A method and device for identifying potential hazards in dikes based on a bionic robot dog

Abstract

This invention relates to the field of water conservancy project safety monitoring, and particularly to a method and device for identifying levee hazards based on a bionic robot dog. The method includes: controlling a bionic robot dog to acquire visible light images and infrared thermal images of multiple levee monitoring points along an optimal path; determining fusion features based on the visible light images and infrared thermal images; wherein the fusion features are used to characterize the hazard features of the levee monitoring points; inputting the fusion features into a preset deep learning model to determine the hazard categories and confidence levels of the multiple monitoring points; determining a state network diagram of levee hazards based on the hazard categories and confidence levels of the multiple monitoring points; and determining the hazard status of the levee based on the state network diagram. Thus, this invention can improve the efficiency of levee hazard identification while ensuring the accuracy of levee hazard identification.

Description

Technical Field

[0001] This invention relates to the field of safety monitoring of water conservancy projects, and in particular to a method and device for identifying potential hazards in dikes based on a bionic robot dog. Background Technology

[0002] The main hidden dangers of dikes include defects (cracks, collapses, scour pits, etc.). These hidden dangers are characterized by their high degree of concealment, rapid development, and high degree of harm. If they are not identified and dealt with in a timely manner, they can easily lead to major disasters such as dike breaches and breaches.

[0003] Currently, the inspection of potential hazards along dikes mainly relies on traditional manual inspections supplemented by some mechanized and automated equipment, which presents several technical challenges: First, manual inspections are inefficient. Due to the long length, wide span, and complex terrain of the dikes, the average daily inspection mileage per person is limited, making it difficult to achieve full coverage and high-frequency inspections, and the rate of missed inspections is relatively high. Second, the operation is risky. Dikes often involve slopes, muddy areas, and water-related areas, making manual inspections prone to accidents such as slipping and drowning, especially during severe weather such as heavy rain and floods. Third, the accuracy of hazard identification is insufficient. Existing technologies mostly rely on data from single sensors and lack the ability to fuse and analyze multi-source data, making it difficult to achieve collaborative identification of surface and internal hazards. Furthermore, the identification results are highly subjective, and early warnings are delayed.

[0004] Based on this, the present invention proposes a method and device for identifying potential hazards in dikes based on a bionic robot dog to solve the above-mentioned technical problems. Summary of the Invention

[0005] This invention describes a method and device for identifying potential hazards in dikes based on a bionic robot dog, which can improve the efficiency of identifying potential hazards while ensuring the accuracy of the identification.

[0006] According to a first aspect, the present invention provides a method for identifying potential hazards in dikes based on a bionic robot dog, comprising:

[0007] The bionic robot dog is controlled to acquire visible light images and infrared thermal images of multiple dike monitoring points according to the optimal path.

[0008] Based on the visible light image and the infrared thermal image, fusion features are determined; wherein, the fusion features are used to characterize the potential hazard features at the dike monitoring points.

[0009] The fused features are input into a preset deep learning model to determine the hazard categories and confidence levels of multiple monitoring points;

[0010] Based on the hazard categories and confidence levels of multiple monitoring points, a state network diagram of levee hazards is determined;

[0011] Based on the aforementioned state network diagram, the potential hazard status of the dike is determined.

[0012] According to a second aspect, the present invention provides a levee hazard identification device based on a bionic robot dog, comprising:

[0013] The acquisition unit is configured to control the bionic robot dog to acquire visible light images and infrared thermal images of multiple dike monitoring points according to the optimal path;

[0014] The second data processing unit is configured to determine fusion features based on the visible light image and the infrared thermal image; wherein, the fusion features are used to characterize the potential hazard features of the dike monitoring points.

[0015] The third data processing unit is configured to input the fused features into a preset deep learning model to determine the hazard categories and confidence levels of multiple monitoring points;

[0016] The fourth data processing unit is configured to determine a state network diagram of levee hazards based on the hazard categories and confidence levels of multiple monitoring points;

[0017] The fifth data processing unit is configured to determine the potential danger status of the dike based on the state network diagram.

