An unmanned aerial vehicle based emergency rescue equipment and system

Abstract

The application provides an unmanned aerial vehicle-based emergency rescue equipment and system, and relates to the technical field of emergency rescue.The emergency rescue equipment comprises an unmanned aerial vehicle carrier, an image recognition module and an image acquisition device arranged on the unmanned aerial vehicle carrier; the image recognition module comprises an initial feature processing module, a multi-stage lightweight feature extraction module, a global feature refining module and a classification decision module, and is used for identifying the current disaster type according to the image acquired by the image acquisition device.The unmanned aerial vehicle-based emergency rescue equipment can improve the accuracy of acquired disaster information.

Description

Technical Field

[0001] This application relates to the field of emergency rescue technology, specifically to an emergency rescue equipment and system based on unmanned aerial vehicles (UAVs). Background Technology

[0002] In the field of emergency rescue for natural disasters and sudden accidents, timely and accurate acquisition of disaster information is a crucial prerequisite for carrying out effective rescue operations. However, traditional manual surveys and fixed monitoring equipment face many limitations in practical applications, severely restricting the accuracy and timeliness of disaster assessment.

[0003] Currently, the mainstream methods for obtaining disaster information mainly rely on on-site investigations by ground personnel and monitoring by fixed video cameras. These methods reveal significant shortcomings in complex and ever-changing disaster environments: firstly, disaster sites often involve traffic disruptions and hazardous environments, making it difficult for rescue personnel to reach the core areas in a timely manner, leading to delays in disaster information acquisition; secondly, fixed monitoring equipment, limited by its installation location and viewing angle, cannot comprehensively cover the disaster area, especially in large-scale disaster scenarios, failing to provide a holistic disaster assessment. Furthermore, conventional monitoring methods also have significant shortcomings in terms of the real-time nature and mobility of data collection, making it difficult to provide dynamically updated on-site information for command and decision-making.

[0004] While existing drone monitoring solutions have improved the mobility of information acquisition to some extent, they still have significant shortcomings in the intelligence level of disaster identification. Most systems can only achieve simple image acquisition and transmission, lacking the ability to deeply analyze disaster characteristics. This forces the command center to rely on manual interpretation, which is not only inefficient but also susceptible to subjective factors affecting accuracy. Especially in disaster sites with low visibility and complex environments, conventional image processing methods often fail to effectively extract key disaster features, leading to information identification errors or omissions. In summary, existing technologies suffer from the technical problem of inaccurately acquiring disaster information. Summary of the Invention

[0005] Firstly, this invention provides an emergency rescue equipment based on unmanned aerial vehicles (UAVs). Its image recognition module can capture key detailed features in disaster scenarios through hierarchical stacking processing, eliminate redundant information through spatial compression, strengthen the correlation between features through semantic fusion, and finally establish a high-precision association between abstract features and disaster types. Through a three-level progressive mechanism of initial feature processing, multi-stage feature deepening, and global feature optimization, the generated global feature vector can comprehensively represent the typical features of disasters. It can identify both obvious disaster features such as floods and fires, as well as subtle changes such as landslides and building cracks, thereby improving the accuracy of disaster information acquisition.

[0006] The emergency rescue equipment includes an unmanned aerial vehicle (UAV) carrier and an image acquisition device mounted on the UAV carrier. The image acquisition device is equipped with an image recognition module, which includes an initial feature processing module, a multi-stage lightweight feature extraction module, a global feature refinement module, and a classification decision module. This module is used to identify the current disaster type based on the images acquired by the image acquisition device. The process by which the image recognition module identifies the current disaster type based on the original images acquired by the image acquisition device includes: performing preliminary feature extraction on the images using the initial feature processing module to obtain a basic feature map; performing multi-stage stacking processing on the basic feature map using the multi-stage lightweight feature extraction module to obtain a high-level feature map; performing spatial compression and semantic fusion on the high-level feature map using the global feature refinement module to obtain a global feature vector; and mapping the global feature vector to a prediction result for the disaster category using the classification decision module.

[0007] According to one embodiment of the present invention, the initial feature processing module includes an initial feature transformation unit and a spatial response enhancement unit; the initial feature transformation unit is used to extract local structural features from the original image, control the feature size reduction ratio of the local structural features, and output an intermediate feature map that retains the local structural features; the spatial response enhancement unit is used to perform spatial enhancement on the local structural features in the intermediate feature map through a depthwise convolution operation to obtain the basic feature map.

