be regarded as limiting the scope of the present application.

[0056] Figure 1 It is a schematic flow chart of a method for monitoring soil displacement of a deep foundation pit slope provided by an embodiment of the present application;

[0057] Figure 2 An example of the on-site image of a deep foundation pit slope provided by an embodiment of the present application;

[0058] Figure 3 Another example of the on-site image of a deep foundation pit slope provided by an embodiment of the present application;

[0059] Figure 4 An example of the performance graph during the training of an improved YOLO model provided by an embodiment of the present application;

[0060] Figure 5 Another example of the performance graph during the training of an improved YOLO model provided by an embodiment of the present application;

[0061] Figure 6 An example of the data analysis graph after the training of an improved YOLO model provided by an embodiment of the present application;

[0062] Figure 7 An example of the recognition result graph of an improved YOLO model provided by an embodiment of the present application;

[0063] Figure 8 Another example of the recognition result graph of an improved YOLO model provided by an embodiment of the present application;

[0064] Figure 9 Another example of the recognition result graph of an improved YOLO model provided by an embodiment of the present application;

[0065] Figure 10 A structural block diagram of a soil displacement monitoring device for a deep foundation pit slope provided by an embodiment of the present application;

[0066] Figure 11 A structural block diagram of a computer device provided by an embodiment of the present application. Detailed implementation manners

[0067] In the following, only some exemplary embodiments are briefly described. As those skilled in the art can recognize, the described embodiments can be modified in various different ways without departing from the spirit or scope of the present application. Therefore, the drawings and descriptions are considered to be exemplary in nature rather than restrictive.

[0068] Figure 1 A flowchart showing a method for monitoring soil displacement of a deep foundation pit slope according to an embodiment of the present application.

[0069] As Figure 1 shown, the method may include the following steps:

[0070] S110. Obtain real-time high-definition images covering the slope of the target deep foundation pit and environmental sensing data of the slope of the target deep foundation pit.

[0071] In one implementation, a high-resolution camera can be used to periodically capture the slope of the target deep foundation pit to obtain the real-time high-definition images. Among them, the shooting range of the high-resolution camera covers the slope of the target deep foundation pit. In this way, minute cracks and subtle deformations on the slope of the target deep foundation pit can be captured by the high-resolution camera, facilitating subsequent improvement of the accuracy of the deep learning model (such as the improved YOLO model) in identifying cracks and the ability to distinguish small cracks. Among them, the real-time high-definition images can be color images in RGB three-channel format.

[0072] As an example, the high-resolution camera can be a camera with at least 20 million pixels to have sufficient resolution to capture minute cracks and subtle deformations on the slope of the target deep foundation pit.

[0073] In specific implementation, the image acquisition frequency of the high-resolution camera can be adjusted according to the activity level and historical data of the deep foundation pit project. For example, it can be taken once a day or once an hour during the period of frequent activities. The embodiments of the present application do not limit this.

[0074] In one implementation, environmental sensors can be used to collect relevant data of the slope of the target deep foundation pit to obtain the environmental sensing data. Among them, the environmental sensing data can include displacement sensing data of the slope of the target deep foundation pit. That is, the environmental sensors can include displacement sensors.

[0075] As an example, the environmental sensing data can also include sensing data such as temperature, humidity, and rainfall. These sensing data are crucial for analyzing the environmental factor impacts on the soil cracks and soil displacements of the slope of the target deep foundation pit. That is, the environmental sensors can also include temperature sensors, humidity sensors, and rainfall sensors.

[0076] S120. Preprocess the real-time high-definition images to obtain processed real-time high-definition images.

[0077] In one implementation, the real-time high-definition images can be denoised, adjusted in brightness and contrast, and cropped and scaled to obtain preliminarily processed real-time high-definition images.

[0078] In specific implementation, an image denoising algorithm (such as the Gaussian blur algorithm) or a median filter can be used to remove the noise points in the real-time high-definition images.

[0079] Exemplarily, when using a median filter to remove noise from the real-time high-definition image, the denoising process can be as follows: Call an image editing software or a dedicated image processing tool to open the real-time high-definition image; Select the denoising tool, i.e., the median filter, to denoise the real-time high-definition image, observe the denoising effect of the real-time high-definition image, and stop denoising when ensuring that the key features (such as soil cracks and soil displacements) in the real-time high-definition image are clearly visible and the background noise is effectively suppressed.

[0080] In specific implementation, an image editing software can be used to automatically adjust the brightness and contrast of the real-time high-definition image to ensure the consistency of the real-time high-definition image under different lighting conditions.

[0081] Exemplarily, in the image editing software, select the options for adjusting brightness and contrast. Then, according to the performance of the real-time high-definition image under different lighting conditions, gradually adjust the brightness and contrast sliders until the real-time high-definition image reaches an ideal visual effect, and then stop adjusting the brightness and contrast.

[0082] In specific implementation, the real-time high-definition image can be cropped to the area of interest (i.e., the area only containing the target deep foundation pit slope), which can reduce the consumption of computing resources. Then, the cropped real-time high-definition image can be appropriately scaled to meet the input requirements of the trained improved YOLO model.

[0083] Exemplarily, a dedicated image processing tool can be used to select the area of interest of the real-time high-definition image for cropping, remove the unnecessary background part, and focus on key areas such as the foundation pit slope. Then, adjust the size of the cropped real-time high-definition image according to the input requirements of the trained improved YOLO model. For example, if the trained improved YOLO model requires an input size of 416x416 pixels, then the size of the cropped real-time high-definition image can be scaled to this size.

