Radar monitoring and early warning method and system based on mine field slope instability trend

Abstract

The invention relates to the technical field of slope monitoring, in particular to a mine field slope instability trend-based radar monitoring and early warning method and system, and the method comprises the following steps: recognizing a slope region with an instability risk on a geological image through a pre-trained Fast R CNN model, and taking the slope region as an instability risk region; scanning the instability risk area in real time through a slope radar to obtain a radar image sequence of the instability risk area; and acquiring a geological image sequence of the instability risk area with the same time sequence as the radar image sequence, determining an instability trend of the instability risk area based on the radar image sequence and the geological image sequence, and then allocating an early warning level according to the instability trend. According to the method, locking of instability factors in advance is realized on the basis of the Fast R CNN model, falling rock and soil loosening sliding track trends are quickly analyzed in real time on the actual falling rock and soil loosening track obtained through multi-modal data solution on the basis of the LSTM network, and thus integrated display, focusing monitoring and graded early warning of mining area risk points are realized.

Description

Technical Field

[0001] The present invention relates to the technical field of slope monitoring, and particularly to a radar monitoring and early warning method and system based on the instability trend of mine slopes. Background Art

[0002] In the monitoring of mine slopes, radar technology is widely used. By monitoring the displacement, deformation, cracks, etc. of slopes, potential slope landslide hazards can be detected in a timely manner. The radar early warning mechanism process for mine slopes can take different countermeasures according to different early warning levels, effectively reducing the probability of accidents.

[0003] With the rise and development of computer vision algorithms based on deep learning and the improvement of computer computing power, slope disaster monitoring technology has entered a new research stage. Convolutional Neural Networks (CNN) is one of the classic deep learning algorithms and can well extract feature information such as texture, color, and shape on two-dimensional images. Therefore, in slope disaster recognition, algorithms such as object recognition and semantic segmentation based on CNN are mainly used to process slope disaster image data, which mainly comes from high-resolution satellite images, UAV images, etc. Xu et al. combined the cloud platform Google Earth Engine (GEE) with the U-Net semantic segmentation network to segment and identify Landsat image data of post-earthquake landslides, achieving earthquake landslide interpretation and playing a key role in post-disaster reconstruction work.

[0004] Although the existing technology can achieve certain results in slope disaster recognition by using deep learning technology, it is unable to lock in the instability factors in advance (i.e., the areas of falling rocks and loose soil prone to slope disaster states), and cannot quickly and real-time analyze the instability trend (i.e., the sliding trajectories of falling rocks and loose soil) after the instability factors become unstable, corresponding to the dynamic and static combined slope monitoring process of hierarchical early warning. Therefore, the current slope monitoring and early warning adopts undifferentiated or non-key area monitoring. The monitoring range is wide and the data processing volume is large. The operation time for deep learning to detect the disaster occurrence area is long or even difficult to detect the disaster occurrence area, that is, the disaster recognition accuracy and efficiency of deep learning technology are insufficient. Moreover, the current disaster recognition is the landslide interpretation based on the established facts of disasters, and it cannot predict the possible impacts of disasters a short time after the disasters occur and conduct hierarchical early warning according to the disaster impacts. Summary of the Invention

[0005] The object of the present invention is to provide a radar monitoring and early warning method and system based on the instability trend of mine slopes, so as to solve the technical problems in the prior art that the monitoring range of slope disasters is wide and the data processing volume is large, resulting in insufficient accuracy and efficiency of disaster identification, and it is difficult to predict the possible impacts of disasters for hierarchical early warning, and the flexibility of early warning is limited.

