Detection method and system for early fire behavior

Abstract

The invention relates to a detection method and system for an early fire behavior, and belongs to the technical field of image processing, and the method comprises the steps: obtaining an image group of a to-be-detected region shot by a camera, and forming a target tensor; obtaining first mapping data and second mapping data; performing Tucker decomposition on the first mapping data to obtain a core tensor, and a first factor matrix and a second factor matrix which are orthogonal; respectively inputting the first mapping data and the second mapping data into a convolutional network, and fusing an output result with the core tensor, the first factor matrix and the second factor matrix to obtain a target fusion feature; and a detection head is adopted to detect the target fusion feature so as to detect and position the fire point. The application provides advanced sensing detection for early fire points, fire misjudgment can be reduced, tiny fire points can be effectively identified, personnel can conveniently and timely deal with the fire points, and accurate positioning for the fire points can be realized through the image and the detection head.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to a detection method and system for early fire Background Art

[0002] Most traditional fire-fighting equipment is passive and can only function after a fire breaks out, unable to effectively prevent and control fires. With the continuous development of electronic information technology, some intelligent fire control systems with automatic control and autonomous monitoring functions have emerged. Relying on the development of modern information technology, these intelligent fire control systems combine sensors, monitoring devices, data processing, and communication technologies to achieve real-time monitoring, early warning, and control of fires.

[0003] Existing fire-fighting intelligent control systems mainly achieve data collection, processing, and analysis through interface services of fire-fighting equipment, network equipment, and other devices. However, it has been found that although existing control systems have a certain ability of autonomous detection and autonomous processing, the handling of fire situations is not accurate, mainly manifested as follows: (1) Currently, most monitoring mainly identifies and alarms through smoke detection. Due to the fluidity of smoke, multiple smoke detectors may be triggered, resulting in multiple fire extinguishing sprinklers performing fire extinguishing operations. However, in some cases, detecting smoke does not necessarily mean a fire, such as water vapor, steam, or a high-temperature environment. In addition, when an open fire appears in an office but the diameter of the open fire is less than 10 cm and the smoke is scarce, the existing smoke monitoring system cannot detect the occurrence of a fire in advance, and thus cannot timely notify the fire-fighting person in charge for handling.

[0004] (2) Implementing fire situation monitoring through the interfaces of fire-fighting equipment, network equipment, and other devices cannot quickly locate the fire point, and it can only be known through manual investigation or when the fire spreads and triggers a smoke detector. This is fatal for an accident like a fire where every second counts.

[0005] It can be seen that the current fire situation monitoring methods have defects such as misjudgment of fire situations, inability to detect small open fires in advance, and the need for manual location of the fire point. Summary of the Invention

[0006] To solve the above-mentioned problems of the prior art, the present invention provides a detection method and system for early fire

[0007] In a first aspect, an embodiment of the present application provides a detection method for early fire, including: obtaining an image group captured by a camera for an area to be detected and forming a target tensor; wherein, the image group includes multiple frames of images, and each frame of image includes an identified image background and a detection object; obtaining first mapping data and second mapping data; wherein, the first mapping data is the mapping of all image backgrounds in the image group in the target tensor, and the second mapping data is the mapping of all detection objects in the image group in the target tensor; performing Tucker decomposition on the first mapping data to obtain a core tensor, an orthogonal first factor matrix, and a second factor matrix; inputting the first mapping data and the second mapping data into a convolutional network respectively, and fusing the output results with the core tensor, the first factor matrix, and the second factor matrix to obtain a target fusion feature; using a detection head to detect the target fusion feature to detect and locate a fire point.

[0008] Optionally, the step of inputting the first mapping data and the second mapping data into a convolutional network respectively, and fusing the output results with the core tensor, the first factor matrix, and the second factor matrix to obtain a target fusion feature includes: inputting the first mapping data into the convolutional network to obtain a first feature; inputting the second mapping data into the convolutional network to obtain a second feature; wherein, the output results include the first feature and the second feature; fusing the second feature with the core tensor, the first factor matrix, and the second factor matrix to obtain a first fusion feature; adding the first fusion feature and the first feature to obtain the target fusion feature.

[0009] Optionally, when the scales of the first fusion feature and the first feature are not unified, the method further includes: performing a zero-padding operation on the feature with a smaller scale among the first fusion feature and the first feature.

[0010] Optionally, the method further includes: if a fire point is detected, determining the target fire-fighting equipment closest to the fire point based on the positioning data of the fire point, and uploading detection data to a control terminal; wherein, the target fire-fighting equipment is the fire-fighting equipment closest to the fire point; wherein, the detection data includes an approved control instruction for the target fire-fighting equipment and the positioning data of the fire point.