[0084] It should be noted that the embodiments of the present application do not limit the order of the above preprocessing operations, that is, do not limit the order of the three preprocessing processes of image denoising, brightness and contrast adjustment, and cropping and scaling.

[0085] In one implementation manner, by performing normalization processing on the preliminarily processed real-time high-definition image, the processed real-time high-definition image can be obtained.

[0086] Exemplarily, by performing normalization processing on the preliminarily processed real-time high-definition image, the pixel values of the preliminarily processed real-time high-definition image can be scaled from [0, 255] to [0, 1].

[0087] In the embodiments of the present application, by performing step S120, the processed real-time high-definition image can be made more suitable for the input requirements of the trained improved YOLO model, facilitating the improvement of the overall effect and accuracy of the trained improved YOLO model for monitoring the target deep foundation pit slope, and contributing to more efficient and reliable foundation pit slope monitoring.

[0088] S130. Invoke the trained improved YOLO model to perform recognition and analysis on the processed real-time high-definition image to obtain the visual information of the target deep foundation pit slope.

[0089] In one implementation, the visual information includes displacement recognition data. Specifically, the visual information may also include information such as bounding boxes, categories, and confidences.

[0090] As an example, after obtaining the visual information of the target deep foundation pit slope, it can be stored in a database for further analysis and real-time monitoring.

[0091] In one implementation, the trained improved YOLO model can be obtained through the following training process:

[0092] S131. Obtain an image sample set, which includes multiple deep foundation pit slope images.

[0093] Specifically, the multiple deep foundation pit slope images may correspond to the same foundation pit slope or different foundation pit slopes, and the embodiments of the present application do not limit this.

[0094] As an example, the multiple deep foundation pit slope images may be images obtained by a high-resolution camera historically photographing the target deep foundation pit slope.

[0095] S132. Preprocess each deep foundation pit slope image in the image sample set to obtain each processed deep foundation pit slope image.

[0096] Specifically, the implementation process of step S132 may be the same as or similar to the implementation process of the above step S120, and will not be elaborated here.

[0097] As an example, in step S132, image denoising can also be performed by using LabelImg (an open-source image annotation tool) in combination with other tools, and the specific process can be as follows:

[0098] A. Use LabelImg for image annotation

[0099] a1. Open the image: Open LabelImg and load each deep foundation pit slope image to be processed;

[0100] a2. Draw bounding boxes: Use the mouse to draw bounding boxes at the locations of the key features in each deep foundation pit slope image. Add labels to each bounding box, such as "crack" or "displacement", and save the annotation results. Among them, LabelImg will save the coordinates and class information of the bounding boxes in an XML file in the same directory as each deep foundation pit slope image.

[0101] B. Use OpenCV (a cross-platform computer vision library) for image denoising

[0102] b1. Read images: Use OpenCV to read each deep foundation pit slope image annotated by LabelImg;

[0103] b2. Apply median filtering: Use the cv2.medianBlur() function in OpenCV for median filtering denoising. Select an appropriate filter size k, usually 3 or 5;

[0104] b3. Save each processed deep foundation pit slope image: Use the cv2.imwrite() function in OpenCV to save each processed deep foundation pit slope image;

[0105] b4. Observe the effect: Check each processed deep foundation pit slope image to ensure that key features such as soil cracks and soil displacements are clearly visible, while background noise is effectively suppressed.

[0106] In the embodiment of the present application, by performing step S132, it is possible to ensure that each deep foundation pit slope image has a consistent brightness and contrast level after adjustment, so as to reduce the variability in the training of the improved YOLO model, and to ensure that each deep foundation pit slope image after cropping and scaling meets the input size requirements of the improved YOLO model, and by normalizing each deep foundation pit slope image, it helps to improve the training stability and convergence speed of the improved YOLO model.

[0107] That is, by performing step S132, the quality of each deep foundation pit slope image can be optimized, irrelevant noise and variables can be reduced, it can be ensured that each deep foundation pit slope image is more suitable for the requirements of the improved YOLO model, which helps to improve the recognition ability of the improved YOLO model for important features, thereby improving the overall effect and accuracy of deep foundation pit monitoring, and helping to achieve more efficient and reliable foundation pit slope monitoring. For example, through preprocessing, the recognition accuracy of cracks by the improved YOLO model can be increased from the original 70% to more than 85%, greatly improving the recognition accuracy of cracks and the ability to distinguish fine cracks by the improved YOLO model.

[0108] S133. Mark each processed deep foundation pit slope image, mark the key features in each deep foundation pit slope image, and obtain a labeled image sample set.

[0109] In specific implementation, the key features include soil cracks and soil displacements.

[0110] In specific implementation, step S133 may include the following process:

[0111] (1) Annotation process

[0112] Open LabelImg: First, open the LabelImg tool and load the image set to be annotated, that is, the processed image sample set.

[0113] Select an image: Select one processed deep foundation pit slope image from the file list of the processed image sample set.

[0114] Draw a bounding box: Use the mouse to draw a bounding box (such as a rectangular box) at the location of the key features in each processed deep foundation pit slope image, so that key features such as soil cracks and soil displacements in each processed deep foundation pit slope image can be framed; for each drawn bounding box, add a label to this bounding box, such as "crack" or "displacement", and save the coordinates and class information of the bounding box in an XML file in the same directory as each processed deep foundation pit slope image. These XML files will be converted into the text format required by the improved YOLO model.