[0018] Thirdly, embodiments of this specification also provide an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, it implements the method described in any embodiment of this specification.

[0019] Fourthly, embodiments of this specification also provide a computer-readable storage medium having a computer program stored thereon, which, when executed in a computer, causes the computer to perform the methods described in any embodiment of this specification.

[0020] According to the present invention, a method and apparatus for identifying potential hazards in dikes based on a bionic robot dog acquires visible light images and infrared thermal images of multiple dike monitoring points along an optimal path. The optimal path achieves a dynamic balance between sampling quality and robot dog energy consumption, ensuring comprehensive image acquisition coverage and data validity, while adapting to complex dike terrain and improving mobility. Subsequently, based on the visible light images and infrared thermal images, fusion features are determined; these fusion features characterize the potential hazard features of each dike monitoring point. Specifically, Gaussian filtering and image enhancement techniques are first used to eliminate noise in the visible light images and enhance the contours of surface defects such as cracks and collapses. Temperature calibration and initial screening of abnormal areas are performed on the infrared thermal images to capture temperature field differences caused by seepage. Then, through feature fusion, texture and shape features from the visible light images are extracted, along with temperature gradients and abnormal distribution features from the infrared thermal images, forming fusion features that combine surface details with information related to hidden hazards. These features comprehensively characterize the potential hazard features of each dike monitoring point. The aforementioned fused features are input into a pre-defined deep learning model trained and optimized using a sample database of levee hazards. The model, through feature matching and classification reasoning, outputs the hazard category and corresponding confidence level for each monitoring point. The confidence level quantifies the reliability of the identification results and effectively filters out false positives. Based on the hazard category and confidence level of each monitoring point, a levee hazard status network diagram is constructed. Finally, the hazard status of the levee is determined through the status network diagram. Thus, this invention, through path optimization and dual-modal fusion, significantly reduces the cost of manual intervention while ensuring the accuracy of levee hazard identification, improving efficiency by 5-8 times compared to traditional manual inspections, achieving efficient, accurate, and automated identification of levee hazards. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 A flowchart illustrating a method for identifying potential hazards in dikes based on a bionic robot dog, according to one embodiment, is shown.

[0023] Figure 2 A schematic block diagram of a levee hazard identification device based on a bionic robot dog according to one embodiment is shown. Detailed Implementation

[0024] The solution provided by the present invention will now be described with reference to the accompanying drawings.

[0025] Figure 1This diagram illustrates a flowchart of a method for identifying potential hazards in dikes based on a bionic robot dog, according to one embodiment. It is understood that this method can be executed by any device, equipment, platform, or cluster of devices with computing and processing capabilities. Figure 1 As shown, the method includes:

[0026] Step 100: Control the bionic robot dog to acquire visible light images and infrared thermal images of multiple dike monitoring points according to the optimal path;

[0027] Step 102: Based on the visible light image and the infrared thermal image, determine the fusion features; whereby the fusion features are used to characterize the hidden danger features of the levee monitoring points.

[0028] Step 104: Input the fused features into the preset deep learning model to determine the hazard categories and confidence levels of multiple monitoring points;

[0029] Step 106: Based on the hazard categories and confidence levels of multiple monitoring points, determine the state network diagram of levee hazards;

[0030] Step 108: Based on the state network diagram, determine the potential hazard status of the dike.