[0006] To solve the above technical problems, the present invention specifically provides the following technical solutions: A radar monitoring and early warning method based on the instability trend of mine slopes, comprising the following steps: Collect geological images of the open-pit mine slope, and identify the slope areas with instability risks on the geological images through a pre-trained Fast R CNN model as instability risk areas; Set the instability risk areas as the monitoring areas of the slope radar, and obtain a radar image sequence of the instability risk areas by scanning the instability risk areas in real time with the slope radar; Obtain a geological image sequence of the instability risk areas with the same time sequence as the radar image sequence, determine the instability trend of the instability risk areas based on the radar image sequence and the geological image sequence, and then allocate early warning levels according to the instability trend; The method for determining the instability trend includes: Construct an image conversion model for converting geological images into radar images through a CycleGAN network structure; Convert the geological images in the geological image sequence of the instability risk areas into radar images through the image conversion model to generate a radar conversion image sequence of the instability risk areas; Fuse the radar conversion image sequence of the instability risk areas with the radar image sequence of the instability risk areas to generate a radar multi-modal image sequence of the instability risk areas; Solve the local instability trajectories of falling rocks and loose soil representing the instability trend in the radar multi-modal image sequence of the instability risk areas, and predict the global instability trajectories of falling rocks and loose soil based on the local instability trajectories through an LSTM network structure; Conduct hierarchical early warning on the instability risk areas on the open-pit mine according to the early warning levels.

[0007] As a preferred solution of the present invention, the pre-training method of the Fast R CNN model includes: Obtain multiple geological images, and mark the areas containing falling rocks and loose soil in each geological image as instability risk areas; Divide the data set composed of multiple geological images into a test set and a training set; On the training set, use the geological images as input items and the instability risk areas of the geological images as output items to train the Fast R CNN model; On the test set, the performance of the Fast R CNN model in identifying the instability risk area is evaluated.

[0008] As a preferred embodiment of the present invention, the method for constructing the image conversion model includes: Obtain multiple geological images of the instability risk area, and obtain multiple radar images of the instability risk area at the same time as the geological images; Divide the data set composed of multiple geological images of the instability risk area and multiple radar images of the instability risk area into a training set and a test set; On the training set, use the first GAN network structure in the CycleGAN network structure to convert the geological image of the instability risk area into the radar image of the instability risk area, and use the second GAN network structure in the CycleGAN network structure to convert the radar image of the instability risk area into the geological image of the instability risk area, and train the CycleGAN network structure to obtain the image conversion model; On the test set, evaluate the performance of the image conversion model in image modality conversion; Among them, The loss function for training the CycleGAN network structure is supplemented with the coordinate consistency loss L of falling rocks and loose soil on the basis of the cycle consistency loss Lcycle and the adversarial generation loss L GAN ; P ; ; In the formula, is the radar image of the instability risk area converted from the geological image of the instability risk area, is the geological image of the instability risk area converted from the radar image of the instability risk area, x is the geological image of the instability risk area, y is the radar image of the instability risk area, is the position coordinates of falling rocks and loose soil identified by the Fast R CNN model on ; is the position coordinates of falling rocks and loose soil identified by the Fast R CNN model on ; is the position coordinates of falling rocks and loose soil identified by the Fast R CNN model on x, is the position coordinates of falling rocks and loose soil identified by the Fast R CNN model on y, is the L1 norm formula, and are both mathematical expectations.

[0009] As a preferred embodiment of the present invention, the method for generating the radar multi-modal image sequence of the instability risk area includes: Through the cross-attention mechanism, the radar image sequence is mapped into a query matrix, the radar transformed image sequence is mapped into key and value matrices, and the attention weights of the radar image sequence are calculated; Through the cross-attention mechanism, the radar transformed image sequence is mapped into a query matrix, the radar image sequence is mapped into key and value matrices, and the attention weights of the radar transformed image sequence are calculated; The radar image sequence weighted by the attention weights of the radar image sequence is added to the radar transformed image sequence weighted by the attention weights of the radar transformed image sequence to obtain the radar multimodal image sequence.

[0010] As a preferred solution of the present invention, the calculation and solution method of the local instability trajectory includes: Performing phase difference processing on two adjacent radar multimodal images in the radar multimodal image sequence in turn to generate an interferogram sequence , where is the interferogram at the t-th time sequence, is the phase information of the radar multimodal image at the (t + 1)-th time sequence, is the phase information of the radar multimodal image at the t-th time sequence, and m is the total number of time sequences of the radar multimodal image sequence; Based on the phase difference information in the interferogram sequence, calculate the movement trajectories of falling rocks and loose soil in the instability risk area as the local instability trajectory , where is the movement displacement of falling rocks and loose soil at the t-th time sequence, is the radar wavelength.