[0115] (2) Output format of the annotation

[0116] The annotation format required by the improved YOLO model usually includes the class index and four parameters of the bounding box (such as the center coordinates and width and height). These data need to be converted from the XML format to a simple text format, with each line representing an object, and the format is as follows:

[0117] <class_id><x_center><y_center> <width> <height>

[0118] Among them, the coordinates and dimensions are normalized to the interval [0, 1], relative to the width and height of each processed deep foundation pit slope image.

[0119] In specific implementation, by using LabelImg to label each processed deep foundation pit slope image, it is convenient to generate an annotation format compatible with the improved YOLO model.

[0120] It should be noted that during the preprocessing of each deep foundation pit slope image, if LabelImg is combined with other tools for image denoising, it is necessary to annotate each deep foundation pit slope image first and then perform denoising. For details, see the above description.

[0121] S134. Train a preset YOLO model based on the labeled image sample set, and during the training process, use GPU to accelerate the training process, and continuously adjust the network structure and loss function of the preset YOLO model to obtain a trained and improved YOLO model.

[0122] In specific implementation, the adjustment of the network structure of the preset YOLO model includes: adjusting the aspect ratio and size of the anchor boxes of the preset YOLO model, the number of convolutional layers of the preset YOLO model, the filter size of the preset YOLO model, and the image input size of the preset YOLO model.

[0123] As an example, the process of adjusting the aspect ratio and size of the anchor boxes of the preset YOLO model can be as follows:

[0124] Adjust and optimize the aspect ratio and size of the anchor boxes of the preset YOLO model through the following formulas (1) and (2).

[0125] AR = w / h (1)

[0126] S = w × h (2)

[0127] Among them, AR represents the aspect ratio of the anchor box, S represents the size of the anchor box, w represents the width of the anchor box, and h represents the height of the anchor box.

[0128] In specific implementation, adjusting the anchor box strategy in the preset YOLO model is a key step to improve the detection accuracy of soil cracks and soil displacements. In this application, by adjusting and optimizing the aspect ratio and size of the anchor boxes in the improved YOLO model, the sensitivity of the loss function to small objects can be optimized and enhanced, ensuring that the anchor boxes can better match the shape and size of the actual monitoring objects, and improving the monitoring ability for small-scale deformations or cracks. Such adjustments help the trained improved YOLO model to more accurately locate and identify smaller or peculiarly shaped cracks when making bounding box predictions. In other words, the trained improved YOLO model can have accurate target localization ability.

[0129] As an example, the process of adjusting the number of convolutional layers in the preset YOLO model can be as follows: Add additional convolutional layers to the network structure of the preset YOLO model.

[0130] As an example, the process of adjusting the filter size of the preset YOLO model can be as follows: Use smaller filters in these additionally added convolutional layers.

[0131] In specific implementation, by adding convolutional layers and adjusting the filter size in the network structure of the preset YOLO model, the improved YOLO model can identify more subtle textures and changes in the primary processing stage, thereby improving the overall detection performance of the trained improved YOLO model.

[0132] As an example, the process of adjusting the image input size of the preset YOLO model can be as follows:

[0133] Adopt a multi-scale training method by dynamically adjusting the input image size of the preset YOLO model at different scales (k) (for example, Dimension at scale k = 146×1.1 k ), which can improve the real-time processing speed of the improved YOLO model and enable the improved YOLO model to adapt to monitoring objects of different sizes. This not only improves the adaptability of the improved YOLO model to changing sizes but also enhances the application flexibility of the improved YOLO model in actual monitoring scenarios.

[0134] In specific implementation, the loss function of the preset YOLO model consists of several parts, involving position, size, confidence, and classification probability. Based on this, in order to further optimize the performance of the model, especially in terms of accuracy and robustness, the adjustment of the loss function of the preset YOLO model includes: the adjustment of the position loss (error), the size loss (error), the confidence loss (error), and the class probability loss (error) of the loss function. Each part will be introduced in detail below on how to adjust to meet the specific requirements of deep foundation pit monitoring.

[0135] As an example, the process of adjusting the position loss of the loss function can be as follows:

[0136] The position loss of the loss function is represented by the following formula (3).

[0137]

[0138] Among them, L coord represents the position loss of the loss function, is an indicator function used to indicate that the j-th bounding box in the i-th image is responsible for predicting the object, λ coord represents the weight of the position parameter, (x i , y i ) and are the predicted coordinates and the actual coordinates respectively, and B represents the number of bounding boxes predicted in the i-th image.

[0139] In specific implementation, for small-size targets such as soil cracks and soil displacements, it is very important to improve the accuracy of their position prediction. The position loss of the loss function mainly focuses on the accurate prediction of the center position of the bounding box, and the weight of the position loss is usually large because accurate positioning is the key to object detection. Based on this, in the embodiments of the present application, the weight of the center coordinates can be adjusted by adjusting λ coord to strengthen the learning accuracy of the improved YOLO model for the crack position, so that the improved YOLO model pays more attention to these details.

[0140] As an example, the process of adjusting the size loss of the loss function can be as follows:

[0141] The size loss of the loss function is represented by the following formula (4).

[0142]

[0143] Among them, L size represents the size loss of the loss function, λ size用 is for adjusting the weights of width and height prediction, (w i , h i ) represents the predicted width and height expressed as anchor boxes, represents the actual width and height of the anchor boxes.

