Landslide rockfall detection method, device and equipment

Abstract

The embodiment of the invention provides a landslide rockfall detection method, device and equipment. The method comprises the steps of obtaining a landslide rockfall image of a to-be-detected geology and a system resource occupation state; calculating texture complexity and edge density of the landslide rockfall image data; the scene complexity is judged according to the texture complexity and the edge density; determining an initial detection model from a plurality of to-be-selected detection models according to the system resource occupation state and the scene complexity; determining a target detection model based on the comprehensive score of the initial detection model; based on the target detection model, landslide rockfall detection is carried out on the to-be-detected geology, and a target detection result is obtained. According to the invention, effective balance between real-time performance and detection precision is realized; a time-precision comprehensive evaluation index S is adopted to guide model optimization, and the detection precision is maximized while the real-time requirement is met.

Description

Technical Field

[0001] The present invention relates to the field of geological disaster monitoring, and relates to but is not limited to a landslide and rockfall detection method, device and equipment. Background Art

[0002] Landslides and rockfalls are common geological disasters that pose a serious threat to human life and property. Traditional monitoring methods mainly rely on manual inspections and simple sensor monitoring, which have problems such as poor real-time performance, limited coverage, and low accuracy. In recent years, with the development of deep learning and computer vision technology, image-based target detection methods have gradually been applied to the field of geological disaster monitoring. However, existing target detection models still have the following shortcomings when dealing with complex geological targets such as landslides and rockfalls:

[0003] (1) It is difficult for a single model to achieve both real-time performance and accuracy: high-precision models (such as Faster R-CNN) have high computational complexity and are difficult to meet real-time monitoring requirements; while lightweight models (such as YOLOv8) are fast but lack detection accuracy, especially in complex environments.

[0004] (2) Insufficient resource utilization and inability to dynamically optimize performance: Existing methods are usually fixed models that fail to dynamically adjust according to the system resource status. When resources are sufficient, computing power cannot be fully utilized to improve detection accuracy; when resources are tight, the model may not meet real-time requirements, resulting in low resource utilization;

[0005] (3) The model has poor adaptability to environmental changes: Under different lighting, weather, and terrain conditions, the model’s detection performance fluctuates greatly and lacks robustness. In addition, most target detection models are designed for general objects and do not fully consider the characteristics of landslides and falling rocks, resulting in unsatisfactory detection results. Summary of the invention

[0006] The present invention provides a landslide and rockfall detection method, device and equipment, aiming to solve the problem of difficulty in balancing the real-time performance and detection accuracy in the prior art of landslide and rockfall detection.

[0007] The technical method of the embodiment of the present invention is implemented as follows:

[0008] In a first aspect, an embodiment of the present invention provides a landslide and rockfall detection method, the method comprising:

[0009] Obtain the landslide and rockfall images of the geological structure to be detected and the system resource occupancy status;

[0010] Calculating the texture complexity and edge density of the landslide rockfall image data;

[0011] Determining scene complexity according to the texture complexity and the edge density;

[0012] Determining an initial detection model from a plurality of detection models to be selected according to the system resource occupancy status and the scene complexity;

[0013] Determining a target detection model based on the comprehensive score of the initial detection model;

[0014] Based on the target detection model, landslide and rockfall detection is performed on the geology to be detected to obtain a target detection result.

[0015] In a second aspect, an embodiment of the present invention provides a landslide and rockfall detection device, the device comprising:

[0016] An acquisition module is used to obtain the landslide and rockfall images of the geological objects to be monitored and the system resource occupancy status;

[0017] A calculation module, used for calculating the texture complexity and edge density of the landslide and rockfall image data;

[0018] A judging module, used for judging scene complexity according to the texture complexity and the edge density;

[0019] A determination module, used to determine an initial detection model from a plurality of detection models to be selected according to the system resource occupancy status and the scene complexity;

[0020] The determination module is further used to determine the target detection model based on the comprehensive score of the initial detection model;

[0021] The detection module is used to perform landslide and rockfall detection on the geology to be monitored based on the target detection model to obtain a target detection result.

[0022] In some embodiments, the acquisition module is also used to compare the system resource occupancy status with a preset occupancy rate threshold to obtain a first comparison result; if the first comparison result is that the system resource occupancy status exceeds the preset occupancy rate threshold, it is determined that the system resources are insufficient; if the first comparison result is that the system resource occupancy status does not exceed the preset occupancy rate threshold, it is determined that the system resources are sufficient.

[0023] In some embodiments, the judgment module is also used to compare the texture complexity and the edge density with a preset complexity threshold and a preset density threshold, respectively, to obtain a second comparison result; if the second comparison result is that the texture complexity and the edge density are both smaller than the complexity threshold and the preset density threshold, then it is determined that the scene complexity is low; if the second comparison result is that the texture complexity and the edge density are both greater than the complexity threshold and the preset density threshold, then it is determined that the scene complexity is high.

[0024] In some embodiments, the system resource occupancy status includes insufficient system resources and sufficient system resources; the scene complexity includes low scene complexity and high scene complexity; the multiple detection models to be selected include a lightweight YOLOv8 model and a high-precision landslide and rockfall detection model; the determination module is also used to determine that the initial detection model is the lightweight YOLOv8 model if the scene complexity is high or the system resources are insufficient; if the scene complexity is low and the system resources are sufficient, determine that the initial detection model is the high-precision landslide and rockfall detection model.

[0025] In some embodiments, the determination module is also used to determine the comprehensive score based on the weight coefficient, average accuracy and average inference time of the initial detection model; the comprehensive score is used to characterize the model performance of the initial detection model; the comprehensive score is compared with a preset score target value to obtain a third comparison result; if the third comparison result is that the comprehensive score is greater than the preset score target value, then it is determined that the model performance of the initial detection model is better, and the initial detection model is determined as the target detection model; if the third comparison result is that the comprehensive score is less than the preset score target value, then it is determined that the model performance of the initial detection model is poor, and based on the corresponding optimization strategy, the initial detection model is optimized and iterated to obtain the target detection model.