[0011] As a preferred solution of the present invention, the prediction method of the global instability trajectory includes: Using the LSTM network structure to perform subsequent prediction on the local instability trajectory to obtain the global instability trajectory , where is the movement displacement of falling rocks and loose soil at the t-th time sequence, m is the total number of time sequences of the radar multimodal image sequence, and n is the total number of subsequent prediction time sequences.

[0012] As a preferred solution of the present invention, the distribution method of the warning level includes: Calculate The distance from the mine protection area. When The distance from the mine protection area is less than the preset safety distance, a high-risk level is assigned to the instability risk area; When If the distance from the mine protection area is greater than or equal to the preset safety distance, a medium - low risk level is assigned to the instability risk area.

[0013] The present invention provides a radar monitoring and early warning system based on the instability trend of a mine slope. According to the above - mentioned radar monitoring and early warning method based on the instability trend of a mine slope, the system includes: A vision device for collecting geological images; A slope radar for collecting radar images; A data processor for identifying, through a pre - trained Fast R CNN model, the slope areas with instability risks in the geological images as instability risk areas; determining the instability trend of the instability risk areas based on the radar image sequence and the geological image sequence, and then assigning an early warning level according to the instability trend; A display device loaded with software for monitoring and grading early warning of an open - pit mine slope, used to display the radar images of the open - pit mine, and perform grading marking, instability trend display, and early warning pop - up windows on the early warning areas in the radar images.

[0014] As a preferred solution of the present invention, the vision device includes a remote sensing image acquisition device and a drone camera device.

[0015] The present invention has the following beneficial effects compared with the prior art: Based on the Fast R CNN model, the present invention realizes early locking of instability factors (i.e., the areas of falling rocks and loose soil in a state prone to slope disasters), and based on the LSTM network, quickly and real - time analyzes the sliding trajectory trend of falling rocks and loose soil on the actual trajectory of falling rocks and loose soil obtained from multi - modal data, so as to perform corresponding grading early warning, realizing the integrated display, focused monitoring, and grading early warning of risk points in the mining area, and improving the effect of slope monitoring. Description of the Drawings

[0016] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings described below are only exemplary, and for those of ordinary skill in the art, without creative efforts, other implementation drawings can be obtained according to the provided drawings.

[0017] Figure 1 It is a flowchart of the radar monitoring and early warning method based on the instability trend of a mine slope provided by an embodiment of the present invention; Figure 2 It is a structural block diagram of the radar monitoring and early warning system based on the instability trend of a mine slope provided by an embodiment of the present invention. Detailed Embodiments

[0018] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying 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 the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0019] As Figure 1 shown, the present invention provides a radar monitoring and early warning method based on the instability trend of a mine slope, including the following steps: Collect geological images of the open-pit mine slope, and identify the slope areas with instability risks on the geological images through a pre-trained Fast R-CNN model as instability risk areas; Set the instability risk area as the monitoring area of the slope radar, and obtain a radar image sequence of the instability risk area by scanning the instability risk area in real time with the slope radar; Obtain a geological image sequence of the instability risk area with the same time sequence as the radar image sequence, determine the instability trend of the instability risk area based on the radar image sequence and the geological image sequence, and then allocate an early warning level according to the instability trend; Perform hierarchical early warning on the instability risk areas on the open-pit mine according to the early warning level.

[0020] In the silent period of the open-pit mine slope, the present invention collects its geological images (such as satellite remote sensing images, drone-shot images, etc.). On the geological images, a region containing disaster elements such as falling rocks and loose soil is divided through a Fast R-CNN model with image segmentation function. In this region, the falling rocks and loose soil elements are more likely to cause instability and disasters, such as rolling down to the mining protection area and injuring people or damaging house construction equipment. That is to say, this region has instability risks. Therefore, the present invention identifies such regions with instability risks on the slope during the silent period, locks in the instability risk areas in advance, and then performs radar scanning and real-time monitoring on them to achieve zonal monitoring of slope early warning.