[0144] In specific implementation, cracks are usually slender, and reducing dimensional errors is particularly important for cracks. Whether the width and height of the bounding box can be accurately predicted is very important for accurately evaluating the actual size of the object, especially when the object size has an important impact on subsequent processing processes (such as tracking or behavior analysis). Based on this, in this application, by adjusting λ size to adjust the weights of the width and height predictions of the anchor boxes to emphasize the sensitivity to small-sized cracks, so that the improved YOLO model can more accurately predict the actual size of the cracks.

[0145] As an example, the process of adjusting the confidence loss of this loss function can be as follows:

[0146] The confidence loss needs to consider both the cases with and without targets. Based on this, the confidence loss of this loss function can be expressed by the following formula (5).

[0147]

[0148] where L conf represents the confidence loss of this loss function, λ noobj represents the weight of the confidence parameter, C i represents the predicted confidence, represents the actual confidence, is an indicator function used to indicate that the i-th bounding box in the i-th image does not contain a target.

[0149] In specific implementation, the confidence indicates whether the predicted bounding box contains a target and the prediction accuracy of the bounding box. Based on this, in this application, by adjusting λ noobj to measure the confidence of whether the bounding box contains a target and the prediction quality of the bounding box, the confidence loss can help the improved YOLO model distinguish between the bounding boxes containing targets and the background, so that the improved YOLO model can improve the prediction accuracy of the bounding boxes.

[0150] As an example, the process of adjusting the class probability loss of this loss function can be as follows:

[0151] The class probability loss of this loss function is expressed by the following formula (6).

[0152]

[0153] where L class represents the class probability loss of this loss function, p ic Denoted as the predicted class probability, Denoted as the actual class probability, λ class Denoted as the weight of the class probability.

[0154] In specific implementation, it may not be the main concern for specific applications because there may be only one class in crack monitoring, but it is still necessary to ensure the accuracy of class prediction. When the improved YOLO model needs to classify the objects within the bounding box, the class probability error becomes particularly important. Based on this, in this application, the class probability can be improved by adjusting λ class so as to ensure that the improved YOLO model can correctly identify and classify the detected objects through confidence loss, enabling the improved YOLO model to improve the detection accuracy of the class of the bounding box.

[0155] In summary, in specific implementation, the loss function of the improved YOLO model is the result of combining the location loss, size loss, confidence loss, and class probability loss. This combination process involves weighted summation of various different types of errors to form a comprehensive metric that reflects the performance of the improved YOLO model in various aspects. This is done to simultaneously optimize the performance of the improved YOLO model in terms of localization accuracy, size prediction, confidence accuracy, and classification effect. It can also be understood that the improved YOLO model divides the input image into an S×S grid, each grid cell predicts B bounding boxes, and each bounding box includes location (center coordinates (x, y), width w, height h), confidence, and class probability distribution.

[0156] Among them, the total loss function of the improved YOLO model can generally be expressed as the following formula (7):

[0157] L = L coord + L size + L conf + L class (7)

[0158] In practical applications, the weights of these losses may be adjusted according to the requirements of specific tasks. For example, in applications with high-precision localization requirements, the value of λ coord may be increased to emphasize the accuracy of location prediction more strongly. Similarly, if classification is not the main concern, the value of λ class may be decreased. In this way, the total loss function can comprehensively reflect the performance of the improved YOLO model in multiple important aspects, thereby guiding the overall optimization of the performance of the improved YOLO model during training. That is, by optimizing the loss function, it directly promotes the improved YOLO model to move towards the expected performance goal.

[0159] In this application, by continuously adjusting the network structure and loss function of the preset YOLO model to improve the preset YOLO model. For example, through the comprehensive improvement in step S134 above, the improved YOLO model is not only theoretically more suitable for the deep foundation pit slope monitoring task, but also can provide higher accuracy and reliability in practical applications. The adjustment of these network structures and the optimization of the loss function work together to significantly improve the adaptability and detection efficiency of the improved YOLO model for complex monitoring scenarios.

[0160] In specific implementation, step S134 can be executed online during idle time, that is, the preset YOLO model can be trained online.

[0161] For example, the process of online training of the preset YOLO model can include the following steps:

[0162] (1) Data preparation:

[0163] (1.1) Prepare a labeled image sample set;

[0164] (1.2) Data loader implementation: Use Python in combination with a deep learning framework (such as PyTorch or TensorFlow) to write a data loading script, which can automatically read image files and annotation data from a directory and convert them into a format suitable for input to the preset YOLO model;

[0165] (1.3) Data augmentation: Implement data augmentation techniques in the data loader, such as random rotation, scaling, color jitter, etc., to enhance the adaptability and generalization ability of the preset YOLO model to various environmental changes.

[0166] (2) Configure training parameters to optimize the training process and model performance:

[0167] (2.1) Batch Size configuration: Select a value between 16 and 32. Considering the GPU memory limit and model complexity, ensure efficient use of computing resources;

[0168] (2.2) Learning Rate configuration: Initially set to 0.001, and adopt a learning rate decay strategy, such as reducing the learning rate by 10% every 20 training epochs, to help the model more finely adjust the weights in the later stage of training;

[0169] (2.3) Epochs configuration: Set to 50 to 100 epochs, and adjust according to the performance of the model on the validation set to avoid overfitting.

[0170] (3) GPU acceleration

[0171] (3.1) Configure the GPU environment: Ensure that all necessary GPU drivers and support libraries are installed and properly configured;

[0172] (3.2) Specify GPU resources: Clearly specify the GPU devices to be used in the training script to optimize resource allocation;

[0173] (3.3) Parallel processing: If conditions permit, use multiple GPUs for data parallel processing to further improve training efficiency.