[0026] In some embodiments, the corresponding optimization strategies include: a detection accuracy optimization strategy and an inference time optimization strategy; the determination module is also used to analyze the key factors for the poor model performance of the initial detection model; if the key factor is insufficient detection accuracy, then based on the detection accuracy optimization strategy, the initial detection model is optimized and iterated until the corresponding comprehensive score of the detection model after the detection accuracy is optimized is greater than the preset score target value, the iteration is terminated, and the detection model after the detection accuracy is optimized is determined as the target detection model; the detection accuracy optimization strategy is to increase the model depth or width, adopt advanced data enhancement methods, and optimize model hyperparameters; if the key factor is that the inference time is too long, then based on the inference time optimization strategy, the initial detection model is optimized and iterated until the corresponding comprehensive score of the detection model after the inference time is optimized is greater than the preset score target value, the iteration is terminated, and the detection model after the inference time is optimized is determined as the target detection model; the inference time optimization strategy is to increase the model depth or width, adopt advanced data enhancement methods, and optimize model hyperparameters.

[0027] In some embodiments, the device also includes: a verification module, used to verify the performance of the initial detection model and obtain a verification result; if the verification result is a high false detection rate, the confidence threshold of the initial detection model is increased; if the verification result is a high missed detection rate, the confidence threshold is reduced; when the geology to be detected is falling rocks, the input resolution of the initial detection model is increased; when the geology to be detected is a landslide, the input resolution is reduced.

[0028] In some embodiments, the landslide and rockfall detection method is applied to a high-precision landslide and rockfall detection model, and the high-precision landslide and rockfall detection model is constructed in the following manner: adding an SE module after each residual block of the ResNet50 neural network model, adaptively assigning weights to each channel, emphasizing important features, and suppressing irrelevant features; constructing a CNN branch and a Transformer branch to extract local features and global features respectively; and fusing the features of the two branches in the channel dimension; gradually optimizing the bounding box by connecting cascade detectors in series; each level of the detector makes fine adjustments to the candidate box based on the previous level; using classification loss Focal Loss and bounding box regression loss GIoU Loss to enhance the learning of difficult-to-classify samples and bounding box regression, and obtain a high-precision landslide and rockfall detection model; wherein the first-level IoU threshold is 0.5, the second-level IoU threshold is 0.6, and the third-level IoU threshold is 0.7.

[0029] In a third aspect, an embodiment of the present invention provides an electronic device, comprising: a memory for storing executable instructions; and a processor for implementing the above-mentioned landslide and rockfall detection method when executing the executable instructions stored in the memory.

[0030] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium storing executable instructions for causing a processor to execute the executable instructions to implement the above-mentioned landslide and rockfall detection method.

[0031] The landslide rockfall detection method, device and equipment provided by the embodiment of the present invention first obtain the landslide rockfall image of the geological body to be detected and the system resource occupancy status; calculate the texture complexity and edge density of the landslide rockfall image data; judge the scene complexity according to the texture complexity and edge density; determine the initial detection model from multiple detection models to be selected according to the system resource occupancy status and scene complexity; determine the target detection model based on the comprehensive score of the initial detection model; perform landslide rockfall detection on the geological body to be detected based on the target detection model to obtain the target detection result. In this way, the present invention dynamically adjusts the model selection and parameter settings according to the system resource status and scene complexity, combining the lightweight model and the high-precision detection model, and achieves the best balance between real-time performance and detection accuracy; when resources are sufficient, the high-precision detection model proposed by the present invention is used to improve the landslide rockfall detection accuracy, give full play to the computing power of the system, and avoid resource waste; by designing a cyclic optimization framework based on time-precision comprehensive evaluation, the model can dynamically adjust the detection strategy according to the characteristics of different regions and data to improve applicability. On the other hand, the present invention achieves an effective balance between real-time performance and detection accuracy; the time-accuracy comprehensive evaluation index S is used to guide model optimization, thereby maximizing detection accuracy while meeting real-time requirements. The mean average precision (mAP) of the model detection results reaches 90.2%, and the average reasoning time for a single landslide or rockfall image is controlled within 50ms. The cyclic optimization method based on the time-accuracy comprehensive evaluation enables the model to achieve optimal performance in different regions and data conditions, providing a technical framework with wide applicability for geological disaster monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 is a structural schematic diagram of a landslide and rockfall detection system provided by an embodiment of the present invention;

[0033] Figure 2 It is a schematic diagram of a landslide and rockfall detection process provided by an embodiment of the present invention;

[0034] Figure 3 It is a flow chart of a landslide and rockfall detection method based on a time-precision adaptive cyclic evaluation model provided by an embodiment of the present invention;

[0035] Figure 4 is a schematic diagram of landslide and rockfall detection results provided by an embodiment of the present invention;

[0036] Figure 5 It is a schematic diagram of the structure of a landslide and rockfall detection device provided in an embodiment of the present invention;

[0037] Figure 6 It is a schematic diagram of the composition structure of the landslide and rockfall detection equipment provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0038] In order to make the purpose, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with the accompanying drawings. The described embodiments should not be regarded as limiting the present invention. All other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present invention.

[0039] In the following description, reference is made to "some embodiments", which describe a subset of all possible embodiments, but it is understood that "some embodiments" may be the same subset or different subsets of all possible embodiments, and may be combined with each other without conflict. Unless otherwise defined, all technical and scientific terms used in the embodiments of the present invention have the same meaning as those commonly understood by those skilled in the art to which the embodiments of the present invention pertain. The terms used in the embodiments of the present invention are only for the purpose of describing the embodiments of the present invention and are not intended to limit the present invention.

[0040] The following describes an exemplary application of the landslide and rockfall detection device of an embodiment of the present invention. The landslide and rockfall detection device provided by the embodiment of the present invention can be implemented as a terminal or a server. In one implementation, the landslide and rockfall detection device provided by the embodiment of the present invention can be implemented as various types of terminals such as laptop computers, tablet computers, desktop computers, mobile devices, etc.; in another implementation, the landslide and rockfall detection device provided by the embodiment of the present invention can also be implemented as a server, wherein the server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content distribution networks (CDN, Content Delivery Network), and big data and artificial intelligence platforms. The terminal and the server can be directly or indirectly connected via wired or wireless communication, which is not limited in the embodiment of the present invention. Below, an exemplary application of the landslide and rockfall detection device when it is implemented as a server will be described.