[0031] In this embodiment, a bionic robot dog is controlled to acquire visible light images and infrared thermal images of multiple levee monitoring points along an optimal path. The optimal path achieves a dynamic balance between sampling quality and robot dog energy consumption, ensuring comprehensive image acquisition coverage and data validity, while adapting to the complex terrain of the levee and improving mobility. Subsequently, based on the visible light images and infrared thermal images, fusion features are determined; these fusion features characterize the potential hazards at each levee monitoring point. Specifically, Gaussian filtering and image enhancement techniques are first used to eliminate noise in the visible light images and enhance the contours of surface defects such as cracks and collapses. Temperature calibration and initial screening of abnormal areas are performed on the infrared thermal images to capture temperature field differences caused by seepage. Then, through feature fusion, texture and shape features from the visible light images are extracted, along with temperature gradients and abnormal distribution features from the infrared thermal images, forming fusion features that combine surface details with information related to hidden hazards. These features comprehensively characterize the potential hazards at each levee monitoring point. The aforementioned fused features are input into a pre-defined deep learning model trained and optimized using a sample database of levee hazards. The model outputs the hazard category and corresponding confidence level for each monitoring point through feature matching and classification reasoning. The confidence level quantifies the reliability of the identification results and effectively filters out false positives. Based on the hazard category and confidence level of each monitoring point, a network diagram of levee hazard status is constructed.

[0032] Finally, the state network diagram is used to determine the status of potential hazards in the dike. Thus, this invention, through path optimization and dual-modal fusion, significantly reduces the cost of manual intervention while ensuring the accuracy of dike hazard identification. It improves efficiency by 5-8 times compared to traditional manual inspection, achieving efficient, accurate, and automated identification of dike hazards.

[0033] In one embodiment of the present invention, a state network diagram of levee hazards is determined based on the hazard categories and confidence levels of multiple monitoring points, including:

[0034] Based on the location of the monitoring points, a point cloud map of the area to be monitored is constructed;

[0035] The corresponding hazard categories and confidence levels are assigned to the point cloud map to obtain the state network diagram of levee hazards;

[0036] The categories of potential hazards include cracks, collapses, and scour pits.

[0037] In this embodiment, firstly, relying on the LiDAR data collected by the bionic robot dog and combined with the precise GNSS coordinates of each monitoring point, a three-dimensional point cloud map of the embankment section to be monitored is constructed. This fully restores the surface topography, structural outline, and spatial distribution of monitoring points of the embankment, providing a spatial carrier for mapping hazard information. Subsequently, the hazard categories (including three core surface hazards: cracks, collapses, and scour pits) and corresponding confidence levels of each monitoring point, identified by a deep learning model, are accurately assigned to the corresponding coordinate points in the point cloud map. Simultaneously, differentiated feature labels are marked for different categories of hazards, and points with confidence levels below a set threshold are initially screened and labeled, ultimately forming a network map of the embankment hazard status.

[0038] In one embodiment of the present invention, determining the potential hazard status of a dike based on a state network diagram includes:

[0039] The state network graph is subjected to feature extraction to obtain the state network graph features;

[0040] Based on the characteristics of the state network diagram, the potential hazard status of the dike is determined;

[0041] Among them, the features of the state network diagram include point cloud topology features, hazard distribution density features, hazard location coordinate features, similar hazard association features, dissimilar hazard collaboration features, confidence gradient features, and hazard magnitude features.

[0042] In this embodiment, targeted feature extraction is first performed on the state network diagram, which integrates the spatial coordinates of monitoring points, hazard categories, and confidence levels. Core features characterizing the distribution, correlation, and severity of hazards are then extracted to form a standardized feature set. These features include: point cloud topology features, reflecting the spatial connectivity and distance between monitoring points, and showing a three-dimensional spread trend of hazards related to the slope of the dike terrain; hazard distribution density features, quantifying the number and clustering of hazard points per unit area to locate concentrated hazard areas; hazard point coordinate features, pinpointing the specific locations of hazards at the top, slope, and toe of the dike based on three-dimensional coordinates to clarify the risk area attribution; correlation features of similar hazards, identifying the spatial connectivity and spread direction of hazards of the same category to determine whether a large-scale hazard zone has formed; synergistic features of dissimilar hazards, analyzing the spatial superposition relationship of different categories of hazards to assist in identifying compound hazard risks; confidence level gradient features, capturing differences in confidence changes between adjacent points to screen hazard boundaries and suspected areas of hidden hazards; and hazard magnitude features, combining point cloud data to derive hazard size parameters to quantify the severity of cracks, collapses, and scour pits.