[0008] According to one embodiment of the present invention, the multi-stage lightweight feature extraction module includes multiple stage structural units, each stage structural unit including a lightweight feature construction submodule and a feature compression and fusion submodule; the lightweight feature construction submodule is composed of multiple structurally consistent residual enhancement lightweight feature construction units stacked together, each residual enhancement lightweight feature construction unit being used to perform local spatial awareness, in-channel dynamic modeling, and short-circuiting operations; in the lightweight feature construction submodule, the stacking depth and number of channels of the residual enhancement lightweight feature construction units are configured according to the stage of feature extraction; the feature compression and fusion unit is used to splice and fuse intermediate features within the current stage and perform spatial downsampling to obtain the high-level feature map.

[0009] According to one embodiment of the present invention, the emergency rescue equipment further includes a point cloud acquisition module disposed on the UAV carrier, the point cloud acquisition module being used to acquire point cloud data at the disaster site.

[0010] According to one embodiment of the present invention, the unmanned aerial vehicle carrier is a drone using a high-lift three-bladed propeller or a racing drone using a four-in-one ESC system.

[0011] According to one embodiment of the present invention, the image acquisition device is a visible light pod, an infrared pod, or a dual-light pod.

[0012] According to one embodiment of the present invention, the emergency rescue equipment further includes an active power supply module disposed on the UAV carrier. The active power supply module includes an airborne power step-down module and an optoelectronic integrated cable device. The airborne power step-down module is used to convert high-voltage DC voltage into the operating voltage required by the UAV carrier. The optoelectronic integrated cable device includes an optoelectronic integrated cable and a cable winch. The optoelectronic integrated cable is used to transmit current and communication signals. The cable winch is used to manage the cable winding and unwinding. The cable winch integrates an optoelectronic conversion module for connecting to a ground workstation to send the airborne data of the UAV carrier to the ground workstation.

[0013] According to one embodiment of the present invention, the emergency rescue equipment further includes an audible and visual warning device, which is used to send warning information.

[0014] Secondly, the present invention also provides an emergency rescue system based on unmanned aerial vehicles (UAVs), including the emergency rescue equipment described in the above embodiments; the emergency rescue system further includes a ground workstation, an emergency rescue command vehicle, and ground power supply equipment; wherein, the ground workstation is communicatively connected to the emergency rescue equipment and is used to receive carrier data from the UAV carrier and process the carrier data; the emergency rescue command vehicle is communicatively connected to the ground workstation and is used to maintain data interaction with the ground workstation; the ground power supply equipment is connected to the UAV carrier through an optoelectronic integrated cable device and is used to supply power to the UAV carrier.

[0015] According to one embodiment of the present invention, the emergency rescue command vehicle integrates one or more of the following: an integrated power supply system, a video matrix, a video conferencing system, an audio frequency divider, a network system, and meteorological equipment; wherein, the power supply system is used to supply power to the emergency rescue command vehicle; the video matrix is ​​used to send multiple video or image output signals to different video receiving terminals and to switch the interconnection relationship of video sources to the video receiving terminals; the video conferencing system is used to establish video conferences; the audio frequency divider is used for multi-channel audio management and distribution; the network system is used to provide a network environment; and the meteorological equipment includes a six-element meteorological instrument, a visibility meteorological instrument, a road surface temperature collector, and a road surface water film thickness measuring instrument, used to acquire multi-source meteorological data.

[0016] Compared with existing technologies, the beneficial effects of this application are as follows: by capturing key detailed features in disaster scenes through hierarchical stacking of image recognition modules, redundant information is eliminated through spatial compression, and the correlation between features is strengthened by semantic fusion. Finally, a high-precision association is established between abstract features and disaster types. Through a three-level progressive mechanism of initial feature processing, multi-stage feature deepening, and global feature optimization, the generated global feature vector can comprehensively represent the typical features of disasters. It can identify both obvious disaster features such as floods and fires, as well as subtle changes such as landslides and building cracks, thereby improving the accuracy of disaster information acquisition. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the structure of an emergency rescue equipment based on a drone, provided in an embodiment of this application.

[0018] Figure 2 This is a schematic diagram of the network structure of the image recognition module provided in an embodiment of this application.

[0019] Figure 3 This is a schematic diagram of the network structure of the initial feature processing module provided in the embodiments of this application.

[0020] Figure 4 A schematic diagram of the network structure of the multi-stage lightweight feature extraction module provided in the embodiments of this application.

[0021] Figure 5 This is a schematic diagram of the structure of the global feature refinement module provided in an embodiment of this application.

[0022] Figure 6 This is a schematic diagram of the connection of an emergency rescue system provided in an embodiment of this application.

[0023] Figure 7 A schematic diagram of video streaming of emergency rescue equipment provided in the embodiments of this application.

[0024] Figure 8 This is a schematic diagram illustrating the operation of ground power supply equipment providing uninterrupted power to emergency rescue equipment, as provided in an embodiment of this application. Detailed Implementation

[0025] The present application will now be described in further detail with reference to experimental examples and specific embodiments. However, this should not be construed as limiting the scope of the subject matter of the present application to the following embodiments. All technologies implemented based on the content of the present application fall within the scope of protection of the present application.