[0041] See also Figure 1 , Figure 11 is a schematic diagram of the structure of the landslide and rockfall detection system 10 provided in an embodiment of the present invention. In order to detect geological landslides and rockfalls, an embodiment of the present invention can provide a landslide and rockfall detection platform, which can be implemented as a landslide and rockfall detection application. The landslide and rockfall detection system 10 provided in an embodiment of the present invention includes a terminal 110, a network 120 and a server 130, wherein the server 130 is a server of the landslide and rockfall detection application. The server 130 can constitute the landslide and rockfall detection device of the embodiment of the present invention. The terminal 110 is connected to the server 130 via the network 120, and the network 120 can be a wide area network or a local area network, or a combination of the two.

[0042] In some embodiments, please refer to Figure 1 When performing landslide and rockfall detection, the terminal 110 sends the landslide and rockfall image and the system resource occupancy status to the server 130 through the network 120. The server 130 receives the landslide and rockfall image and the system resource occupancy status sent by the terminal 110, calculates the texture complexity and edge density of the landslide and rockfall image data; judges the scene complexity according to the texture complexity and edge density; determines the initial detection model from multiple detection models to be selected according to the system resource occupancy status and the scene complexity; determines the target detection model based on the comprehensive score of the initial detection model; performs landslide and rockfall detection on the geology to be detected based on the target detection model to obtain the target detection result. After obtaining the target detection result, the server 130 sends the target detection result to the terminal 110 through the network 120.

[0043] The target detection result method provided in the embodiment of the present invention can also be implemented based on a cloud platform and through cloud technology. For example, the server 130 can be a cloud server. A high-precision landslide and rockfall detection model is constructed through the cloud server to implement landslide and rockfall detection of the geological structure to be detected.

[0044] It should be noted here that cloud technology refers to a hosting technology that unifies hardware, software, network and other resources in a wide area network or local area network to achieve data computing, storage, processing and sharing. Cloud technology is a general term for network technology, information technology, integration technology, management platform technology, application technology, etc. based on the cloud computing business model. It can form a resource pool that can be used on demand and is flexible and convenient. Cloud computing technology will become an important support. The backend services of the technical network system require a large amount of computing and storage resources, such as video websites, picture websites and more portal websites. With the high development and application of the Internet industry, in the future, each item may have its own identification mark, and all need to be transmitted to the backend system for logical processing. Data of different levels will be processed separately. All kinds of industry data require strong system backing support, which can only be achieved through cloud computing.

[0045] The present invention provides a method for detecting landslide and rockfall. Figure 2 , Figure 2 is a flow chart of a landslide and rockfall detection method provided by an embodiment of the present invention, which is combined with Figure 2 The steps shown are explained.

[0046] Step S210, obtaining the landslide and rockfall image of the geological structure to be detected and the system resource occupancy status.

[0047] In some embodiments, landslide and rockfall images refer to images used to monitor and identify risks of natural disasters such as geological landslides and rockfalls. These images are usually obtained through drone aerial photography, satellite remote sensing, ground monitoring cameras, etc.

[0048] In some embodiments, the system resource occupancy status refers to the resource occupancy status of the current system, including but not limited to CPU usage, memory occupancy, GPU utilization, etc., to evaluate the processing capacity of the system.

[0049] Step S220, calculating the texture complexity and edge density of the landslide and rockfall image data.

[0050] In some embodiments, texture complexity refers to the richness and variation of texture information in an image, and is used to reflect the roughness, directionality, regularity, and other characteristics of the surface of a landslide or rockfall in an image. In practical applications, algorithms such as gray level co-occurrence matrix and local binary pattern can be used to calculate the texture complexity of an image. These algorithms can extract texture features in an image and quantify the complexity of the texture by counting and analyzing these features.

[0051] In some embodiments, edge density refers to the ratio of edge pixels in an image, reflecting the richness of the outline and detail information of an object in the image. In practical applications, an edge detection algorithm (such as a Canny edge detector) can be used to extract edges in an image, and the ratio of the number of edge pixels to the total number of pixels can be calculated to obtain edge density.

[0052] Step S230: determining scene complexity according to the texture complexity and the edge density.

[0053] In some embodiments, the calculated texture complexity and edge density are compared with a preset threshold or classification rule, and the scene complexity is classified into different levels (such as low complexity, medium complexity, high complexity, etc.) according to the comparison result. Of course, more complex judgment strategies can also be used, such as combining the weighted average or product of texture complexity and edge density to judge the scene complexity. This application does not make specific restrictions on this.

[0054] Step S240: determining an initial detection model from a plurality of detection models to be selected according to the system resource occupancy status and the scene complexity.

[0055] In the present invention, an initial detection model can be determined from multiple detection models to be selected according to the system resource occupancy status and scene complexity. This can be achieved through a predefined strategy, for example, selecting a high-precision model when resources are sufficient and the scene complexity is high, and selecting a lightweight model when resources are limited.

[0056] Step S250, determining a target detection model based on the comprehensive score of the initial detection model.

[0057] In some embodiments, the target detection model refers to a detection model with the best detection performance and the most suitable for the current application scenario. In the present invention, based on the comprehensive score of the initial detection model, the optimal target detection model is further determined.

[0058] Step S260: Based on the target detection model, landslide and rockfall detection is performed on the geology to be detected to obtain a target detection result.

[0059] In the present invention, a determined target detection model is used to detect landslides and rocksfalls on the geological conditions to be detected, and target detection results are obtained, which may include information such as the location, size, and quantity of landslides and rocksfalls, for subsequent early warning and response.