[0043] In one embodiment of the present invention, the optimal path includes a first path and a second path, which respectively correspond to two different sampling paths of the bionic robot dog. Each monitoring point on the dike corresponds to at least two preset sampling points, and different sampling points correspond to different sampling angles and sampling distances.

[0044] The first path was determined through the following steps:

[0045] Step 200: Take any sampling point corresponding to the edge of the dike monitoring point as the current first moving point;

[0046] Step 202: Calculate the comprehensive cost of each sampling point corresponding to the current first moving point to the next levee monitoring point; whereby the comprehensive cost is used to balance the sampling quality and the energy consumption of the bionic robot dog.

[0047] Step 204: Select the sampling point with the lowest overall cost as the next moving point, and control the bionic robot dog to move from the current first moving point to the next moving point;

[0048] Step 206: Repeat steps 202 to 204 until all monitoring points along the dike have been sampled, in order to determine the first path.

[0049] In this embodiment, the optimal path includes a first path and a second path. Each path corresponds to two different bionic robot dogs independently performing sampling tasks. Parallel operation of multiple devices improves inspection efficiency, while multi-angle sampling ensures comprehensive hazard identification. Each preset levee monitoring point is configured with at least two differentiated sampling points. Different sampling points correspond to different sampling angles and distances, allowing for targeted capture of feature information from different areas of the levee surface. For example, some sampling points use close-range vertical sampling to focus on minute cracks and other detailed defects; others use mid-range inclined sampling to cover a larger surface area, avoiding missed detections caused by single-angle or single-distance sampling, and providing data support for fusion feature extraction and hazard identification. The determination of the first path is as follows: Step 200, path initialization: Select any sampling point at the edge of the monitoring point from all levee monitoring points as the initial current first moving point. This edge point can serve as the path starting point, ensuring that subsequent sampling gradually covers the entire monitored area and avoiding path redundancy. Step 202: Calculate the comprehensive cost of each candidate sampling point from the current first moving point to the next levee monitoring point. This comprehensive cost is used to balance sampling quality and the energy consumption of the bionic robot dog, ensuring that the sampling points provide clear and effective image data while also controlling the robot dog's energy consumption. This avoids insufficient battery life due to excessive pursuit of sampling quality or sacrificing recognition accuracy for energy saving. Step 204: Select the point with the minimum comprehensive cost from all candidate sampling points as the next moving point. This point maximizes the effectiveness of the sampling data while minimizing the energy consumption and path travel time during the robot dog's movement. Then, control the bionic robot dog to smoothly move from the current first moving point to the next moving point along the planned trajectory. Step 206: Repeat steps 202 to 204. After completing the movement and data collection of each sampling point, use that point as the new current first moving point and continue to calculate and select the optimal sampling point for the next monitoring point until all levee monitoring points have completed sampling operations, ultimately forming a first path that covers the entire area and balances efficiency and quality.

[0050] In one embodiment of the present invention, the second path is determined by the following steps:

[0051] Step 300: The sampling point corresponding to any edge of the dike monitoring point is taken as the current second moving point; wherein, the sampling points selected by the second path are all different from the sampling points selected by the first path;

[0052] Step 302: Calculate the comprehensive cost of each sampling point corresponding to the current second moving point to the next levee monitoring point;

[0053] Step 304: Select the sampling point with the lowest overall cost as the next moving point, and control the bionic robot dog to move from the current second moving point to the next moving point;

[0054] Step 306: Repeat steps 302 to 304 until all monitoring points along the dike have been sampled, in order to determine the second path.