[0021] Through the real-time scanning of the slope radar, the present invention can obtain the radar image data during the instability outbreak period (falling rocks and loose soil sliding) in the instability risk area in real time. At the same time, obtain the geological image data synchronized with the radar image data during the outbreak period, fuse these two modalities of image data to solve the actual trajectories of the falling rocks and loose soil sliding during the instability outbreak period, and predict the subsequent trajectories of the falling rocks and loose soil sliding based on the actual trajectories of the falling rocks and loose soil sliding during the instability outbreak period through an LSTM network with time series prediction function. The instability trend can be pre-grasped a short time after the outbreak period occurs, that is, the instability trend can be quickly and real-time analyzed after the instability factors become unstable, that is, the sliding trajectories of the falling rocks and loose soil. Whether disasters are caused is judged according to the instability trend, and hierarchical early warning is performed based on the degree of disaster impact.

[0022] The present invention solves the actual trajectories of falling rocks and loose soil sliding during the instability outbreak period by fusing radar image data and geological image data of two modalities, can obtain the accuracy advantage of solving by multi-modal data fusion, improve the accuracy of solving the actual trajectories, so as to accurately master the actual trajectories of falling rocks and loose soil sliding during the outbreak period. Furthermore, the accurate improvement of the actual trajectories can ensure the accuracy improvement of the predicted trajectories of falling rocks and loose soil sliding following the actual trajectories, thus ensuring the accuracy of slope disaster warning.

[0023] The image conversion network constructed by the CycleGAN network structure in the present invention can convert geological images into the form of radar images, thereby converting the geological image data synchronized with the radar image data during the outbreak period into the form of radar images (i.e., radar conversion images), and fusing the radar conversion images and radar images, realizing the fusion of multi-modal data. Using the fused radar multi-modal images to calculate the actual trajectories of falling rocks and loose soil sliding, compared with using only radar images for calculation, the information features provided by multi-modal data are richer, reducing the random error of single-modal data, and the positioning of falling rocks and loose soil is more accurate, thus making the calculation of the actual trajectories of falling rocks and loose soil more accurate.

[0024] The present invention combines the attention mechanism in multi-modal data fusion. Utilizing the key information focusing characteristic of the attention mechanism, it can assign high weights to important information in the radar conversion images and radar images, highlighting the key performances in the two modalities of images, enabling information interaction between different modalities of the two images, improving the cross-modal understanding ability of the model, helping the model better analyze the role features, and improving the quality of image multi-modal fusion. Therefore, through the multi-modal fusion of the attention mechanism, it can better improve the radar conversion images and radar images, retain the key information of their respective modalities, and at the same time interact the position information of falling rocks and loose soil sliding, obtaining accurate position information of falling rocks and loose soil sliding.

[0025] In the present invention, the phase interference method is adopted for calculating the actual trajectories of falling rocks and loose soil sliding. Through a sequence of radar multi-modal images during a short period of time in the outbreak period, the actual trajectories of falling rocks and loose soil sliding are calculated.

[0026] After calculating the actual trajectories of falling rocks and loose soil sliding, the present invention predicts the subsequent trajectories of falling rocks and loose soil sliding based on the actual trajectories of falling rocks and loose soil sliding during the instability outbreak period through the LSTM network with time series prediction function. The instability trend can be pre-grasped a short time after the outbreak period occurs, that is, the instability trend can be quickly and real-time analyzed after the instability factors become unstable.

[0027] After the present invention predicts the trajectories of subsequent falling rocks and loose soil slides, it determines whether their trajectories invade the mine protection area, that is, whether they will cause substantial harm. If so, it will be an emergency event and a high-level warning will be issued. If not, a medium or low-level warning will be issued. The graded warning of the present invention is in line with the actual situation, and the accuracy of the graded warning is improved.

[0028] During the silent period of the open-pit mine slope, the present invention collects its geological images (such as satellite remote sensing images, images taken by drones, etc.), and divides the areas containing disaster elements such as falling rocks and loose soil on its geological images through the Fast R CNN model with image segmentation function, specifically as follows: The pre-training method of the Fast R CNN (Fast Region-based Convolutional Neural Network, an improved object detection model designed to improve the speed and efficiency of object detection while maintaining or improving detection accuracy) model includes: Obtain multiple geological images, and mark the areas containing falling rocks and loose soil in each geological image as unstable risk areas; Divide the data set composed of multiple geological images into a test set and a training set; On the training set, use the geological image as the input item and the unstable risk area of the geological image as the output item to train the Fast R CNN model; On the test set, conduct a performance evaluation of the Fast R CNN model for identifying unstable risk areas.