[0011] Optionally, the convolutional network includes three convolutional layers.

[0012] Optionally, the method further includes: obtaining a first image of the area to be detected; inputting the first image into an improved CSPNet model to obtain a first feature map; wherein, the improved CSPNet model adopts a convolutional module based on regional attention; using the detection head to detect the first feature map to detect smoke.

[0013] Optionally, the method further includes: obtaining all cameras in the coverage area; wherein each camera captures an area to be detected; constructing a fire warning map based on all cameras in the coverage area; wherein in the fire warning map, each camera is used as a node, and an edge is established between two nodes that are regionally connected; predicting the fire spread direction based on the fire warning map.

[0014] Optionally, the edges in the fire warning map are set with weights; the weights of the edges are determined by the connection type between two nodes, the physical distance between two nodes, and / or the flammability attribute of the materials between two nodes.

[0015] Optionally, predicting the fire spread direction based on the fire warning map includes: when a target node with a fire is detected, determining the fire spread direction based on the weights of the edges between other nodes adjacent to the target node in the fire warning map.

[0016] In a second aspect, the present application provides a detection system for early fire situations, including: a first acquisition module, configured to acquire an image group captured by a camera of an area to be detected and form a target tensor; wherein the image group includes multiple frames of images, and each frame of image includes an identified image background and a detection object; a second acquisition module, configured to acquire first mapping data and second mapping data; wherein the first mapping data is the mapping of all image backgrounds in the image group in the target tensor, and the second mapping data is the mapping of all detection objects in the image group in the target tensor; a first processing module, configured to perform Tucker decomposition on the first mapping data to obtain a core tensor, an orthogonal first factor matrix, and a second factor matrix; a second processing module, configured to input the first mapping data and the second mapping data into a convolutional network respectively, and fuse the output results with the core tensor, the first factor matrix, and the second factor matrix to obtain a target fusion feature; a detection module, configured to use a detection head to detect the target fusion feature to detect and locate a fire point.

[0017] The beneficial effects of the present invention include: The present application provides a detection method for early fire situations. Considering that current fire early warning mainly relies on smoke recognition, which has misjudgment problems and cannot detect tiny fire spots in advance, a dynamic recognition method based on images is proposed. The core of this method lies in "dynamic background modeling based on Tucker decomposition and fire spot detection with multi-feature fusion". Specifically, based on the image background, the first mapping data of the time series is constructed, and then features are extracted through Tucker decomposition. The decomposed background features are more compact and can better distinguish real fire spots from background interferences (such as reflective points and movable objects in the office). Then, the decomposed features and the output results of the convolutional network are fused together to form the target fusion features, which can enhance the saliency of the fire spot features and suppress background redundant information, thereby improving the accuracy of fire spot detection. This method can effectively identify tiny fire spots (specifically, fire spots below 10 cm). That is, the present application provides an early perception detection for early fire spots, which can reduce fire misjudgment, effectively identify tiny fire spots, facilitate personnel to respond and handle in a timely manner, and can accurately locate the fire spots through the image and the detection head. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a flowchart of the steps of a detection method for early fire situations provided by an embodiment of the present invention; Figure 2 It is a flowchart of the steps of another detection method for early fire situations provided by an embodiment of the present invention; Figure 3 It is a schematic diagram of a data processing process provided by an embodiment of the present invention; Figure 4 It is a schematic diagram of an improved CSPNet model provided by an embodiment of the present invention; Figure 5 It is a flowchart of the steps of yet another detection method for early fire situations provided by an embodiment of the present invention; Figure 6 It is a block diagram of the modules of a detection system for early fire situations provided by an embodiment of the present invention; Figure 7 It is a block diagram of the modules of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0019] In the following description, specific details such as specific system structures and technologies are proposed for the purpose of illustration rather than limitation, so as to thoroughly understand the embodiments of the present application. However, those skilled in the art should clearly understand that the present application can also be implemented in other embodiments without these specific details. In other cases, the detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid unnecessary details from interfering with the description of the present application.

[0020] In addition, in the description of the specification and the appended claims of the present application, the terms "first", "second", "third", etc. are only used for distinguishing descriptions and should not be construed as indicating or implying relative importance.