[0174] In this way, the powerful computing power of the GPU can be utilized to accelerate the training process of the preset YOLO model.

[0175] (4) Training execution

[0176] (4.1) Forward propagation: The preset YOLO model receives a batch of data and performs a series of calculations, including convolution, activation, pooling, etc., and finally outputs prediction results;

[0177] (4.2) Loss calculation: Calculate the total loss based on the output of the preset YOLO model and the true labels, including location loss, size loss, confidence loss, and class probability loss.

[0178] (4.3) Backward propagation: Use the calculated total loss to automatically calculate the gradients of each parameter in the network through the chain rule.

[0179] (4.4) Parameter update: Use the Adam optimizer to update the network parameters according to the gradients to minimize the total loss.

[0180] In this way, during the actual training process, the model parameters of the preset YOLO model can be optimized through continuous iteration.

[0181] (5) Model validation:

[0182] (5.1) Setting of cross-validation method

[0183] Cross-validation is a statistical analysis method for evaluating the ability of a model to generalize to an independent dataset, which can effectively avoid model overfitting.

[0184] (5.1.1) Divide the dataset

[0185] Dataset division: Randomly divide the labeled image sample set into K mutually exclusive subsets (usually choose 5 or 10). Each subset should maintain the consistency of data distribution as much as possible, that is, the data type and proportion in each subset should be similar to those of the entire dataset.

[0186] (5.1.2) Perform K times of training and validation

[0187] K - fold iteration: In each iteration, a subset is selected as the validation set, and the remaining K - 1 subsets are combined as the training set. In this way, each subset will be strictly used as the validation set once and as the training set K - 1 times;

[0188] Training and validation: In each iteration, the model is trained using the data from K - 1 subsets, and then the performance of the model is tested on the reserved validation set.

[0189] (5.2) Evaluate the generalization ability and accuracy of the model

[0190] (5.2.1) Determine the model performance metrics:

[0191] Accuracy: Measures the proportion of correctly predicted labels by the improved YOLO model;

[0192] Precision: Measures the proportion of samples that are truly positive among the samples predicted as positive by the improved YOLO model;

[0193] Recall: Measures the proportion of positive samples correctly identified by the improved YOLO model among all positive samples;

[0194] F1 - score: The harmonic mean of precision and recall, which is an important indicator for evaluating the accuracy of the improved YOLO model.

[0195] (5.2.2) Performance analysis

[0196] Recording and analysis: In each validation, record the above - mentioned performance metrics and analyze the performance differences of the improved YOLO model on different validation sets to evaluate the generalization ability of the improved YOLO model;

[0197] Average performance: Calculate the average value of the performance metrics of the improved YOLO model in all iterations to provide a comprehensive performance evaluation.

[0198] (5.3) Adjust the model parameters

[0199] Parameter adjustment: According to the results of cross - validation, especially the average value and standard deviation of those performance metrics, adjust the key parameters (such as learning rate, batch size, etc.) of the improved YOLO model, or modify the network architecture (such as increasing or decreasing the number of layers, adjusting the filter size);

[0200] Iterative optimization: According to the adjusted parameters, conduct another round of cross - validation to verify the effect of parameter adjustment;

[0201] In this application, by performing the above model training and verification steps, it not only helps to optimize the performance of the improved YOLO model, but also lays a foundation for the actual deployment and use of the improved YOLO model.

[0202] To further illustrate, the following uses experimental data to verify that the improved YOLO model can have good performance.

[0203] (1) Use the on-site images of the deep foundation pit slope of a certain project (such as Figure 2 and Figure 3 the on-site images of the deep foundation pit slope shown) as the input images of the improved YOLO model. Then use the above online training method to train the improved YOLO model. During the training, the performance diagrams during training as shown in Figure 4 and Figure 5 can be obtained. According to the performance diagrams during training as shown in Figure 4 and Figure 5 , the data preprocessing effect, the accuracy of the labels and the model performance can be checked.

[0204] (2) After completing the training, the data analysis diagram as shown in Figure 6 can be obtained. The following explains the training and verification accuracy metrics in Figure 6 .

[0205] Precision refers to the proportion of samples that are actually positive among the samples predicted as positive by the improved YOLO model. Its calculation formula is: Precision = TP / (TP + FP), where TP is the true positive and FP is the false positive. Among them, the performance of Precision in Figure 6 is: it fluctuates slightly in the initial stage of training, but generally remains at a relatively high level, indicating that the improved YOLO model has a strong ability to predict positive classes.

[0206] Recall refers to the proportion of samples that are actually positive and are correctly predicted as positive by the improved YOLO model. Its calculation formula is: Recall = TP / (TP + FN), where FN is the false negative. Among them, the performance of Recall in Figure 6 is: similar to Precision, it also has some fluctuations in the initial stage of training, but generally remains at a high level, indicating that the improved YOLO model has a high recognition rate for positive samples.

[0207] mAP50 refers to the average precision of all classes when the IOU threshold is 0.5. The higher the mAP value, the better the overall performance of the improved YOLO model. Among them, the performance of mAP50 in Figure 6 is: it remains in a state close to full marks from the initial stage of training, indicating that the overall detection effect of the improved YOLO model is very good.

[0208] mAP50-95 refers to the average precision of all classes at different IOU thresholds (such as from 0.5 to 0.95 with a step of 0.05). This metric is more stringent and comprehensively measures the detection ability of the improved YOLO model. Among them, mAP50-95 also remains at a relatively high level during Figure 6 the training process, indicating that the improved YOLO model has good detection effects at different IOU thresholds.