[0060] The landslide rockfall detection method, device and equipment provided by the embodiment of the present invention first obtain the landslide rockfall image of the geological body to be detected and the system resource occupancy status; calculate the texture complexity and edge density of the landslide rockfall image data; judge the scene complexity according to the texture complexity and edge density; determine the initial detection model from multiple detection models to be selected according to the system resource occupancy status and scene complexity; determine the target detection model based on the comprehensive score of the initial detection model; perform landslide rockfall detection on the geological body to be detected based on the target detection model to obtain the target detection result. In this way, the present invention dynamically adjusts the model selection and parameter settings according to the system resource status and scene complexity, combining the lightweight model and the high-precision detection model, and achieves the best balance between real-time performance and detection accuracy; when resources are sufficient, the high-precision detection model proposed by the present invention is used to improve the landslide rockfall detection accuracy, give full play to the computing power of the system, and avoid resource waste; by designing a cyclic optimization framework based on time-precision comprehensive evaluation, the model can dynamically adjust the detection strategy according to the characteristics of different regions and data to improve applicability. On the other hand, the present invention achieves an effective balance between real-time performance and detection accuracy; the time-accuracy comprehensive evaluation index S is used to guide model optimization, thereby maximizing detection accuracy while meeting real-time requirements. The mean average precision (mAP) of the model detection results reaches 90.2%, and the average reasoning time for a single landslide or rockfall image is controlled within 50ms. The cyclic optimization method based on the time-accuracy comprehensive evaluation enables the model to achieve optimal performance in different regions and data conditions, providing a technical framework with wide applicability for geological disaster monitoring.

[0061] In some embodiments, the above step S210 can be implemented by the following steps S211 to S213:

[0062] Step S211, comparing the system resource occupancy status with a preset occupancy rate threshold to obtain a first comparison result.

[0063] It should be noted that the preset occupancy rate threshold is determined based on system performance requirements, historical data statistical analysis or user customization.

[0064] Step S212: If the first comparison result is that the system resource occupancy status exceeds the preset occupancy rate threshold, it is determined that the system resources are insufficient.

[0065] In some embodiments, when it is determined that system resources are insufficient, corresponding processing measures may also be triggered, such as issuing an alarm, starting a resource optimization program, limiting the use of some functions, etc.

[0066] Step S213: If the first comparison result is that the system resource occupancy status does not exceed the preset occupancy rate threshold, it is determined that the system resources are sufficient.

[0067] In some embodiments, the above step S230 can be implemented by the following steps S231 to S233:

[0068] Step S231 : Compare the texture complexity and the edge density with a preset complexity threshold and a preset density threshold respectively to obtain a second comparison result.

[0069] Step S232: If the second comparison result is that both the texture complexity and the edge density are smaller than the complexity threshold and the preset density threshold, it is determined that the scene complexity is low.

[0070] Step S233: If the second comparison result is that both the texture complexity and the edge density are greater than the complexity threshold and the preset density threshold, it is determined that the scene complexity is high.

[0071] In some embodiments, the system resource occupancy state includes insufficient system resources and sufficient system resources; the scene complexity includes low scene complexity and high scene complexity; the multiple detection models to be selected include a lightweight YOLOv8 model and a high-precision landslide and rockfall detection model; the above step S240 can be implemented by the following steps S241 to S242:

[0072] Step S241: If the scene complexity is high or the system resources are insufficient, determine that the initial detection model is the lightweight YOLOv8 model.

[0073] Step S242: If the scene complexity is low and the system resources are sufficient, determine that the initial detection model is the high-precision landslide and rockfall detection model.

[0074] In this invention, by designing a cyclic optimization framework based on time-precision comprehensive evaluation, the model can dynamically adjust the detection strategy according to the characteristics of different regions and data, thereby improving applicability.

[0075] In the present invention, when resources are sufficient, the high-precision detection model proposed in the present invention is used to improve the accuracy of landslide and rockfall detection, give full play to the computing power of the system, and avoid waste of resources.

[0076] Based on the above embodiment, the above step S250 can be implemented by the following steps S251 to S254:

[0077] Step S251, determining the comprehensive score based on the weight coefficient, average accuracy and average inference time of the initial detection model; the comprehensive score is used to characterize the model performance of the initial detection model.

[0078] Step S252, comparing the comprehensive score with a preset score target value to obtain a third comparison result.

[0079] Step S253, if the third comparison result is that the comprehensive score is greater than the preset score target value, it is determined that the model performance of the initial detection model is better, and the initial detection model is determined as the target detection model.

[0080] Step S254, if the third comparison result is that the comprehensive score is less than the preset score target value, it is determined that the model performance of the initial detection model is poor, and based on the corresponding optimization strategy, the initial detection model is optimized and iterated to obtain the target detection model.

[0081] In some embodiments, the corresponding optimization strategies include: a detection accuracy optimization strategy and an inference time optimization strategy; the above step S254 can be implemented by the following steps S2541 to S2543:

[0082] Step S2541, analyzing the key factors of the poor model performance of the initial detection model

[0083] Step S2542, if the key factor is insufficient detection accuracy, then based on the detection accuracy optimization strategy, the initial detection model is optimized and iterated until the corresponding comprehensive score of the detection model after the detection accuracy optimization is greater than the preset score target value, and the iteration is ended, and the detection model after the detection accuracy optimization is determined as the target detection model; the detection accuracy optimization strategy is to increase the model depth or width, adopt advanced data enhancement methods, and optimize model hyperparameters.

[0084] Step S2543, if the key factor is that the inference time is too long, then based on the inference time optimization strategy, the initial detection model is optimized and iterated until the corresponding comprehensive score of the detection model after the inference time optimization is greater than the preset score target value, and the iteration is ended, and the detection model after the inference time optimization is determined as the target detection model; the inference time optimization strategy is to increase the model depth or width, adopt advanced data enhancement methods, and optimize model hyperparameters.

[0085] In the present invention, an adaptive mechanism is designed to dynamically adjust model selection and parameter settings according to system resource status and scene complexity by combining lightweight models and high-precision detection models to achieve the best balance between real-time performance and detection accuracy.