[0055] In this embodiment, to achieve parallel full-angle sampling by two machines, the second path adopts a differentiated point selection strategy from the first path to ensure that all different sampling points of each dike monitoring point are covered, further improving the comprehensiveness of data collection. The determination steps are as follows: Step 300, path initialization setting of the current second moving point. Select the sampling point corresponding to any monitoring point at the edge of the dike monitoring points as the initial current second moving point. The constraint is that all sampling points selected by the second path are completely different from the sampling points selected by the first path. This achieves non-overlapping coverage of the two types of sampling points and compensates for the blind spots of single-path sampling based on the complementarity of different sampling angles and distances. Step 302, calculate the comprehensive cost of each candidate sampling point corresponding to the current second moving point to the next dike monitoring point, balancing the effectiveness of sampling data and the energy consumption of the bionic robot dog, ensuring that path planning takes into account both accuracy and endurance. Step 304, optimal point selection and movement control. From all candidate sampling points, the point with the lowest overall cost is selected as the next moving point. This point maximizes sampling quality while minimizing movement energy consumption and time. The corresponding bionic robot dog is then controlled to move smoothly from the current second moving point to the next moving point along a trajectory adapted to the levee terrain, ensuring the stability of the data collection process. Step 306: Iteratively generate the path. Steps 302 to 304 are repeated. After sampling at each point, that point is used as the new current second moving point. The optimal sampling point for the next monitoring point is calculated and selected, until all levee monitoring points have completed differentiated sampling, ultimately forming a second path that complements the first path and covers the entire area.

[0056] In one embodiment of the present invention, the overall cost is calculated using the following formula:

[0057]

[0058]

[0059]

[0060]

[0061] In the formula, For the overall cost, The weights are dynamic, and k is the cost index. For the cost of distance, This represents the straight-line distance between the current moving point and the target sampling point. This represents the maximum distance from the current moving point to all corresponding sampling points for the next monitoring point. This is the set of all sampling points corresponding to the next monitoring point. For the sake of sampling quality, The actual sampling angle of the target sampling point. The actual sampling distance of the target sampling point. To achieve the optimal sampling angle, For the optimal sampling distance, The sampling angle adaptation threshold, The sampling distance adaptability threshold, As a result of energy loss, The energy consumption coefficient per unit distance. The terrain complexity between the current moving point and the target sampling point. This represents the maximum power consumption for a single battery run of the robot dog. This is a terrain dynamic correction function.

[0062] In this embodiment, the formula uses the current movement point. P i To the target sampling point P j As the calculation dimension, the comprehensive cost (range [0,1]) is output by weighted summation of distance cost, sampling quality cost, and energy loss cost with dynamic weighting factors. A smaller value indicates a better path. Here, k=1,2,3 correspond to distance cost, sampling quality cost, and energy loss cost, respectively. These three types of costs are normalized to eliminate dimensional differences and can be directly weighted and fused. Traditional path planning often focuses solely on distance or energy consumption, easily leading to either prioritizing efficiency over quality or prioritizing quality over energy consumption. This formula achieves multi-objective balance by fusing distance, sampling quality, and energy consumption costs with dynamic weights, adapting to the complex terrain of embankments and the need for high-precision sampling. It solves the technical pain point of traditional algorithms in balancing inspection efficiency and hazard identification accuracy. Normalization processing enables cross-dimensional cost fusion: By normalizing the three types of costs respectively, the technical problem of not being able to directly fuse costs with different physical meanings and inconsistent dimensions, such as distance, sampling quality, and energy consumption, is solved. This allows the comprehensive cost to quantify the optimality of the path. Compared with traditional fusion algorithms without normalization, the recognition accuracy and the rationality of path planning are significantly improved.

[0063] In addition, it is adapted to dual-path differentiated sampling scenarios: the formula supports cost calculation for multiple sampling points (different angles and distances) at the same monitoring point, providing algorithmic support for the selection of differentiated points for the first and second paths, realizing parallel full-angle sampling by dual machines, avoiding blind spots in the field of view, and further improving the comprehensiveness and reliability of levee hazard identification compared with the single-path sampling scheme, which meets the actual needs of high precision and high efficiency in levee inspection.