[0029] The image conversion network constructed by the CycleGAN network structure of the present invention can convert the geological image into the form of a radar image, so as to convert the geological image data synchronized with the radar image data during the outbreak period into the form of a radar image (that is, a radar conversion image), fuse the radar conversion image and the radar image, realize the fusion of multi-modal data, and use the fused radar multi-modal image to calculate the actual trajectories of falling rocks and loose soil slides. Compared with using only radar images for calculation, the information features provided by multi-modal data are richer, reducing the random error of single-modal data, and the positioning of falling rocks and loose soil is more accurate, thus making the calculation of the actual trajectories of falling rocks and loose soil more accurate, specifically as follows: The method for determining the instability trend includes: Construct an image conversion model for converting geological images into radar images through the CycleGAN network structure (Cycle-Consistent Generative Adversarial Networks, a deep learning model for unsupervised image-to-image conversion, whose main goal is to achieve image conversion between different domains without paired training data. CycleGAN uses two generators and two discriminators, and utilizes cycle-consistency loss and adversarial loss to ensure the consistency of the main content and structure of the image during the conversion process); Convert the geological images in the geological image sequence of the instability risk area into radar images through the image conversion model, and generate a radar conversion image sequence of the instability risk area; Fuse the radar conversion image sequence of the instability risk area with the radar image sequence of the instability risk area to generate a radar multi-modal image sequence of the instability risk area; Solve the local instability trajectories of falling rocks and loose soil representing the instability trend in the radar multi-modal image sequence of the instability risk area, and predict the global instability trajectories of falling rocks and loose soil based on the local instability trajectories through the LSTM network structure.

[0030] The construction method of the image conversion model includes: Obtain multiple geological images of the instability risk area, and obtain multiple radar images of the instability risk area at the same time as the geological images; Divide the data set composed of multiple geological images of the instability risk area and multiple radar images of the instability risk area into a training set and a test set; On the training set, use the first-layer GAN network structure in the CycleGAN network structure to convert the geological images of the instability risk area into radar images of the instability risk area, and use the second-layer GAN network structure in the CycleGAN network structure to convert the radar images of the instability risk area into geological images of the instability risk area, and train the CycleGAN network structure to obtain an image conversion model; On the test set, perform performance evaluation of image modality conversion on the image conversion model; Among them, the loss function for training the CycleGAN network structure supplements and adds the coordinate consistency loss L of falling rocks and loose soil on the basis of the cycle-consistency loss Lcycle and the adversarial generation loss L GAN ; P ; ; In the formula, is the radar image of the instability risk area converted from the geological image of the instability risk area, The geological image of the instability risk area converted from the radar image of the instability risk area, where x is the geological image of the instability risk area and y is the radar image of the instability risk area. is obtained by the Fast R CNN model to identify the coordinates of falling rocks and loose soil. is obtained by the Fast R CNN model to identify the coordinates of falling rocks and loose soil. is obtained by the Fast R CNN model to identify the coordinates of falling rocks and loose soil on x. is obtained by the Fast R CNN model to identify the coordinates of falling rocks and loose soil on y. is the L1 norm formula. and are both mathematical expectations.

[0031] Where: ; ; In the formula, G is a generator that converts the geological image into a radar image, F is a generator that converts the radar image into a geological image, and D x is a discriminator that determines whether the input belongs to the form of a radar image. D y is a discriminator that determines whether the input belongs to the form of a geological image. pdata(x): the true data distribution of the geological image, pdata(y): the true data distribution of the radar image, E: the expectation operator, indicating taking the average of all samples in the data distribution. : the discriminator D y takes the logarithm of the discrimination result of , and the goal is to maximize this value (encouraging the discriminator to correctly identify real samples). : the discriminator takes the logarithm complement of the discrimination result of the generated , and the goal is to maximize this value (encouraging the generator to deceive the discriminator). Similarly, : the discriminator takes the logarithm of the discrimination result of , and the goal is to maximize this value (encouraging the discriminator to correctly identify real samples). : the discriminator takes the logarithm complement of the discrimination result of the generated , and the goal is to maximize this value (encouraging the generator to deceive the discriminator).