[0086] In some embodiments, the device also includes: a verification module, used to verify the performance of the initial detection model and obtain a verification result; if the verification result is a high false detection rate, the confidence threshold of the initial detection model is increased; if the verification result is a high missed detection rate, the confidence threshold is reduced; when the geology to be detected is falling rocks, the input resolution of the initial detection model is increased; when the geology to be detected is a landslide, the input resolution is reduced.

[0087] In the present invention, the core of the verification is the performance of the initial detection model, which is specifically reflected in the following two key indicators: (1) False detection rate: the frequency with which the detection model mistakenly detects non-landslide rockfall areas as landslides or rockfalls. (2) Missed detection rate: the frequency with which the detection model fails to identify landslides or rockfalls. The verification process is a statistical analysis of the model detection results: comparing the predicted results with the true labels, and calculating the false detection rate and missed detection rate.

[0088] Confidence threshold adjustment under verification results: (1) Raise the confidence threshold when the false positive rate is high: The purpose is to reduce the number of falsely detected objects. Raising the threshold means that the model will be more stringent and only output results with confidence higher than the new threshold, filtering out potential false positives.

[0089] (2) Lowering the confidence threshold when the missed detection rate is high: The purpose is to increase the number of targets identified and reduce the possibility of missed detection. Lowering the threshold means that the model will output more low-confidence targets and minimize the number of unidentified real targets.

[0090] In some embodiments, the landslide and rockfall detection method is applied to a high-precision landslide and rockfall detection model, and the high-precision landslide and rockfall detection model is constructed in the following manner:

[0091] Firstly, an SE module is added after each residual block of the ResNet50 neural network model to adaptively assign weights to each channel, emphasize important features, and suppress irrelevant features. Secondly, a CNN branch and a Transformer branch are constructed to extract local features and global features respectively. The features of the two branches are fused in the channel dimension. Finally, the bounding box is gradually optimized by connecting cascade detectors in series. Each level of the detector makes fine adjustments to the candidate box based on the previous level. The classification loss Focal Loss and the bounding box regression loss GIoU Loss are used to enhance the learning of difficult-to-classify samples and bounding box regression, and a high-precision landslide and rockfall detection model is obtained. The IoU threshold of the first level, the IoU threshold of the second level, and the IoU threshold of the third level are 0.7.

[0092] In the present invention, by introducing the attention mechanism, dual-branch network and cascade detector, the model's adaptability to complex environments (such as illumination changes, target size differences, etc.) is improved, and the stability and accuracy of detection are enhanced.

[0093] The following describes an exemplary application of an embodiment of the present invention in a practical application scenario.

[0094] As attached Figure 3 As shown, the technical solution process of the present invention is as follows: first, data is collected, annotated and enhanced to construct a high-quality landslide and rockfall image dataset; second, an adaptive mechanism framework is designed to dynamically adjust model selection and parameter settings according to system resource status and scene complexity to achieve the best balance between real-time performance and detection accuracy; then, for the model selection part in the adaptive mechanism, a high-precision landslide and rockfall detection model is designed and implemented, including an attention mechanism, a dual-branch fusion model and a cascade detector to improve detection accuracy; finally, based on a cyclic optimization method based on time-precision comprehensive evaluation, the model is optimized and iterated to ensure that the model can achieve optimal performance under different resource conditions and scene complexities.

[0095] 1. Data Collection and Enhancement

[0096] In order to build a high-quality training dataset, a large amount of image data containing landslides and rockfalls was collected, covering different terrains, weather and lighting conditions to ensure data diversity and comprehensiveness. Professional image annotation tools were used to accurately annotate the images, defining category 0 as rockfalls and category 1 as landslides.

[0097] In order to improve the generalization ability of the model, the training data was enhanced by various data enhancements, including geometric transformation (random rotation, translation, scaling, flipping), color adjustment (random adjustment of brightness, contrast, saturation), noise addition (Gaussian noise, blurring), etc. These enhancement methods enrich the diversity of training data and improve the robustness of the model in complex environments.

[0098] 2. Design of Adaptive Mechanism Framework

[0099] In order to achieve the best balance between real-time performance and detection accuracy, an adaptive mechanism framework is designed to dynamically adjust model selection and parameter settings according to system resource status and scenario complexity.

[0100] First, the texture complexity and edge density of the input image are calculated to determine the complexity of the scene, and resource information such as CPU / GPU utilization and memory usage is monitored in real time; second, when resources are tight or the scene is complex, the lightweight model YOLOv8 is selected to ensure real-time performance; when resources are sufficient and the scene is simple, a high-precision landslide and rockfall detection model is selected to improve detection accuracy; finally, according to the detection results, the confidence threshold and input resolution are dynamically adjusted. When the false detection rate is high, the threshold is increased; when the missed detection rate is high, the threshold is reduced. For small targets (such as rockfall), the input resolution is increased; for large targets (such as landslides), the input resolution is reduced to reduce the amount of calculation.

[0101] Criteria for judging whether resources are tight or not: When the CPU / GPU utilization rate exceeds 80% or the available memory is less than 1GB, resources are considered tight; when the CPU / GPU utilization rate is less than 80% and the available memory is more than 1GB, resources are considered sufficient.

[0102] Judgment criteria for scene complexity: The scene analysis module calculates the texture complexity and edge density of the input image. When the texture complexity and edge density are lower than the preset threshold, the scene is considered to be simple; when the texture complexity and edge density are higher than the preset threshold, the scene is considered to be complex.

[0103] 3. Design and implementation of a high-precision landslide and rockfall detection model

[0104] In the adaptive mechanism framework, in order to solve the problem of improving detection accuracy when resources are sufficient, an innovative combination model is designed and implemented, including:

[0105] 1) Attention mechanism fusion: Add a SE module after each residual block of ResNet50 to emphasize important features and suppress irrelevant features by adaptively assigning weights to each channel.

[0106] 2) Dual-branch fusion model: Construct a CNN branch and a Transformer branch to extract local and global features respectively. The features of the two branches are fused in the channel dimension to improve the detection ability of objects of different scales.