[0026] Unless otherwise specified, the terms "upper," "lower," "left," "right," "center," "inner," "outer," and "side" used in the description of specific embodiments of this application to indicate orientation or positional relationships are based on the orientation or positional relationships shown in the accompanying drawings, or the orientation or positional relationship in which the product / equipment / device is usually placed during use. These terms are merely for the purpose of facilitating the description of the solution in this application or simplifying the description in specific embodiments, so as to enable those skilled in the art to quickly understand the solution, and do not indicate or imply that a particular device / component / element must have a specific orientation, or be constructed and operated in a specific positional relationship. Therefore, they should not be construed as limitations on this application.

[0027] In the description of the embodiments of this application, technical terms such as "first" and "second" only distinguish one entity or operation from another, and should not be construed as indicating or implying relative importance or implicitly specifying the number, specific order, or primary or secondary relationship of the indicated technical features. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly defined.

[0028] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0029] Please refer to Figure 1 , Figure 1 This is a schematic diagram of the structure of an unmanned aerial vehicle (UAV)-based emergency rescue equipment provided in an embodiment of this application. The UAV-based emergency rescue equipment 10 may include a UAV carrier 11, an image acquisition device 12 mounted on the UAV carrier 11, and an image recognition module 13. The UAV carrier 11 is the UAV itself, and the image recognition module 13 may be mounted on the UAV carrier 11 or in a server at another end.

[0030] Please refer to Figure 2 , Figure 2This is a schematic diagram of the network structure of the image recognition module provided in this embodiment. The image recognition module 13 includes an initial feature processing module 131, a multi-stage lightweight feature extraction module 132, a global feature refinement module 133, and a classification decision module 134, used to identify the current disaster type based on the images acquired by the image acquisition device 12. In this embodiment, the image recognition module 13 can classify disaster images captured by UAV aerial photography and transmit the detected information to the command center in real time in text form in environments with poor network signals. This program uses a combination of multiple structural modules such as depthwise convolution, grouped pointwise convolution, dense connections, and global response normalization (GRN) to reduce redundant network calculations. It requires fewer parameters and has low computing power, featuring lightweight, high efficiency, low power consumption, and real-time deployment capabilities. It is suitable for high-precision real-time image classification in embedded environments and UAV fields.

[0031] The process by which the image recognition module identifies the current disaster type based on the original image acquired by the image acquisition device includes: performing preliminary feature extraction on the image using the initial feature processing module to obtain a basic feature map; performing multi-stage stacking processing on the basic feature map using the multi-stage lightweight feature extraction module to obtain a high-level feature map; performing spatial compression and semantic fusion on the high-level feature map using the global feature refinement module to obtain a global feature vector; and mapping the global feature vector to a prediction result for the disaster category using the classification decision module.

[0032] For example, please see Figure 3 , Figure 3 This is a schematic diagram of the network structure of the initial feature processing module 131 provided in an embodiment of this application. The initial feature processing module 131 is used to perform preliminary feature extraction on the input image to provide a compact representation for subsequent feature learning. Specifically, the initial feature processing module 131 includes an initial feature transformation unit and a spatial response enhancement unit.

[0033] Let the original image input to the initial feature processing module 131 be:

[0034]

[0035] in, This represents all possible real-valued values. H represents the height dimension of the original image, and W represents the width dimension, both measured in pixels. 3 represents the value of the third dimension, corresponding to the three color channels of a standard RGB color image.

[0036] The initial feature transformation unit is used to extract basic local structures from the original image, control the feature size reduction ratio, and output an intermediate feature map that retains the features of the local structures. The mathematical representation of the initial feature transformation unit is as follows:

[0037]

[0038] in, This represents the inter-channel convolution calculation in a local region, using standard convolution. In one possible implementation, the convolution size can be 3×3, the convolution stride is 2, the image edge padding length is 2, the receptive field dilation rate is 2, and the number of output channels is 40. This indicates a normalization transformation used to suppress scale differences. In this embodiment, batch normalization is used. The nonlinear activation function is represented by the Rectified Linear Unit (ReLU) in this embodiment.

[0039] The spatial response enhancement unit is used to spatially enhance local structural features in the intermediate feature map through depthwise convolution operations to obtain the base feature map. The mathematical representation of the spatial response enhancement unit is:

[0040]

[0041] in, The convolution calculation function for channel-wise spatial feature extraction is represented by depth-wise convolution. In one possible implementation paradigm, the convolution size is 3×3, the convolution stride is 2, and the image edge padding length is 2. and All of these are consistent with the meaning in the mathematical representation of the initial feature transformation unit.