[0032] In the present invention, the coordinate consistency loss of falling rocks and loose soil is additionally supplemented. , it is expected that the position coordinates of falling rocks and loose soil in the radar image generated by converting the geological image can be accurately mapped, so as to ensure that the position coordinates of falling rocks and loose soil are still accurate after the image form conversion, provide accurate position coordinates of falling rocks and loose soil in the subsequent multi-modal image fusion, realize integrating the position of falling rocks and loose soil in the geological image into the radar image, make the more abundant position information provided by multi-modal data, reduce the random error of position data of single-modal data, and make the positioning of falling rocks and loose soil more accurate, so that the actual trajectory calculation of falling rocks and loose soil is more accurate.

[0033] The present invention combines the attention mechanism in multi-modal data fusion. By using the key information focusing feature of the attention mechanism, it can assign high weights to important information in the radar conversion image and the radar image, highlight the key performance in the two modal images, realize information interaction between different modalities of the two images, improve the cross-modal understanding ability of the model, help the model better analyze the character features, and improve the quality of image multi-modal fusion. Therefore, through the multi-modal fusion of the attention mechanism, it can better improve the radar conversion image and the radar image, retain the key information of their respective modalities, and at the same time interact the position information of the falling rocks and loose soil sliding, and obtain accurate position information of the falling rocks and loose soil sliding, as follows: The method for generating a radar multi-modal image sequence of the instability risk area includes: Through the cross-attention mechanism, map the radar image sequence as the query Query matrix, map the radar conversion image sequence as the key Key and value Value matrices, and calculate the attention weights of the radar image sequence; Through the cross-attention mechanism, map the radar conversion image sequence as the query Query matrix, map the radar image sequence as the key Key and value Value matrices, and calculate the attention weights of the radar conversion image sequence; , where Q, K, and V are Query, Key, and Value respectively, is a scaling factor to prevent the dot product result from being too large and causing gradient disappearance.

[0034] Add the radar image sequence weighted by the attention weights of the radar image sequence and the radar conversion image sequence weighted by the attention weights of the radar conversion image sequence to obtain the radar multi-modal image sequence.

[0035] In the present invention, the phase interference method is used to calculate the actual trajectory of the falling rocks and loose soil sliding. Through a short period of the radar multi-modal image sequence during the outbreak period, the actual trajectory of the falling rocks and loose soil sliding is calculated, as follows: The calculation and solution method of the local instability trajectory includes: In the radar multi-mode image sequence, perform phase difference processing on two adjacent radar multi-mode images in sequence to generate an interferogram sequence , where is the interferogram at the t-th time sequence, is the phase information of the radar multi-mode image at the (t + 1)-th time sequence, is the phase information of the radar multi-mode image at the t-th time sequence, and m is the total number of time sequences of the radar multi-mode image sequence; Based on the phase difference information in the interferogram sequence, calculate the movement trajectories of falling rocks and loose soil in the instability risk area as local instability trajectories , where is the movement displacement of falling rocks and loose soil at the t-th time sequence, is the radar wavelength.

[0036] After calculating the actual trajectories of falling rocks and loose soil sliding in the present invention, the LSTM network with time series prediction function predicts the subsequent trajectories of falling rocks and loose soil sliding based on the actual trajectories of falling rocks and loose soil sliding during the instability outbreak period. The instability trend can be pre-grasped a short time after the outbreak period, and the instability trend can be quickly and real-time analyzed after the instability factor becomes unstable. Specifically as follows: The prediction method of the global instability trajectory includes: Use the LSTM network structure (Long Short-Term Memory, a special recurrent neural network designed to solve the long-term dependence problem faced by traditional RNNs in processing long sequence data) to perform subsequent prediction on the local instability trajectory to obtain the global instability trajectory , where is the movement displacement of falling rocks and loose soil at the t-th time sequence, m is the total number of time sequences of the radar multi-mode image sequence, and n is the total number of subsequent prediction time sequences. After predicting the subsequent trajectories of falling rocks and loose soil sliding in the present invention, judge whether their trajectories invade the mine protection area, that is, whether it will cause substantial harm. If it does, it will be an emergency event and a high-level warning will be issued. If not, a medium or low-level warning will be issued. The hierarchical warning of the present invention is in line with the actual situation, and the accuracy of the hierarchical warning is improved. Specifically as follows: The method for allocating warning levels includes: Calculate the distance from to the mine protection area. When the distance from to the mine protection area is less than the preset safety distance, allocate a high-risk level to the instability risk area;