[0107] 3) Cascade R-CNN: Cascade R-CNN uses a cascade detector to gradually optimize the bounding box and improve positioning accuracy by connecting multiple detectors in series. The IoU threshold of the first level is 0.5, the IoU threshold of the second level is 0.6, and the IoU threshold of the third level is 0.7. Each level of the detector further refines the candidate box based on the previous level. The classification loss FocalLoss and the bounding box regression loss GIoU Loss are used to enhance the learning of difficult-to-classify samples and bounding box regression.

[0108] 4. Loop Optimization Based on Time-Accuracy Comprehensive Evaluation

[0109] In order to ensure that the model can achieve optimal performance under different regions and data conditions, a cyclic optimization method based on time-accuracy comprehensive evaluation is proposed, which needs to be adjusted according to the specific data set and application scenario. The specific optimization process is as follows:

[0110] 1) Using the initial training model, calculate the comprehensive score and record the performance:

[0111] S = α × mAP - β × T inf ,

[0112] Among them, α and β are weight coefficients, mAP is the average precision, T inf is the average inference time.

[0113] 2) Analyze performance bottlenecks to determine whether the low overall score is caused by insufficient detection accuracy or long inference time.

[0114] 3) Take corresponding optimization measures according to performance bottlenecks. If the detection accuracy is insufficient, you can increase the network depth or width, use advanced data enhancement methods, or optimize hyperparameters to improve the model's learning ability; if the inference time is too long, you can reduce the number of model parameters and calculations through pruning, quantization, and other methods.

[0115] 4) Retrain or fine-tune the model and implement optimization strategies.

[0116] 5) Calculate the comprehensive score S again and compare it with the previous score.

[0117] 6) If the comprehensive score S is significantly improved and has not reached the preset target, return to step 2 and continue optimization; otherwise, end the cycle and determine the final model.

[0118] The landslide and rockfall detection results of the present invention are shown in the attached Figure 4 As shown in the figure, the present invention achieves an effective balance between real-time performance and detection accuracy. The time-accuracy comprehensive evaluation index S is used to guide model optimization, which achieves maximum detection accuracy while meeting real-time requirements. The mean average precision (mAP) of the model detection results reaches 90.2%, and the average inference time of a single landslide or rockfall image is controlled within 50ms.

[0119] In addition, if Figure 3 As shown, according to general experience:

[0120] 1. Criteria for judging whether resources are tight or not: When the CPU / GPU utilization rate exceeds 80% or the available memory is less than 1GB, resources are considered tight; when the CPU / GPU utilization rate is less than 80% and the available memory is more than 1GB, resources are considered sufficient.

[0121] 2. Judgment criteria for scene complexity: The scene analysis module calculates the texture complexity and edge density of the input image. When the texture complexity and edge density are lower than the preset threshold, the scene is considered simple; when the texture complexity and edge density are higher than the preset threshold, the scene is considered complex. The preset threshold needs to be set according to the actual situation on site.

[0122] 3. The setting standards of mAP and Tinf need to be set according to the actual situation. The higher the accuracy standard is set and the shorter the inference time is set, the more times the model needs to be iterated.

[0123] The cyclic optimization method based on comprehensive time-precision evaluation enables the model to achieve optimal performance under different regions and data conditions, providing a technical framework with wide applicability for geological disaster monitoring.

[0124] Figure 5 FIG. 1 is a schematic diagram of the structure of a landslide and rockfall detection device provided in an embodiment of the present invention. Figure 5 As shown, the landslide and rockfall detection device 500 includes: an acquisition module 501, which is used to obtain the landslide and rockfall image of the geological body to be monitored and the system resource occupancy status; a calculation module 502, which is used to calculate the texture complexity and edge density of the landslide and rockfall image data; a judgment module 503, which is used to judge the scene complexity according to the texture complexity and the edge density; a determination module 504, which is used to determine the initial detection model from multiple detection models to be selected according to the system resource occupancy status and the scene complexity; the determination module 504 is also used to determine the target detection model based on the comprehensive score of the initial detection model; a detection module 505, which is used to perform landslide and rockfall detection on the geological body to be monitored based on the target detection model to obtain a target detection result.

[0125] In some embodiments, the acquisition module 501 is also used to compare the system resource occupancy status with a preset occupancy rate threshold to obtain a first comparison result; if the first comparison result is that the system resource occupancy status exceeds the preset occupancy rate threshold, it is determined that the system resources are insufficient; if the first comparison result is that the system resource occupancy status does not exceed the preset occupancy rate threshold, it is determined that the system resources are sufficient.

[0126] In some embodiments, the judgment module 503 is also used to compare the texture complexity and the edge density with a preset complexity threshold and a preset density threshold, respectively, to obtain a second comparison result; if the second comparison result is that the texture complexity and the edge density are both smaller than the complexity threshold and the preset density threshold, then it is determined that the scene complexity is low; if the second comparison result is that the texture complexity and the edge density are both greater than the complexity threshold and the preset density threshold, then it is determined that the scene complexity is high.

[0127] In some embodiments, the system resource occupancy status includes insufficient system resources and sufficient system resources; the scene complexity includes low scene complexity and high scene complexity; the multiple detection models to be selected include a lightweight YOLOv8 model and a high-precision landslide and rockfall detection model; the determination module 504 is also used to determine that the initial detection model is the lightweight YOLOv8 model if the scene complexity is high or the system resources are insufficient; if the scene complexity is low and the system resources are sufficient, determine that the initial detection model is the high-precision landslide and rockfall detection model.

[0128] In some embodiments, the determination module 504 is also used to determine the comprehensive score based on the weight coefficient, average accuracy and average inference time of the initial detection model; the comprehensive score is used to characterize the model performance of the initial detection model; the comprehensive score is compared with a preset score target value to obtain a third comparison result; if the third comparison result is that the comprehensive score is greater than the preset score target value, then it is determined that the model performance of the initial detection model is better, and the initial detection model is determined as the target detection model; if the third comparison result is that the comprehensive score is less than the preset score target value, then it is determined that the model performance of the initial detection model is poor, and based on the corresponding optimization strategy, the initial detection model is optimized and iterated to obtain the target detection model.