[0209] From Figure 6 the data performance, as the number of training rounds of the improved YOLO model increases, the precision, recall rate, mAP50, and mAP50-95 all remain at relatively high levels, indicating that the improved YOLO model performs well on both training and validation data. For example, regarding the fluctuation situation, although the precision and recall rate fluctuate somewhat in the initial stage of training, this is a normal phenomenon, usually because the improved YOLO model is still learning and adjusting parameters in the initial stage. Regarding the stability, generally speaking, the various indicators of the improved YOLO model tend to be stable during the training process, indicating that the improved YOLO model is well-trained and has good performance.

[0210] That is Figure 6 intuitively shows the change of the accuracy index of the improved YOLO model under different training rounds, providing strong support for the performance evaluation and optimization of the improved YOLO model.

[0211] (3) Finally, combine the on-site images of the deep foundation pit slope to complete the crack image recognition test, and the recognition result example diagram as shown in Figures 7 - 9 can be obtained. Among them, crack in Figures 7 - 9 refers to the crack, and the number behind it represents the corresponding recognition probability.

[0212] From the above description, the above experimental data well verify that the improved YOLO model can have good performance.

[0213] S140. Fusion-compute the visual information and environmental sensing data to obtain the final displacement data of the target deep foundation pit slope.

[0214] In one implementation, the displacement recognition data in the visual information and the displacement sensing data in the environmental sensing data can be substituted into a specified formula for fusion calculation to obtain the final displacement data.

[0215] As an example, the specified formula can be shown as the following formula (8).

[0216] d = αdv + (1 - α)ds (8)

[0217] Among them, dv is the displacement amount corresponding to the displacement recognition data, ds is the displacement amount corresponding to the displacement sensing data, and α is a weight coefficient between 0 and 1, which is used to adjust the influence of the displacement recognition data and the displacement sensing data in the final output. In specific applications, the optimal value of α can be determined according to experimental or historical data analysis to ensure the reasonable fusion of these two data sources, namely the displacement recognition data and the displacement sensing data.

[0218] In the embodiment of the present application, by executing step S140, the visual information recognized by the trained improved YOLO model can be fused with the environmental sensing data, enhancing the accuracy and robustness of detecting the displacement of the soil mass on the slope of the deep foundation pit, providing more comprehensive monitoring information, and achieving the improvement of the accuracy and reliability of the monitoring data of the slope of the deep foundation pit.

[0219] In summary, in the method for monitoring the displacement of the soil mass on the slope of the deep foundation pit of the present application, by executing steps S110 - S140, the following beneficial effects can be achieved:

[0220] First, improve the level of monitoring automation: By using a trained deep learning model (such as the trained improved YOLO model) to automatically analyze the real-time high-definition images of the real-time target deep foundation pit slope, reduce manual intervention, and improve the automation level of the monitoring process.

[0221] Second, expand the monitoring range and accuracy: Obtain real-time high-definition images covering the target deep foundation pit slope. For example, a high-resolution camera can be used to cover a wider area, and combined with the high-precision recognition function of the trained improved YOLO model, fine-grained monitoring of the entire deep foundation pit area can be realized.

[0222] Third, achieve real-time data processing and early warning: The trained improved YOLO model can immediately process image data, timely identify safety risks such as soil cracks and soil displacement on the slope of the deep foundation pit, and quickly give an early warning, greatly improving the response speed and timeliness.

[0223] Fourth, enhance safety: Real-time and comprehensive monitoring can timely detect potential risks, so as to take corresponding safety measures, reduce the probability of accidents, and protect the safety of workers and equipment.

[0224] Fifth, reduce costs: Automated monitoring reduces the dependence on manpower. At the same time, rapid identification and response to potential problems reduce the costs of long-term maintenance and emergency handling.

[0225] Sixth, enhance the accuracy and reliability of monitoring data: By combining multi-source data through data fusion technology, the comprehensiveness and accuracy of monitoring data are improved, making engineering management decisions more scientific and effective.

[0226] Figure 10 The structural block diagram of the deep foundation pit slope soil displacement monitoring device according to an embodiment of the present application is shown. As Figure 10 shown, the device may include:

[0227] A processing unit 210, configured to obtain a real-time high-definition image covering the target deep foundation pit slope and environmental sensing data of the target deep foundation pit slope; preprocess the real-time high-definition image to obtain a processed real-time high-definition image;

[0228] A monitoring unit 220, configured to call the trained improved YOLO model to perform recognition and analysis on the processed real-time high-definition image to obtain visual information of the target deep foundation pit slope, where the visual information includes displacement recognition data; fuse and calculate the visual information and the environmental sensing data to obtain the final displacement data of the target deep foundation pit slope.

[0229] In an implementation manner, when the processing unit 210 is configured to obtain a real-time high-definition image covering the target deep foundation pit slope and environmental sensing data of the target deep foundation pit slope, it is specifically configured to:

[0230] Regularly photograph the target deep foundation pit slope using a high-resolution camera to obtain a real-time high-definition image, and the shooting range of the high-resolution camera covers the target deep foundation pit slope;

[0231] Collect relevant data of the target deep foundation pit slope using an environmental sensor to obtain environmental sensing data.