[0042] For example, please see Figure 4 , Figure 4 This is a schematic diagram of the network structure of the multi-stage lightweight feature extraction module provided in an embodiment of this application. The multi-stage lightweight feature extraction module is composed of multiple stage structural units stacked sequentially, used to progressively construct mid-to-high-level semantic features of the original image while decreasing the spatial size. Each stage structural unit includes a lightweight feature construction submodule and a feature compression and fusion submodule.

[0043] The lightweight feature construction submodule is composed of multiple stacked residual enhancement lightweight feature construction units with consistent structure. Each residual enhancement lightweight feature construction unit is used to perform local spatial awareness, in-channel dynamic modeling and short-circuiting operations. In the lightweight feature construction submodule, the stacking depth and number of channels of the residual enhancement lightweight feature construction units are configured according to the stage of feature extraction.

[0044] Let the input features of the lightweight feature construction submodule be:

[0045]

[0046] in, This indicates the sequence number of the current processing stage in the lightweight feature construction submodule, used to distinguish feature levels of different depths in the network. As the value of i increases, it corresponds to a deeper level of feature abstraction in the network. This indicates that each element in the feature map is a real value. These values ​​have undergone nonlinear transformations in the preceding layers and already contain semantic information from low to high levels. and These represent the height and width dimensions of the feature map in stage i, respectively. These two spatial dimensions gradually decrease as the network depth increases, reflecting the gradual compression and abstraction of spatial information during feature extraction. As a channel dimension, it represents the number of channels in the feature map of the i-th stage. This parameter usually increases gradually with the network depth to accommodate richer feature representations.

[0047] The cascade aggregation operation of the lightweight feature construction submodule is represented as follows:

[0048]

[0049]

[0050]

[0051] in, For the initial state variables, This is the starting point for the entire process, representing the feature map from the previous stage, with a superscript. It indicates the current network depth stage, and this stage number directly affects the configuration parameters of subsequent processing units. As a loop variable, it is used to mark the sequence number of the current processing unit. Its value ranges from 0 to N-1, corresponding to the N residual enhancement lightweight feature building units stacked in this stage. This represents spatially aware enhanced convolution computation based on a local window, using depthwise convolution. In one possible implementation, the convolution size is 3×3, the convolution stride is 1, and the image edge padding is 1. and All of these are consistent with the meanings in the initial feature processing module mentioned above. As the final output, it is marked as To emphasize its stage attributes, this output feature contains both the deep features extracted step by step by all processing units in this stage, and retains the detailed information of the shallow features through cross-layer connections.

[0052] The feature concatenation operation of the lightweight feature construction submodule is represented as follows:

[0053]

[0054] in, The output of the feature concatenation operation. To represent feature concatenation, let the input tensor be:

[0055]

[0056] The splicing operation can then be represented as:

[0057]

[0058] Among them, the input tensor and This represents two feature maps that need to be stitched together, where H and W represent the height and width dimensions of the feature maps, respectively. These two spatial dimensions must be completely consistent for the stitching operation to be performed. and These represent the number of channels in the two feature maps, respectively. These two parameters can be the same or different, and determine the degree of channel expansion of the final concatenated feature map. For concatenation operators abbreviated form, This indicates the dimensional characteristics of the output feature y after concatenation. The spatial dimension H×W remains unchanged, while the channel dimension is expanded to the sum of the number of channels in the two feature maps. This approach emphasizes that the channel dimension is expanded as a whole, rather than being a superposition of spatial dimensions. This representation method precisely distinguishes between channel concatenation and spatial stacking, reflecting the precise control over feature dimensions in deep learning. The entire concatenation process actually constructs a richer feature representation space, allowing subsequent convolutional operations to perform more flexible feature selection and recombination within this expanded channel space. This enhances the network's representational capabilities without increasing spatial computational complexity.

[0059] In one possible implementation, the multi-stage lightweight feature extraction module can be composed of four stacked structural units to ensure that the receptive field is gradually expanded and the semantic meaning of features is enhanced while controlling the module complexity. Each stage's structure consists of a group of lightweight feature construction units and a feature compression and fusion unit connected in series. The number of units in the lightweight feature construction unit group varies in different stages, as mentioned above. i This indicates the stage number, with N representing the number of repeated stackings of the structure. When i When the value is 1 to 3, five residual-enhanced lightweight feature construction units are stacked, i.e. N The value is set to 5 to ensure the full extraction of the mid-layer representation capabilities; i When the value is 4, four residual-enhanced lightweight feature construction units are stacked, i.e. N The value is 4 to prevent computational redundancy and overfitting caused by excessive stacking.

[0060] The feature compression and fusion submodule is used to concatenate and fuse intermediate features within the current stage and perform spatial downsampling to obtain the high-level feature map. This submodule includes channel compression operators (such as grouped pointwise convolution), pooling operations, and global response modulation mechanisms (such as GRN). This module implements feature size control and cross-channel information enhancement.