[0021] Existing fire intelligent control systems mainly realize data collection, processing, and analysis through interface services of devices such as fire-fighting equipment and network equipment. However, it is found that although existing control systems have a certain ability of autonomous detection and autonomous processing, the handling of fire situations is not accurate, mainly manifested as follows: (1) Currently, most monitoring mainly identifies and alarms through smoke detection. Due to the fluidity of smoke, multiple smoke detectors may be triggered, resulting in multiple fire sprinklers performing fire extinguishing operations. However, in some cases, detecting smoke does not mean there is a fire, such as water vapor, steam, and high-temperature environments. In addition, when there is an open fire in an office but the diameter of the open fire is less than 10 cm and the smoke is less, the existing smoke monitoring system cannot sense the occurrence of the fire situation in advance, and thus cannot notify the fire responsible person for handling in a timely manner.

[0022] (2) Implementing fire situation monitoring through the interfaces of devices such as fire-fighting equipment and network equipment cannot quickly locate the fire point. It can only be known through manual investigation or when the fire spreads and triggers the smoke detector. This is fatal for an accident like a fire where every second counts.

[0023] It can be seen that the current fire situation monitoring method has defects such as misjudgment of fire situations, inability to detect in advance when the open fire is small, and the need for manual positioning of the fire point.

[0024] In view of the above problems, the present application proposes the following embodiments to solve the above technical problems.

[0025] Please refer to Figure 1 , an embodiment of the present application provides a detection method for early fire situations, and the method includes: Step 101 to Step 105.

[0026] Step 101: Obtain an image group captured by a camera for the area to be detected and form a target tensor.

[0027] Among them, the image group includes multiple frames of images, and each frame of image includes the identified image background and detection object.

[0028] In the embodiment of the present application, the camera can be a device with a camera function such as a home camera, a network camera, a monitor, etc. The area to be detected is the scene area that the camera can capture, and the area to be detected can specifically be a company office, a corridor, a living room, a computer room, etc.

[0029] Specifically, assuming the image collected by the camera is , each image can be composed of two parts, namely ; wherein, represents the image background, and represents the detection object. The number of the image background and the detection object can be multiple.

[0030] The above-mentioned image background and detection object can be obtained through the top-down pixel aggregation algorithm, and of course, can also be determined through any other recognition algorithm.

[0031] In this embodiment, the detection object can be a dynamic object in the image.

[0032] Specifically, an image group I within the frame rate T per unit time can be taken ( , ,..., ). For example, if T = 30, the image group I includes 15 frames of images.

[0033] In this embodiment, the image frame I can form a target tensor with a scale of 1920×1080×3× .

[0034] Step 102: Obtain the first mapping data and the second mapping data.

[0035] Wherein, the first mapping data is the mapping of all the image backgrounds in the image group in the target tensor, and the second mapping data is the mapping of all the detection objects in the image group in the target tensor.

[0036] Specifically, the first mapping data can be expressed as , that is, the mapping in the target tensor is ; the second mapping data can be expressed as , that is, the mapping in the target tensor is .

[0037] Step 103: Perform Tucker decomposition on the first mapping data to obtain a core tensor, an orthogonal first factor matrix, and a second factor matrix.

[0038] That is, perform Tucker decomposition on the first mapping data to obtain the core tensor and two orthogonal first factor matrices and the second factor matrix .

[0039] Wherein, the representation of the decomposition process can be specifically: ; Wherein,​ and are both regularization parameters, takes the value of 2, represents the norm, represents the Frobenius norm.

[0040] Step 104: Input the first mapping data and the second mapping data into the convolutional network respectively, and fuse the output results with the core tensor, the first factor matrix, and the second factor matrix to obtain the target fusion feature.

[0041] Then, perform data fusion to form the target fusion feature.

[0042] Step 105: Use the detection head to detect the target fusion feature to detect and locate the fire point.

[0043] Among them, the detection head is a module in the neural network specifically responsible for predicting the target position and category. In this step, the target fusion feature is input into the detection head to complete the detection and location of the fire point.

[0044] In summary, the embodiment of the present application provides a detection method for early fire conditions. Considering that the current fire warning mainly has misjudgments through smoke recognition and cannot detect micro fire points in advance, a dynamic recognition method based on images is proposed. The core of this method lies in "fire point detection based on Tucker decomposition for dynamic background modeling and multi-feature fusion". Specifically, based on the image background, the first mapping data of the time series is constructed, and then features are extracted through Tucker decomposition. The decomposed background features are more compact and can better distinguish real fire points from background interferences (such as reflective points and movable objects in the office). Then, the decomposed features and the output results of the convolutional network are fused together to form the target fusion feature, which can enhance the saliency of the fire point features, suppress background redundant information, and improve the accuracy of fire point detection. This method can effectively identify micro fire points (specifically fire points below 10 cm). That is, the embodiment of the present application provides an early perception detection for early fire points, which can reduce fire misjudgments, effectively identify micro fire points, facilitate personnel to respond in a timely manner, and can accurately locate the fire point through the image plus the detection head.