[0129] In some embodiments, the corresponding optimization strategies include: a detection accuracy optimization strategy and an inference time optimization strategy; the determination module 504 is also used to analyze the key factors for the poor model performance of the initial detection model; if the key factor is insufficient detection accuracy, then based on the detection accuracy optimization strategy, the initial detection model is optimized and iterated until the corresponding comprehensive score of the detection model after the detection accuracy is optimized is greater than the preset score target value, the iteration is terminated, and the detection model after the detection accuracy is optimized is determined as the target detection model; the detection accuracy optimization strategy is to increase the model depth or width, adopt advanced data enhancement methods, and optimize model hyperparameters; if the key factor is that the inference time is too long, then based on the inference time optimization strategy, the initial detection model is optimized and iterated until the corresponding comprehensive score of the detection model after the inference time is optimized is greater than the preset score target value, the iteration is terminated, and the detection model after the inference time is optimized is determined as the target detection model; the inference time optimization strategy is to increase the model depth or width, adopt advanced data enhancement methods, and optimize model hyperparameters.

[0130] In some embodiments, the device also includes: a verification module, used to verify the performance of the initial detection model and obtain a verification result; if the verification result is a high false detection rate, the confidence threshold of the initial detection model is increased; if the verification result is a high missed detection rate, the confidence threshold is reduced; when the geology to be detected is falling rocks, the input resolution of the initial detection model is increased; when the geology to be detected is a landslide, the input resolution is reduced.

[0131] In some embodiments, the landslide and rockfall detection method is applied to a high-precision landslide and rockfall detection model, and the high-precision landslide and rockfall detection model is constructed in the following manner: adding an SE module after each residual block of the ResNet50 neural network model, adaptively assigning weights to each channel, emphasizing important features, and suppressing irrelevant features; constructing a CNN branch and a Transformer branch to extract local features and global features respectively; and fusing the features of the two branches in the channel dimension; gradually optimizing the bounding box by connecting cascade detectors in series; each level of the detector makes fine adjustments to the candidate box based on the previous level; using classification loss Focal Loss and bounding box regression loss GIoU Loss to enhance the learning of difficult-to-classify samples and bounding box regression, and obtain a high-precision landslide and rockfall detection model; wherein the first-level IoU threshold is 0.5, the second-level IoU threshold is 0.6, and the third-level IoU threshold is 0.7.

[0132] It should be noted that the description of the device of the embodiment of the present invention is similar to the description of the above method embodiment, and has similar beneficial effects as the same method embodiment, so it will not be repeated. For technical details not disclosed in the embodiment of the device, please refer to the description of the method embodiment of the present invention for understanding.

[0133] It should be noted that, in the embodiment of the present invention, if the above-mentioned landslide and rockfall detection method is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the embodiment of the present invention, or the part that contributes to the relevant technology, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for enabling a terminal to execute all or part of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a magnetic disk or an optical disk. In this way, the embodiment of the present invention is not limited to any specific combination of hardware and software.

[0134] Correspondingly, an embodiment of the present invention provides a landslide and rockfall detection device, Figure 6FIG. 1 is a schematic diagram of the structure of a landslide and rockfall detection device provided in an embodiment of the present invention. Figure 6 As shown, the landslide and rockfall detection device 600 at least includes: a processor 601 and a computer-readable storage medium 602 configured to store executable instructions, wherein the processor 601 generally controls the overall operation of the landslide and rockfall detection device. The computer-readable storage medium 602 is configured to store instructions and applications executable by the processor 601, and can also cache data to be processed or processed by each module in the processor 601 and the landslide and rockfall detection device 600, which can be implemented by flash memory (FLASH) or random access memory (RAM, Random Access Memory).

[0135] An embodiment of the present invention provides a storage medium storing executable instructions, wherein the executable instructions are stored. When the executable instructions are executed by a processor, the processor will be caused to execute the method provided by the embodiment of the present invention, for example, Figure 2 The method shown.

[0136] In some embodiments, the storage medium can be a computer-readable storage medium, for example, a ferroelectric random access memory (FRAM), a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a flash memory, a magnetic surface memory, an optical disk, or a compact disk read-only memory (CD-ROM), etc.; it can also be various devices including one or any combination of the above memories.

[0137] In some embodiments, executable instructions may be in the form of a program, software, software module, script or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and may be deployed in any form, including as a stand-alone program or as a module, component, subroutine or other unit suitable for use in a computing environment.

[0138] As an example, executable instructions may, but need not necessarily, correspond to a file in a file system, may be stored as part of a file storing other programs or data, for example, in one or more scripts in a Hyper Text Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple coordinated files (e.g., files storing one or more modules, subroutines, or code portions). As an example, executable instructions may be deployed to be executed on one electronic device, or on multiple electronic devices located at one location, or on multiple electronic devices distributed at multiple locations and interconnected by a communication network.

[0139] The above description is only an embodiment of the present invention and is not intended to limit the protection scope of the present invention. Any modification, equivalent replacement and improvement made within the spirit and scope of the present invention are included in the protection scope of the present invention.

[0140] It should be understood that "one embodiment" or "an embodiment" mentioned throughout the specification means that specific features, structures or characteristics related to the embodiment are included in at least one embodiment of the present invention. Therefore, "in one embodiment" or "in an embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. In addition, these specific features, structures or characteristics can be combined in one or more embodiments in any suitable manner. It should be understood that in various embodiments of the present invention, the size of the serial number of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention. The serial numbers of the above-mentioned embodiments of the present invention are only for description and do not represent the advantages and disadvantages of the embodiments.

[0141] It should be noted that, in this article, the terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method or device. In the absence of further restrictions, an element defined by the statement "comprises one..." does not exclude the presence of other identical elements in the process, method, article or device including the element. In several embodiments provided by the present invention, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored, or not executed.