[0061] The mathematical representation of channel compression and fusion processing is:

[0062]

[0063] in, As the core output feature, This represents the linear channel fusion convolution computation under the grouping method, which uses grouped point-wise convolution with a convolution size of 1×1 and a number of groups. Defined as:

[0064]

[0065]

[0066] in, This indicates downsampling pooling calculation, which includes two types: max pooling and mean pooling. i Indicates the stage ordinal number. i When the value is 1~3, Max pooling is used. i When it is 4, Mean pooling is used. and These represent the number of channels for the input and output features in the i-th stage, respectively. As a spatially compressed feature output, it reduces spatial resolution while maintaining feature discriminative power. This represents a normalization transformation, specifically Global Response Normalization (GRN). The mathematical representation of Global Response Normalization is:

[0067] Let the input be for:

[0068]

[0069] Channel mean for:

[0070]

[0071] Variance normalization:

[0072]

[0073] Channel response recalibration (learnable scalar):

[0074]

[0075] in, and It is the spatial size after multiple downsampling. The increasing channel dimension reflects the network's attention shift from spatial features to channel features. Representing spatial location in the feature map The activation value of the c-th channel. These are the standardized eigenvalues. This is a very small constant introduced to prevent division by zero errors and to avoid over-amplification of low-variance channels. The final refined feature quantity is the output. and These are learnable channel-level scalar parameters. The training process automatically learns the importance of each channel, thereby amplifying key features and suppressing redundant features. The channel bias term provides the necessary translational degrees of freedom for the normalized features, preventing the network's expressive power from being overly constrained. Both of these parameters are vectors with the same dimension as the number of channels C. They are automatically optimized through backpropagation during training, which reflects the technical concept of trading lightweight parameters for performance improvement.

[0076] For example, please see Figure 5 , Figure 5 This is a schematic diagram of the structure of the global feature refinement module provided in an embodiment of this application.

[0077] The input feature map for the global feature refinement module 133 is:

[0078]

[0079] The spatial compression and feature specification of the global feature refinement module 133 are represented as follows:

[0080]

[0081] in This indicates a spatial compression transformation within the channel; : indicates that Reduced to The adaptive aggregation operation is used to compress global two-dimensional spatial features into vectors; here, global adaptive mean pooling is used.

[0082] The final classification decision output is:

[0083]

[0084] Where W is the weight matrix of the fully connected layer; b is a bias vector of the same dimension, providing a baseline threshold for each class to enhance the model's fitting ability. This represents the activation function of the output layer, which uses the Softmax activation function to output the class probability. The design of the entire classification decision module 134 embodies the optimization idea of ​​lightweight networks. It significantly reduces the parameter size of the fully connected layers through global feature compression, and then achieves fast classification through efficient matrix operations.

[0085] In some optional embodiments, the emergency rescue equipment 10 also includes a point cloud acquisition module mounted on the UAV carrier 11. This module acquires point cloud data from the disaster site. Specifically, the point cloud acquisition module uses a lidar device. The lidar integrates a high-precision laser, a POS (Position and Orientation System), and a camera, and has a remote switch. The lidar can be controlled via a ground station, effectively preventing the UAV from collecting invalid data during takeoff and return. The lidar saves the acquired distance data, UAV position and attitude data, camera data, and timestamps to a memory card. It can also transmit this data back to the ground workstation in real time. The ground workstation software can browse and save the point cloud data in real time. Based on the relationship between the camera data and the lidar point cloud data, the software assigns color attributes to each lidar point, making it easier to identify terrain features and structures during real-time and result browsing.

[0086] In some alternative embodiments, the unmanned aerial vehicle carrier 11 in the emergency rescue equipment 10 can be a drone using a high-lift three-bladed propeller or a racing drone using a four-in-one ESC system.

[0087] Specifically, for emergency rescue needs at high altitudes, the drone can employ a high-lift three-bladed propeller, enabling stable takeoff and landing at altitudes of 4000m; it uses ultra-low temperature lithium batteries as a power source, allowing normal operation in temperatures as low as -20℃; and it employs an intelligent FOC magnetic field-guided motor ESC system, capable of adjusting the magnitude and angle of current to control motor output, with a maximum pull of 8.45kg per motor. The maximum takeoff weight of the high-altitude drone exceeds 15kg, enabling rapid movement at 60km / h and withstanding maximum wind speeds of 12m / s.

[0088] To address the emergency rescue needs of highways, drones can be lightweight racing drones with a four-in-one ESC system, providing stronger power and flight stability; they can also use the Flightone closed-source flight control system with carefully calibrated parameters for easier flight; and they can have a built-in flight path planning system for easier disaster early warning missions on highways.