[0064] In one embodiment of the present invention, the dynamic weights are determined by the following formula:

[0065]

[0066] In the formula, The basic weights for the k-th class cost. is the weight adjustment coefficient for the cost of class k, and m is the summation index.

[0067] In this embodiment, the weight adjustment coefficient of the k-th type of cost is used to control the sensitivity of the weight to terrain complexity. The larger the value of λ, the more significant the change in weight with terrain. The terrain complexity between the current moving point and the target sampling point comprehensively reflects terrain features such as slope and mud level between the two points; the larger the value, the more complex the terrain. The base weight of the k-th type of cost is the initial weight when the terrain is flat, providing a benchmark for dynamic adjustment. Reference value. It is 0.4. It is 0.3. The value is 0.3. This invention introduces a dynamic weighting mechanism to improve terrain adaptability. Unlike the traditional weighted formula with fixed weights, the dynamic weights in this formula are dynamically adjusted according to the terrain complexity between the current moving point and the target sampling point. Smooth nonlinear optimization is achieved through the natural exponential function, enabling path planning to adaptively adjust the cost weights according to the terrain (flat, sloping, muddy, etc.). In complex terrain, sampling quality is prioritized, while energy consumption and efficiency are optimized in flat terrain.

[0068] The foregoing has described specific embodiments of the invention. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired results. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0069] According to another embodiment, the present invention provides a levee hazard identification device based on a bionic robot dog. Figure 2A schematic block diagram of a levee hazard identification device based on a bionic robotic dog, according to one embodiment, is shown. It will be understood that this device can be implemented using any device, equipment, platform, or cluster of devices with computing and processing capabilities. Figure 2 As shown, the device includes: an acquisition unit 400, a first data processing unit 402, a second data processing unit 404, a third data processing unit 406, and a fourth data processing unit 408. The main functions of each component are as follows:

[0070] The acquisition unit 400 is configured to control the bionic robot dog to acquire visible light images and infrared thermal images of multiple dike monitoring points according to the optimal path.

[0071] The first data processing unit 402 is configured to determine fusion features based on the visible light image and the infrared thermal image; wherein, the fusion features are used to characterize the hidden danger features of the dike monitoring points.

[0072] The second data processing unit 404 is configured to input the fused features into a preset deep learning model to determine the hazard categories and confidence levels of multiple monitoring points.

[0073] The third data processing unit 406 is configured to determine a state network diagram of levee hazards based on the hazard categories and confidence levels of multiple monitoring points.

[0074] The fourth data processing unit 408 is configured to determine the potential danger status of the dike based on the state network diagram.

[0075] In one embodiment of the present invention, the third data processing unit 406 is configured to perform the following operations:

[0076] Based on the location of the monitoring points, a point cloud map of the area to be monitored is constructed;

[0077] The corresponding hazard category and confidence level are assigned to the point cloud map to obtain the state network diagram of the levee hazard;

[0078] The categories of potential hazards include cracks, collapses, and scour pits.

[0079] In one embodiment of the present invention, the fourth data processing unit 408 is configured to perform the following operations:

[0080] The state network graph is subjected to feature extraction to obtain state network graph features;

[0081] Based on the characteristics of the state network diagram, the potential danger status of the dike is determined;

[0082] The state network diagram features include point cloud topology features, hazard distribution density features, hazard location coordinate features, similar hazard association features, dissimilar hazard collaboration features, confidence gradient features, and hazard magnitude features.