[0142] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A method for detecting landslide and rockfall, characterized in that: The method comprises: Obtain the landslide and rockfall images of the geological structure to be detected and the system resource occupancy status; Calculating the texture complexity and edge density of the landslide rockfall image data; Determining scene complexity according to the texture complexity and the edge density; Determining an initial detection model from a plurality of detection models to be selected according to the system resource occupancy status and the scene complexity; Determining a target detection model based on the comprehensive score of the initial detection model; Based on the target detection model, landslide and rockfall detection is performed on the geology to be detected to obtain a target detection result.

2. The method according to claim 1, characterized in that Get the system resource usage status, including: Comparing the system resource occupancy state with a preset occupancy rate threshold to obtain a first comparison result; If the first comparison result is that the system resource occupancy state exceeds the preset occupancy rate threshold, it is determined that the system resources are insufficient; If the first comparison result is that the system resource occupancy status does not exceed the preset occupancy rate threshold, it is determined that the system resources are sufficient.

3. The method according to claim 1, characterized in that The determining the scene complexity according to the texture complexity and the edge density includes: Respectively comparing the texture complexity and the edge density with a preset complexity threshold and a preset density threshold to obtain a second comparison result; If the second comparison result is that both the texture complexity and the edge density are smaller than the complexity threshold and the preset density threshold, it is determined that the scene complexity is low; If the second comparison result is that both the texture complexity and the edge density are greater than the complexity threshold and the preset density threshold, it is determined that the scene complexity is high.

4. The method according to claim 1, characterized in that: The system resource occupancy state includes insufficient system resources and sufficient system resources; the scene complexity includes low scene complexity and high scene complexity; the multiple detection models to be selected include a lightweight YOLOv8 model and a high-precision landslide and rockfall detection model; The determining of the initial detection model from a plurality of detection models to be selected according to the system resource occupancy state and the scene complexity includes: If the scene complexity is high or the system resources are insufficient, determining that the initial detection model is the lightweight YOLOv8 model; If the scene complexity is low and the system resources are sufficient, the initial detection model is determined to be the high-precision landslide and rockfall detection model.

5. The method according to claim 1, characterized in that The step of determining the target detection model based on the comprehensive score of the initial detection model comprises: Determine the comprehensive score based on the weight coefficient, average accuracy and average inference time of the initial detection model; the comprehensive score is used to characterize the model performance of the initial detection model; Comparing the comprehensive score with a preset score target value to obtain a third comparison result; If the third comparison result is that the comprehensive score is greater than the preset score target value, it is determined that the model performance of the initial detection model is better, and the initial detection model is determined as the target detection model; If the third comparison result is that the comprehensive score is less than the preset score target value, it is determined that the model performance of the initial detection model is poor, and based on the corresponding optimization strategy, the initial detection model is optimized and iterated to obtain the target detection model.

6. The method according to claim 5, characterized in that The corresponding optimization strategies include: detection accuracy optimization strategy and reasoning time optimization strategy; If the third comparison result is that the comprehensive score is less than the preset score target value, it is determined that the model performance of the initial detection model is poor, and based on the corresponding optimization strategy, the initial detection model is optimized and iterated to obtain the target detection model, including: Analyze key factors that cause poor model performance of the initial detection model; If the key factor is insufficient detection accuracy, then based on the detection accuracy optimization strategy, the initial detection model is optimized and iterated until the corresponding comprehensive score of the detection model after the detection accuracy optimization is greater than the preset score target value, and the iteration is terminated, and the detection model after the detection accuracy optimization is determined as the target detection model; the detection accuracy optimization strategy is to increase the model depth or width, adopt advanced data enhancement methods, and optimize model hyperparameters; If the key factor is that the inference time is too long, the initial detection model is optimized and iterated based on the inference time optimization strategy until the corresponding comprehensive score of the detection model after the inference time optimization is greater than the preset score target value, and the iteration is ended, and the detection model after the inference time optimization is determined as the target detection model; the inference time optimization strategy is to increase the depth or width of the model, adopt advanced data enhancement methods, and optimize model hyperparameters.

7. The method according to claim 1, characterized in that The method further comprises: Verifying the performance of the initial detection model to obtain a verification result; If the verification result shows that the false positive rate is high, increasing the confidence threshold of the initial detection model; If the verification result shows that the missed detection rate is high, lowering the confidence threshold; When the geology to be detected is rockfall, the input resolution of the initial detection model is increased; when the geology to be detected is landslide, the input resolution is reduced.

8. The method according to claim 1, characterized in that The landslide and rockfall detection method is applied to a high-precision landslide and rockfall detection model, and the high-precision landslide and rockfall detection model is constructed in the following way: Add a SE module after each residual block of the ResNet50 neural network model to adaptively assign weights to each channel, emphasize important features, and suppress irrelevant features; Construct a CNN branch and a Transformer branch to extract local features and global features respectively; and fuse the features of the two branches in the channel dimension; By connecting cascade detectors in series, the bounding box is gradually optimized; each level of detector makes fine adjustments to the candidate box based on the previous level; The classification loss Focal Loss and bounding box regression loss GIoU Loss are used to enhance the learning of difficult-to-classify samples and bounding box regression, and a high-precision landslide and rockfall detection model is obtained; among them, the first-level IoU threshold is 0.5, the second-level IoU threshold is 0.6, and the third-level IoU threshold is 0.

7.

9. A landslide and rockfall detection device, characterized in that: The device comprises: An acquisition module is used to obtain the landslide and rockfall images of the geological objects to be monitored and the system resource occupancy status; A calculation module, used for calculating the texture complexity and edge density of the landslide and rockfall image data; A judging module, used for judging scene complexity according to the texture complexity and the edge density; A determination module, used to determine an initial detection model from a plurality of detection models to be selected according to the system resource occupancy status and the scene complexity; The determination module is further used to determine the target detection model based on the comprehensive score of the initial detection model; The detection module is used to perform landslide and rockfall detection on the geology to be monitored based on the target detection model to obtain a target detection result.

10. An electronic device, characterized in that: include: A memory for storing executable instructions; The processor is used to implement the landslide and rockfall detection method as described in any one of claims 1 to 8 when executing the executable instructions stored in the memory.

Images