[0089] In some optional embodiments, the image acquisition device 12 can be a video pod, which can specifically be a visible light pod, an infrared pod, or a dual-light pod. A visible light pod can be used to acquire data in well-lit areas; an infrared pod can be used to acquire data in poorly lit areas; and a dual-light pod can be used for long-term monitoring or operation in areas with significant differences in lighting conditions.

[0090] In some optional embodiments, the emergency rescue equipment 10 also includes an active power supply module mounted on the UAV carrier 11, the active power supply module including an airborne power step-down module and an optoelectronic integrated cable device;

[0091] The airborne power step-down module is used to convert the high-voltage DC voltage into the operating voltage required by the UAV carrier 11;

[0092] The optoelectronic integrated cable device includes an optoelectronic integrated cable and a cable winch. The optoelectronic integrated cable is used to transmit current and communication signals. The cable winch is used to manage the cable winding and unwinding. The cable winch is integrated with an optoelectronic conversion module for connecting to a ground workstation to send the airborne data of the UAV carrier 11 to the ground workstation.

[0093] Specifically, the ground power supply equipment rectifies and converts the 220V AC power supplied by the generator or mains power into 380V~420V high-voltage DC power to reduce the loss of long-distance power supply cables. It also has power failure alarm and power supply status display functions.

[0094] The aforementioned integrated optoelectronic cable device consists of an integrated optoelectronic cable and a cable winch. The integrated optoelectronic cable integrates power supply and communication functions, continuously transmitting 3-6A of current, and also includes a communication optical fiber for information transmission between the UAV and ground equipment. The integrated cable is lightweight and encapsulated with insulating material to reduce its weight. The integrated cable is wound onto the cable winch. The cable winch has automatic cable feeding and reeling functions. Before operation, press the cable feeding button on the winch. During UAV takeoff, if its lift exceeds a certain threshold, the winch begins feeding the cable; when the lift is detected to be below another threshold, the winch stops feeding the cable. After the UAV reaches the predetermined altitude, press the lock button on the winch to initiate hovering operations. After operation, press the reeling button on the winch, and the winch will reel in the cable as the UAV descends.

[0095] The airborne power step-down module can be a DC-DC module, which converts the high-voltage DC voltage from the ground power supply equipment into the voltage required for the UAV to operate, enabling the UAV to hover for extended periods. Simultaneously, the power module can charge the onboard backup battery, allowing the UAV to safely return or make an emergency landing in case of an emergency. The airborne power module also provides a photoelectric conversion module, which can transmit data from the video pod and flight control system to the ground workstation via fiber optic cable.

[0096] In some optional embodiments, the emergency rescue equipment 10 also includes an audible and visual warning device for sending warning information. Specifically, the audible and visual warning device has functions such as red and blue flashing warnings, brief text display, and high-decibel sound warnings. It is characterized by its light weight, high brightness, and loud sound. When an accident or disaster occurs, a drone carrying this device can quickly transmit disaster information to the surrounding area. Especially on highways, when monitoring detects damage or collapse of roadbeds and bridges, the management can set the drone's flight path and launch the drone from upstream of the disaster point to warn vehicles that a disaster has occurred ahead, thereby preventing vehicles from continuing to move forward.

[0097] In the above implementation process, key detailed features in disaster scenes are captured by layered stacking of image recognition modules. Then, redundant information is eliminated by spatial compression, and the correlation between features is strengthened by semantic fusion. Finally, high-precision association is established between abstract features and disaster types. Through a three-level progressive mechanism of initial feature processing, multi-stage feature deepening, and global feature optimization, the generated global feature vector can comprehensively represent the typical features of disasters. It can identify both obvious disaster features such as floods and fires, as well as subtle changes such as landslides and building cracks, thereby improving the accuracy of disaster information acquisition.

[0098] Based on the same concept, this application also provides an emergency rescue system based on unmanned aerial vehicles (UAVs). Please refer to [link / reference]. Figure 6 , Figure 6 This is a schematic diagram of the connection of an emergency rescue system provided in an embodiment of this application.

[0099] In addition to the emergency rescue equipment 10 described above, the emergency rescue system may also include a ground workstation, an emergency rescue command vehicle, and ground power supply equipment. In this embodiment, "UAV" refers to all UAVs capable of participating in rescue applications; "airborne equipment" refers to all data acquisition, communication, additional services, and early warning equipment required for rescue; the ground workstation is responsible for receiving and processing data at the rescue site and outputting rescue auxiliary decisions; the emergency rescue command vehicle is a modified vehicle equipped with the necessary rescue equipment and tools, possessing good mobility. Specifically, for disasters such as landslides and collapses, UAVs can be equipped with terrain modeling data acquisition equipment such as lidar to collect data from the disaster area. After processing the data, the ground workstation generates a disaster model. Software tools can measure the model to determine the scale of the disaster and intelligently output a plan for the necessary manpower, material resources, and traffic diversion for the command center's reference.