[0232] In an implementation manner, when the processing unit 210 is configured to preprocess the real-time high-definition image to obtain a processed real-time high-definition image, it is specifically configured to:

[0233] Perform denoising, brightness and contrast adjustment, cropping and scaling processing on the real-time high-definition image to obtain a preliminarily processed real-time high-definition image;

[0234] Perform normalization processing on the preliminarily processed real-time high-definition image to obtain a processed real-time high-definition image.

[0235] In an implementation manner, the trained improved YOLO model is obtained by the monitoring unit 220 through the following training process:

[0236] Obtain an image sample set, where the image sample set includes multiple deep foundation pit slope images;

[0237] Preprocess each deep foundation pit slope image in the image sample set to obtain a processed deep foundation pit slope image for each;

[0238] Mark each processed deep foundation pit slope image, mark the key features in each deep foundation pit slope image, and obtain a marked image sample set. The key features include soil cracks and soil displacements.

[0239] Train the preset YOLO model based on the marked image sample set. During the training process, use GPU to accelerate the training process, and continuously adjust the network structure and loss function of the preset YOLO model to obtain a trained and improved YOLO model.

[0240] In one implementation, the adjustment of the network structure of the preset YOLO model includes: adjusting the aspect ratio and size of the anchor boxes of the preset YOLO model, the number of convolutional layers of the preset YOLO model, the filter size of the preset YOLO model, and the image input size of the preset YOLO model.

[0241] In one implementation, the adjustment of the loss function of the preset YOLO model includes: adjusting the location loss of the loss function, the size loss of the loss function, the confidence loss of the loss function, and the class probability loss of the loss function.

[0242] In one implementation, when the monitoring unit 220 is used to fuse and calculate visual information and environmental sensing data to obtain the final displacement data of the target deep foundation pit slope, it is specifically used for:

[0243] Substitute the displacement recognition data in the visual information and the displacement sensing data in the environmental sensing data into a specified formula for fusion calculation to obtain the final displacement data.

[0244] The functions of the units in the deep foundation pit slope soil displacement monitoring device according to the embodiments of the present application can be referred to the corresponding descriptions in the above method, and will not be elaborated here.

[0245] Figure 11 Show the structural block diagram of a computer device according to an embodiment of the present application. As Figure 11 shown, the computer device includes: a memory 310 and a processor 320. Instructions are stored in the memory 310, and the instructions are loaded and executed by the processor 320 to implement the deep foundation pit slope soil displacement monitoring method in the above embodiments. The number of the memory 310 and the processor 320 can be one or more.

[0246] The computer device further includes:

[0247] A communication interface 330, used to communicate with external devices for data interaction and transmission.

[0248] If the memory 310, the processor 320, and the communication interface 330 are implemented independently, the memory 310, the processor 320, and the communication interface 330 can be interconnected through a bus and communicate with each other. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, an Extended Industry Standard Architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 11 it is represented by only a thick line in Figure 11 , but it does not mean that there is only one bus or one type of bus.

[0249] Optionally, in a specific implementation, if the memory 310, the processor 320, and the communication interface 330 are integrated on a single chip, the memory 310, the processor 320, and the communication interface 330 can communicate with each other through an internal interface.

[0250] The embodiment of the present application provides a computer-readable storage medium, in which a computer program is stored. When the computer program runs on a computer, the method provided in the embodiment of the present application is implemented.

[0251] The embodiment of the present application also provides a chip, which includes a processor for calling and running instructions stored in a memory, so that a communication device installed with the chip executes the method provided in the embodiment of the present application.

[0252] The embodiment of the present application also provides a chip, including: an input interface, an output interface, a processor, and a memory. The input interface, the output interface, the processor, and the memory are connected through an internal connection path. The processor is used to execute the code in the memory. When the code is executed, the processor is used to execute the method provided in the embodiment of the application.

[0253] It should be understood that the above-mentioned processor can be a Central Processing Unit (CPU), or 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. The general-purpose processor can be a microprocessor or any conventional processor, etc. It is worth noting that the processor can be a processor that supports the Advanced RISC Machines (ARM) architecture.

[0254] Further, optionally, the above-mentioned memory can include a read-only memory and a random access memory, and can also include a non-volatile random access memory. The memory can be a volatile memory or a non-volatile memory, or can include both volatile and non-volatile memories. Among them, the non-volatile memory can include a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory can include a random access memory (RAM), which is used as an external cache. By way of example but not limitation, many forms of RAM are available. For example, static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct rambus random access memory (DR RAM).

[0255] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions according to the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium.

[0256] In the description of this specification, the descriptions with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples", etc. mean 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 application. Moreover, the specific features, structures, materials, or characteristics described can be combined in a suitable manner in any one or more embodiments or examples. In addition, without contradiction, those skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0257] In the description of the present application, the meaning of "multiple" is two or more, unless otherwise specifically defined.

[0258] Any process or method description represented in the flowchart or described in other ways herein can be understood to represent a module, segment, or part of code including one or more executable instructions for implementing a specific logical function or process. And the scope of the preferred embodiments of the present application includes additional implementations, where the functions can be executed in a substantially simultaneous manner or in the reverse order according to the involved functions, rather than in the order shown or discussed.

[0259] The logic and / or steps represented in the flowchart or described in other ways herein, for example, can be considered as a sequenced list of executable instructions for implementing a logical function, and can be specifically implemented in any computer-readable medium for use by an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other systems that can fetch and execute instructions from the instruction execution system, apparatus, or device), or in combination with these instruction execution systems, apparatus, or devices.

[0260] It should be understood that various parts of the present application can be implemented by hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. All or part of the steps of the above-described method embodiments can be completed by a program instructing relevant hardware. The program can be stored in a computer-readable storage medium. When the program is executed, it includes one or a combination of the steps of the method embodiments.