[0037] If Figure 2As shown in the figure, the present invention provides a radar monitoring and early warning system based on the instability trend of mine slopes. According to the above-mentioned radar monitoring and early warning method based on the instability trend of mine slopes, the system includes: A vision device for collecting geological images; A slope radar for collecting radar images; A data processor for identifying, through a pre-trained Fast R CNN model, the slope areas with instability risks in the geological images as instability risk areas; determining the instability trend of the instability risk areas based on the radar image sequence and the geological image sequence, and then assigning warning levels according to the instability trend; A display device loaded with software for monitoring and grading early warning of open-pit mine slopes, used to display the radar images of the open-pit mine, integrally display the instability risk areas, and perform grading markings, instability trend display, and warning pop-ups on the warning areas in the radar images.

[0038] The vision device includes remote sensing image acquisition devices, UAV camera devices, etc., so as to obtain high-resolution satellite remote sensing images, UAV captured images, etc.

[0039] The present invention realizes the early locking of instability factors (i.e., the areas of falling rocks and loose soil that are prone to slope disaster states) based on the Fast R CNN model, and quickly and real-time analyzes the sliding trajectory trends of the falling rocks and loose soil on the actual trajectories of the falling rocks and loose soil obtained by solving multi-modal data based on the LSTM network, so as to perform corresponding grading early warnings, realizing the integrated display, focused monitoring, and grading early warning of the risk points in the mining area, and improving the effect of slope monitoring.

[0040] The above embodiments are only exemplary embodiments of the present application and are not used to limit the present application. The protection scope of the present application is defined by the claims. Those skilled in the art can make various modifications or equivalent replacements within the essence and protection scope of the present application, and such modifications or equivalent replacements should also be regarded as falling within the protection scope of the present application.

Claims

1. A radar monitoring and early warning method based on the instability trend of a mine slope, characterized in that Including the following steps: Collect geological images of the slope of an open-pit mine, and identify the slope area with instability risk on the geological images through a pre-trained Fast R CNN model as the instability risk area; Set the instability risk area as the monitoring area of the slope radar, and obtain a radar image sequence of the instability risk area by scanning the instability risk area in real time with the slope radar; Obtain a geological image sequence of the same time series as the radar image sequence of the instability risk area, determine the instability trend of the instability risk area based on the radar image sequence and the geological image sequence, and then assign a warning level according to the instability trend; The determination method of the instability trend includes: Construct an image conversion model for converting geological images into radar images through the CycleGAN network structure; Convert the geological images in the geological image sequence of the instability risk area into radar images through the image conversion model to generate a radar conversion image sequence of the instability risk area; Fuse the radar conversion image sequence of the instability risk area with the radar image sequence of the instability risk area to generate a radar multi-modal image sequence of the instability risk area; Solve the local instability trajectories of falling rocks and loose soil representing the instability trend in the radar multi-modal image sequence of the instability risk area, and predict the global instability trajectories of falling rocks and loose soil based on the local instability trajectories through the LSTM network structure; Carry out hierarchical warning on the instability risk areas on the open-pit mine according to the warning level.

2. The radar monitoring and early warning method based on the instability trend of the mine slope according to claim 1, wherein The pre-training method of the Fast R CNN model includes: Obtain multiple geological images, and mark the areas containing falling rocks and loose soil in each geological image as the instability risk areas; Divide the data set composed of multiple geological images into a test set and a training set; On the training set, use the geological images as input items and the instability risk areas of the geological images as output items to train the Fast R CNN model; On the test set, conduct performance evaluation on the Fast R CNN model for identifying the instability risk areas.