[0100] For search and rescue missions, drones can be equipped with visible light and infrared pods to collect image data of the search and rescue area, which is then transmitted in real time to a ground station via image transmission equipment. Please see [link / reference]. Figure 7 , Figure 7 This is a schematic diagram illustrating the video streaming of emergency rescue equipment provided in an embodiment of this application. For search and rescue missions, drones can carry visible light and infrared pods to collect image data of the search and rescue area, and the data is transmitted in real time to a ground workstation via image transmission equipment. For example... Figure 7 As shown, the ground workstation can push video data to the conference system of the emergency rescue command vehicle for the command center to make decisions; push it to the ground station monitoring software so that personnel at the rescue site can understand the disaster situation; push it to the remote controller so that the pilot can operate the drone; and push it to the video recognition algorithm to identify objects with certain characteristics.

[0101] The ground workstation is connected to the image acquisition device 12 in the emergency rescue equipment 10 and is also connected to the emergency rescue equipment 10 for communication purposes. It is used to receive carrier data from the UAV carrier 11 and to process the carrier data.

[0102] Data from the image acquisition module is transmitted to the ground workstation via the image transmission system. The ground workstation, acting as a server, saves the received data and can process it as follows: 1. Data is streamed to the pilot's remote controller for the pilot to operate the drone; 2. Data is streamed to the monitoring interface of the ground workstation for on-site command personnel to view and make decisions; 3. Data is streamed to the image recognition program of the ground workstation, which can extract features of traffic incidents and geological disasters; 4. Data is streamed to the vehicle-mounted video matrix, and the images captured by the drone are transmitted to the emergency command center through the vehicle-mounted conferencing system.

[0103] In addition, the ground workstation can push video data to the conference system of the emergency rescue command vehicle for the command center to make decisions; push it to the ground station monitoring software so that personnel at the rescue site can understand the disaster situation; push it to the remote controller so that the drone operator can use it; and push it to the video recognition algorithm to identify objects with certain characteristics.

[0104] The emergency rescue command vehicle is communicatively connected to the ground workstation for maintaining data interaction with the ground workstation; the ground power supply equipment is connected to the UAV carrier 11 via an optoelectronic integrated cable device for supplying power to the UAV carrier 11.

[0105] Specifically, the power supply system allows users to choose between generator power or mains power via a toggle switch. Both power supply modes are combined with a UPS to meet equipment power needs in emergency situations. Please refer to [link / reference]. Figure 8 , Figure 8 This is a schematic diagram illustrating the operation of ground power supply equipment providing uninterrupted power to emergency rescue equipment, as provided in an embodiment of this application.

[0106] To address the rescue needs of long-duration hovering monitoring, an uninterrupted power supply (UPS) mode is adopted to improve the drone's endurance. The power take-off generator integrated in the emergency rescue command vehicle generates AC220V power, which is input to a power rectifier, rectifying the power to DC370V~DC420V. This DC power is then supplied to the onboard power step-down module via a composite cable. The output of the step-down module is connected in parallel with a backup battery to the drone's power input. The backup battery ensures the drone can make an emergency landing in the event of a UPS failure. The fiber optic cable is used to transmit image data from the pod.

[0107] The video matrix can direct various video and image output signals to different video receivers, and the interconnection between video sources and receivers can be flexibly switched via buttons. The video conferencing system can easily initiate video conferences and can also be called upon by others to join other conference rooms. During video conferences, images from multiple video sources can be transmitted to the remote location in a multi-view format, meeting the needs of multi-angle viewing and saving bandwidth resources. In emergency rescue operations, the situation inside the vehicle, video collected by drones, and data from handheld video devices at ground stations can be simultaneously transmitted to a remote emergency command center, allowing the command center to quickly make decisions based on multi-angle information.

[0108] The network system is divided into mobile communication networks and satellite communication networks to ensure uninterrupted network connectivity in various environments.

[0109] Meteorological equipment, including six-element weather instruments, visibility weather instruments, road surface temperature collectors, and road surface water film thickness measuring instruments, allows the command center to refer to the data from these devices to formulate more reasonable decision-making plans.

[0110] It should be understood that when the various modules of the system provided in the above embodiments are working, the division of each functional module in the above description is only used as an example. In actual applications, the above functions can be assigned to different functional modules as needed. That is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above.