[0261] In addition, in each embodiment of the present application, the functional units can be integrated into one processing module, or each unit can exist physically alone, or two or more units can be integrated into one module. The above integrated module can be implemented in the form of hardware or in the form of a software functional module. If the above integrated module is implemented in the form of a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. The storage medium can be a read-only memory, a magnetic disk, an optical disk, or the like.

[0262] The above is only the specific embodiment of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present application can easily think of various changes or substitutions, and these should all be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.< / height> < / width>

Claims

1. A method for monitoring soil displacement of a deep foundation pit slope, characterized in that: include: Acquire a real-time high-definition image covering a target deep foundation pit slope and environmental sensing data of the target deep foundation pit slope, wherein the environmental sensing data includes displacement sensing data; Preprocessing the real-time high-definition image to obtain a processed real-time high-definition image; Calling the trained improved YOLO model to perform recognition analysis on the processed real-time high-definition image to obtain visual information of the target deep foundation pit slope, wherein the visual information includes displacement recognition data; Substituting the displacement recognition data in the visual information and the displacement sensing data in the environmental sensing data into a specified formula for fusion calculation to obtain the final displacement data of the target deep foundation pit slope; Wherein, the specified formula is: d=αdv+(1-α)ds Among them, dv is the displacement corresponding to the displacement recognition data, ds is the displacement corresponding to the displacement sensing data, and α is a weight coefficient between 0 and 1, which is used to adjust the influence of the displacement recognition data and the displacement sensing data in the final output; The trained improved YOLO model is obtained through the following training process: Acquire an image sample set, wherein the image sample set includes a plurality of deep foundation pit slope images; Preprocessing each deep foundation pit slope image in the image sample set to obtain each processed deep foundation pit slope image; Marking each processed deep foundation pit slope image, marking key features in each deep foundation pit slope image, and obtaining a marked image sample set, wherein the key features include soil cracks and soil displacement; Training a preset YOLO model according to the labeled image sample set, and during the training process, using a GPU to accelerate the training process, continuously adjusting the network structure and loss function of the preset YOLO model, to obtain the trained improved YOLO model; Among them, the adjustment of the loss function of the preset YOLO model includes: adjusting the position loss of the loss function, the size loss of the loss function, the confidence loss of the loss function and the category probability loss of the loss function.

2. The method according to claim 1, characterized in that: Acquiring real-time high-definition images covering a target deep foundation pit slope and environmental sensing data of the target deep foundation pit slope includes: Using a high-resolution camera to regularly photograph the target deep foundation pit slope to obtain the real-time high-definition image, the shooting range of the high-resolution camera covering the target deep foundation pit slope; The environmental sensor is used to collect the relevant data of the target deep foundation pit slope to obtain the environmental sensing data.

3. The method according to claim 1, characterized in that: Preprocessing the real-time high-definition image to obtain the processed real-time high-definition image includes: De-noising, adjusting brightness and contrast, cropping and scaling the real-time high-definition image to obtain a real-time high-definition image after preliminary processing; The real-time high-definition image after the preliminary processing is normalized to obtain the processed real-time high-definition image.

4. The method according to claim 1, characterized in that The adjustment of the network structure of the preset YOLO model includes: adjusting the anchor box aspect ratio and anchor box size of the preset YOLO model, the number of convolutional layers of the preset YOLO model, the filter size of the preset YOLO model, and the image input size of the preset YOLO model.

5. A deep foundation pit slope soil displacement monitoring device, characterized in that: include: A processing unit, used to obtain a real-time high-definition image covering a target deep foundation pit slope and environmental sensing data of the target deep foundation pit slope; Preprocessing the real-time high-definition image to obtain a processed real-time high-definition image; The monitoring unit is used to call the trained improved YOLO model to perform recognition analysis on the processed real-time high-definition image to obtain visual information of the target deep foundation pit slope, wherein the visual information includes displacement recognition data; the displacement recognition data in the visual information and the displacement sensing data in the environmental sensing data are substituted into a specified formula for fusion calculation to obtain the final displacement data of the target deep foundation pit slope; Wherein, the specified formula is: d=αdv+(1-α)ds Among them, dv is the displacement corresponding to the displacement recognition data, ds is the displacement corresponding to the displacement sensing data, and α is a weight coefficient between 0 and 1, which is used to adjust the influence of the displacement recognition data and the displacement sensing data in the final output; The trained improved YOLO model is obtained by the monitoring unit through the following training process: obtaining an image sample set, the image sample set including a plurality of deep foundation pit slope images; preprocessing each deep foundation pit slope image in the image sample set to obtain each processed deep foundation pit slope image; marking each processed deep foundation pit slope image, marking key features in each deep foundation pit slope image, and obtaining a marked image sample set, wherein the key features include soil cracks and soil displacement; training a preset YOLO model based on the marked image sample set, and during the training process, using a GPU to accelerate the training process, and continuously adjusting the network structure and loss function of the preset YOLO model to obtain the trained improved YOLO model; Among them, the adjustment of the loss function of the preset YOLO model includes: adjusting the position loss of the loss function, the size loss of the loss function, the confidence loss of the loss function and the category probability loss of the loss function.

6. A computer device, characterized in that: include: A memory and a processor, wherein the memory stores instructions, and the instructions are loaded and executed by the processor to implement the method according to any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed on a computer, the method according to any one of claims 1 to 4 is implemented.

Images