3. The radar monitoring and early warning method based on the instability trend of the mine slope according to claim 2, characterized in that, The construction method of the image conversion model includes: Obtain multiple geological images of the instability risk area, and obtain multiple radar images of the instability risk area at the same time as the geological images; Divide the data set composed of multiple geological images of the instability risk area and multiple radar images of the instability risk area into a training set and a test set; On the training set, use the first layer GAN network structure in the CycleGAN network structure to convert the geological images of the instability risk area into radar images of the instability risk area, and the second layer GAN network structure in the CycleGAN network structure to convert the radar images of the instability risk area into geological images of the instability risk area to train the CycleGAN network structure to obtain the image conversion model; On the test set, conduct performance evaluation on the image conversion model for image modality conversion; Among them, the loss function for training the CycleGAN network structure supplements and adds the coordinate consistency loss L GAN of falling rocks and loose soil on the basis of the cycle consistency loss Lcycle and the adversarial generation loss L P ; ; In the formula, is the radar image of the instability risk area converted from the geological image of the instability risk area, is the geological image of the instability risk area converted from the radar image of the instability risk area, x is the geological image of the instability risk area, and y is the radar image of the instability risk area, is the coordinates of the positions of falling rocks and loose soil identified by the Fast R CNN model on ; is the coordinates of the positions of falling rocks and loose soil identified by the Fast R CNN model on ; is the coordinates of the positions of falling rocks and loose soil identified by the Fast R CNN model on x, is the coordinates of the positions of falling rocks and loose soil identified by the Fast R CNN model on y, is the L1 norm formula, and are both mathematical expectations.

4. The radar monitoring and early warning method based on the instability trend of the mine slope according to claim 3, characterized in that, The generation method of the radar multi-modal image sequence of the instability risk area includes: Through the cross-attention mechanism, the radar image sequence is mapped into a query matrix, and the radar transformed image sequence is mapped into key and value matrices, and the attention weights of the radar image sequence are calculated; Through the cross-attention mechanism, the radar transformed image sequence is mapped into a query matrix, and the radar image sequence is mapped into key and value matrices, and the attention weights of the radar transformed image sequence are calculated; The radar image sequence weighted by the attention weights of the radar image sequence is added to the radar transformed image sequence weighted by the attention weights of the radar transformed image sequence to obtain the radar multimodal image sequence.

5. The radar monitoring and early warning method based on the instability trend of the mine slope according to claim 4, characterized in that, The calculation and solution method of the local instability trajectory includes: In the radar multi-mode image sequence, perform phase difference processing on two adjacent radar multi-mode images in sequence to generate an interferogram sequence , where is the interferogram at the t-th time sequence, is the phase information of the radar multi-mode image at the (t + 1)-th time sequence, is the phase information of the radar multi-mode image at the t-th time sequence, and m is the total number of time sequences of the radar multi-mode image sequence; Based on the phase difference information in the interferogram sequence, calculate the movement trajectories of falling rocks and loose soil in the instability risk area as local instability trajectories , where is the movement displacement of falling rocks and loose soil at the t-th time series, is the radar wavelength.

6. The radar monitoring and early warning method based on the instability trend of the mine slope according to claim 5, characterized in that, The prediction method of the global instability trajectory includes: Subsequent prediction of the local instability trajectory is carried out using the LSTM network structure to obtain the global instability trajectory , where is the movement displacement of the rockfall and loose soil at the t-th time series, m is the total number of time series of the radar multi-mode image sequence, and n is the total number of subsequent prediction time series.

7. The radar monitoring and early warning method based on the instability trend of the mine slope according to claim 6, wherein The method for allocating the warning level includes: Calculation The distance to the mine protection area, when The distance to the mine protection area is less than the preset safety distance, then assign a high-risk level to the instability risk area; When the distance from the mining area protection zone is greater than or equal to the preset safety distance, a medium - low risk level is assigned to the instability risk area.

8. A radar monitoring and early warning system based on the instability trend of the mine slope, characterized in that, The radar monitoring and warning method based on the instability trend of the mine slope according to any one of claims 1-7 includes: A vision device for collecting geological images; A slope radar for collecting radar images; A data processor for identifying, through a pre-trained Fast R CNN model, a slope area with instability risk on the geological image as an instability risk area; determining the instability trend of the instability risk area based on the radar image sequence and the geological image sequence, and then allocating a warning level according to the instability trend; A display device, loaded with software for monitoring and grading warnings of open-pit mine slopes, for displaying radar images of the open-pit mine, and performing grading markings, instability trend display, and warning pop-ups on the warning areas on the radar images.

9. The radar monitoring and early warning system based on the instability trend of the mine slope according to claim 8, characterized in that, The vision device includes a remote sensing image acquisition device and a drone camera device.

Images