[0111] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. Emergency rescue equipment based on unmanned aerial vehicles (UAVs), characterized in that: The emergency rescue equipment includes an unmanned aerial vehicle (UAV) carrier and an image acquisition device mounted on the UAV carrier. The emergency rescue equipment is equipped with an image recognition module, which includes an initial feature processing module, a multi-stage lightweight feature extraction module, a global feature refinement module, and a classification decision module, used to identify the current disaster type based on the images acquired by the image acquisition device. The process by which the image recognition module identifies the current disaster type based on the original images acquired by the image acquisition device includes: The initial feature processing module performs preliminary feature extraction on the image to obtain a basic feature map. The basic feature map is processed by stacking multiple stages according to the multi-stage lightweight feature extraction module to obtain an advanced feature map. The global feature refinement module performs spatial compression and semantic fusion on the high-level feature map to obtain a global feature vector. The classification decision module maps the global feature vector to a prediction result for the disaster category. The multi-stage lightweight feature extraction module includes multiple stage structural units, and each stage structural unit includes a lightweight feature construction sub-module and a feature compression and fusion sub-module. The lightweight feature construction submodule is composed of multiple stacked residual enhancement lightweight feature construction units with identical structures. Each residual enhancement lightweight feature construction unit is used to perform local spatial awareness, in-channel dynamic modeling, and short-circuiting operations. In the lightweight feature construction submodule, the stacking depth and number of channels of the residual enhancement lightweight feature construction units are configured according to the feature extraction stage. The feature compression and fusion submodule is used to splice and fuse intermediate features in the current stage and perform spatial downsampling to obtain the high-level feature map; The feature compression and fusion submodule includes a global response normalization mechanism, which is expressed as follows: in, and It is the spatial size after multiple downsampling. For the ever-increasing number of channel dimensions; Representing spatial location in the feature map The activation value of the c-th channel; These are the standardized eigenvalues. To introduce a minimal constant, The final refined feature quantity is the output. and These are learnable channel-level scalar parameters.

2. The emergency rescue equipment based on unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, The initial feature processing module includes an initial feature transformation unit and a spatial response enhancement unit; The initial feature transformation unit is used to extract local structural features from the original image, control the feature size reduction ratio of the local structural features, and output an intermediate feature map that retains the local structural features. The spatial response enhancement unit is used to spatially enhance the local structural features in the intermediate feature map through a depthwise convolution operation to obtain the base feature map.

3. The emergency rescue equipment based on unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, The emergency rescue equipment also includes a point cloud acquisition module mounted on the UAV carrier, which is used to acquire point cloud data at the disaster site.

4. The emergency rescue equipment based on unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, The drone carrier is either a drone with a high-lift three-bladed propeller or a racing drone with a four-in-one ESC system.

5. The emergency rescue equipment based on unmanned aerial vehicles according to claim 1, characterized in that, The image acquisition equipment is a visible light pod, an infrared pod, or a dual-light pod.

6. The emergency rescue equipment based on unmanned aerial vehicles according to claim 1, characterized in that, The emergency rescue equipment also includes an active power supply module mounted on the UAV carrier, the active power supply module including an airborne power step-down module and an optoelectronic integrated cable device; The airborne power step-down module is used to convert high-voltage DC voltage into the operating voltage required by the UAV carrier. The optoelectronic integrated cable device includes an optoelectronic integrated cable and a cable winch. The optoelectronic integrated cable is used to transmit current and communication signals; the cable winch is used to manage the winding and unwinding of the cable. The cable winch integrates a photoelectric conversion module for connecting to a ground workstation and sending the onboard data of the UAV carrier to the ground workstation.

7. The emergency rescue equipment based on unmanned aerial vehicles according to claim 1, characterized in that, The emergency rescue equipment also includes an audible and visual warning device, which is used to send warning information.

8. An emergency rescue system based on unmanned aerial vehicles (UAVs), characterized in that, The system includes the emergency rescue equipment as described in any one of claims 1-7; the emergency rescue system further includes a ground workstation, an emergency rescue command vehicle, and ground power supply equipment; wherein, the ground workstation is communicatively connected to the emergency rescue equipment and is used to receive carrier data from the UAV carrier and process the carrier data; the emergency rescue command vehicle is communicatively connected to the ground workstation and is used to maintain data interaction with the ground workstation; the ground power supply equipment is connected to the UAV carrier through an optoelectronic integrated cable device and is used to supply power to the UAV carrier.

9. The emergency rescue system according to claim 8, characterized in that, The emergency rescue command vehicle integrates one or more of the following: an integrated power supply system, a video matrix, a video conferencing system, an audio splitter, a network system, and meteorological equipment. The power supply system powers the emergency rescue command vehicle. The video matrix transmits multiple video or image output signals to different video receivers and switches the interconnection relationships between video sources at the video receivers. The video conferencing system is used to establish video conferences. The audio splitter manages and distributes multi-channel audio. The network system provides a network environment. The meteorological equipment includes a six-element meteorological instrument, a visibility meteorological instrument, a road surface temperature collector, and a road surface water film thickness measuring instrument, used to acquire multi-source meteorological data.

Images