[0083] In one embodiment of the present invention, the optimal path includes a first path and a second path, the first path and the second path respectively corresponding to two different sampling paths of the bionic robot dog, and each of the dike monitoring points corresponds to at least two preset sampling points, with different sampling points corresponding to different sampling angles and sampling distances:

[0084] The device further includes a fifth data processing unit, which is configured to perform the following operations:

[0085] Step 200: Take the sampling point corresponding to any edge of the dike monitoring point as the current first moving point;

[0086] Step 202: Calculate the comprehensive cost of each sampling point corresponding to the current first moving point to the next levee monitoring point; wherein, the comprehensive cost is used to balance the sampling quality and the energy consumption of the bionic robot dog;

[0087] Step 204: Select the sampling point with the lowest overall cost as the next moving point, and control the bionic robot dog to move from the current first moving point to the next moving point;

[0088] Step 206: Repeat steps 202 to 204 until all monitoring points along the dike have been sampled, in order to determine the first path.

[0089] In one embodiment of the present invention, the apparatus further includes a sixth data processing unit, the sixth data processing unit being configured to perform the following operations:

[0090] Step 300: Take the sampling point corresponding to any edge of the dike monitoring point as the current second moving point; wherein, the sampling point selected by the second path is different from the sampling point selected by the first path;

[0091] Step 302: Calculate the comprehensive cost of each sampling point corresponding to the current second moving point to the next levee monitoring point;

[0092] Step 304: Select the sampling point with the lowest overall cost as the next moving point, and control the bionic robot dog to move from the current second moving point to the next moving point;

[0093] Step 306: Repeat steps 302 to 304 until all monitoring points along the dike have been sampled, in order to determine the second path.

[0094] In one embodiment of the present invention, the comprehensive cost is calculated using the following formula:

[0095]

[0096]

[0097]

[0098]

[0099] In the formula, For the aforementioned comprehensive cost, The weights are dynamic, and k is the cost index. For the cost of distance, This represents the straight-line distance between the current moving point and the target sampling point. This represents the maximum distance from the current moving point to all corresponding sampling points for the next monitoring point. This is the set of all sampling points corresponding to the next monitoring point. For the sake of sampling quality, The actual sampling angle of the target sampling point. The actual sampling distance of the target sampling point. To achieve the optimal sampling angle, For the optimal sampling distance, The sampling angle adaptation threshold, The sampling distance adaptability threshold, As a result of energy loss, The energy consumption coefficient per unit distance. The terrain complexity between the current moving point and the target sampling point. This represents the maximum power consumption for a single battery run of the robot dog. This is a terrain dynamic correction function.

[0100] In one embodiment of the present invention, the dynamic weight is determined by the following formula:

[0101]

[0102] In the formula, The basic weights for the k-th class cost. is the weight adjustment coefficient for the cost of class k, and m is the summation index.

[0103] According to another embodiment, a computer-readable storage medium is also provided, on which a computer program is stored, which, when executed in a computer, causes the computer to perform a combination Figure 1 The method described.

[0104] According to another embodiment, an electronic device is also provided, including a memory and a processor, wherein the memory stores executable code, and when the processor executes the executable code, it implements a combination... Figure 1 The method described.

[0105] The various embodiments in this invention are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the apparatus embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0106] Those skilled in the art will recognize that, in one or more of the examples above, the functions described in this invention can be implemented using hardware, software, firmware, or any combination thereof. When implemented in software, these functions can be stored in a computer-readable medium or transmitted as one or more instructions or code on a computer-readable medium.

[0107] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made on the basis of the technical solution of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for identifying potential hazards in dikes based on a bionic robot dog, characterized in that, include: The bionic robot dog is controlled to acquire visible light images and infrared thermal images of multiple dike monitoring points according to the optimal path. Based on the visible light image and the infrared thermal image, fusion features are determined; wherein, the fusion features are used to characterize the potential hazard features at the dike monitoring points. The fused features are input into a preset deep learning model to determine the hazard categories and confidence levels of multiple monitoring points; Based on the hazard categories and confidence levels of multiple monitoring points, a state network diagram of levee hazards is determined; Based on the aforementioned state network diagram, the potential hazard status of the dike is determined; The optimal path includes a first path and a second path, which correspond to two different sampling paths of the bionic robot dog. Each monitoring point on the dike corresponds to at least two preset sampling points, and different sampling points correspond to different sampling angles and sampling distances. The first path was determined through the following steps: Step 200: Take the sampling point corresponding to any edge of the dike monitoring point as the current first moving point; Step 202: Calculate the comprehensive cost of each sampling point corresponding to the current first moving point to the next levee monitoring point; wherein, the comprehensive cost is used to balance the sampling quality and the energy consumption of the bionic robot dog; Step 204: Select the sampling point with the lowest overall cost as the next moving point, and control the bionic robot dog to move from the current first moving point to the next moving point; Step 206: Repeat steps 202 to 204 until all monitoring points along the dike have been sampled, in order to determine the first path; The determination of the state network diagram of dike hazards based on the hazard categories and confidence levels of multiple monitoring points includes: Based on the location of the monitoring points, a point cloud map of the area to be monitored is constructed; The corresponding hazard category and confidence level are assigned to the point cloud map to obtain the state network diagram of the levee hazard; The categories of potential hazards include cracks, collapses, and scour pits; The process of determining the potential hazard status of the dike based on the state network diagram includes: The state network graph is subjected to feature extraction to obtain state network graph features; Based on the characteristics of the state network diagram, the potential danger status of the dike is determined; The state network diagram features include point cloud topology features, hazard distribution density features, hazard location coordinate features, similar hazard association features, dissimilar hazard collaboration features, confidence gradient features, and hazard magnitude features.

2. The method according to claim 1, characterized in that, The second path was determined through the following steps: Step 300: Take the sampling point corresponding to any edge of the dike monitoring point as the current second moving point; wherein, the sampling point selected by the second path is different from the sampling point selected by the first path; Step 302: Calculate the comprehensive cost of each sampling point corresponding to the current second moving point to the next levee monitoring point; Step 304: Select the sampling point with the lowest overall cost as the next moving point, and control the bionic robot dog to move from the current second moving point to the next moving point; Step 306: Repeat steps 302 to 304 until all monitoring points along the dike have been sampled, in order to determine the second path.

3. The method according to claim 2, characterized in that, The overall cost is calculated using the following formula: In the formula, For the aforementioned comprehensive cost, The weights are dynamic, and k is the cost index. For the cost of distance, This represents the straight-line distance between the current moving point and the target sampling point. This represents the maximum distance from the current moving point to all corresponding sampling points for the next monitoring point. This is the set of all sampling points corresponding to the next monitoring point. For the sake of sampling quality, The actual sampling angle of the target sampling point. The actual sampling distance of the target sampling point. To achieve the optimal sampling angle, For the optimal sampling distance, The sampling angle adaptation threshold, The sampling distance adaptability threshold, As a result of energy loss, The energy consumption coefficient per unit distance. The terrain complexity between the current moving point and the target sampling point. This represents the maximum power consumption for a single battery run of the robot dog. This is a terrain dynamic correction function.

4. The method according to claim 3, characterized in that, The dynamic weights are determined by the following formula: In the formula, The basic weights for the k-th class cost. is the weight adjustment coefficient for the cost of class k, and m is the summation index.

5. A levee hazard identification device based on a bionic robot dog, characterized in that, For performing the method as described in any one of claims 1-4, comprising: The acquisition unit is configured to control the bionic robot dog to acquire visible light images and infrared thermal images of multiple dike monitoring points according to the optimal path; The first data processing unit is configured to determine fusion features based on the visible light image and the infrared thermal image; wherein, the fusion features are used to characterize the potential hazard features of the dike monitoring points. The second data processing unit is configured to input the fused features into a preset deep learning model to determine the hazard categories and confidence levels of multiple monitoring points; The third data processing unit is configured to determine the state network diagram of levee hazards based on the hazard categories and confidence levels of multiple monitoring points; The fourth data processing unit is configured to determine the potential danger status of the dike based on the state network diagram.

6. An electronic device, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method as described in any one of claims 1-4.

7. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed in a computer, causes the computer to perform the method described in any one of claims 1-4.

Images