[0045] Please refer to Figure 2 , the above step of inputting the first mapping data and the second mapping data into the convolutional network respectively, and fusing the output results with the core tensor, the first factor matrix, and the second factor matrix to obtain the target fusion feature can specifically include: Step 201 to Step 204.

[0046] Step 201: Input the first mapping data into the convolutional network to obtain the first feature.

[0047] Specifically, input the first mapping data into the convolutional network to obtain the first feature .

[0048] Step 202: Input the second mapping data into the convolutional network to obtain the second feature.

[0049] Specifically, input the second mapping data into the convolutional network to obtain the second feature .

[0050] Among them, the output result includes the first feature and the second feature .

[0051] Step 203: Fuse the second feature with the core tensor, the first factor matrix, and the second factor matrix to obtain the first fused feature.

[0052] Specifically, the fusion method can refer to the following formula: ; Among them, represents the first fused feature, represents a hyperparameter, represents the L1 norm, represents the natural logarithm of the determinant of the matrix, and the explanations of the remaining parameters can be referred to in the aforementioned formula and will not be elaborated here.

[0053] It should be noted that in the embodiments of the present application, the information amount of the factor matrix is constrained by the logarithm of the determinant to suppress background redundant information and enhance the saliency of the fire point feature.

[0054] Step 204: Add the first fused feature to the first feature to obtain the target fused feature.

[0055] Finally, directly add the first fused feature to the first feature , and the target fused feature can be obtained (corresponding to + ).

[0056] Please refer to Figure 3 , and the data flow will be shown below. Perform Tucker decomposition on the first mapping data to obtain the core tensor , the orthogonal first factor matrix and the second factor matrix . At the same time, input the first mapping data into the convolutional network to obtain the first feature . The second mapping data Input into the convolutional network to get the second feature Then, the second feature With core tensors , the first factor matrix And the second factor matrix fusion , get the first fusion feature Finally, the first fusion feature With the first feature Add them together to get the target fusion feature.

[0057] In one embodiment, when the scales of the first fused feature and the first feature are not uniform, the method further includes: performing an edge-filling zero value operation on the first fused feature and a feature with a smaller scale in the first feature.

[0058] That is, the edge filling operation with zero values ​​can make the fused features cover all spatial positions, thus preventing the fire point positions from being omitted due to size clipping.

[0059] Optionally, in one embodiment, the method further includes: if a fire point is detected, determining the target fire-fighting equipment closest to the fire point based on the positioning data of the fire point, and uploading the detection data to the control terminal.

[0060] The target fire-fighting equipment is the fire-fighting equipment closest to the fire point; and the detection data includes the approval control instructions for the target fire-fighting equipment and the positioning data of the fire point.

[0061] Among them, in the embodiment of the present application, the fire fighting equipment can be specifically a sprinkler head, a water gun, a fire fighting robot, etc. In the embodiment of the present application, by detecting the location of the fire point in real time, automatically matching the nearest target fire fighting equipment, and uploading the detection data to the control terminal to shorten the response control data, this solution can trigger the equipment to start within seconds and suppress the initial spread of the fire.

[0062] Optionally, the above convolutional network may include three convolutional layers.

[0063] In summary, the existing system cannot detect open flames when the diameter of the open flame is less than 10 cm and there is little smoke. The present invention uses the images collected by the network camera to detect the fire point to achieve early detection of the fire. According to the characteristics of the fire point on the image, the present application uses Tucker decomposition to decompose the image background obtained by the pixel aggregation algorithm into core tensors and two orthogonal factor matrices and , then fuse the image background after three layers of convolution, the decomposed orthogonal components and the core tensor, and then fuse them with the target image after three layers of convolution. Finally, the detection head is used to detect whether there is a fire point to implement corresponding processing measures.

[0064] Optionally, the embodiments of the present application also provide an auxiliary detection method for early fire caused by smoke. That is, the above method may further include: obtaining a first image of the area to be detected; inputting the first image into an improved CSPNet model to obtain a first feature map; wherein, the improved CSPNet model uses a convolutional module based on regional attention; using a detection head to detect the first feature map to detect smoke.

[0065] That is, in the embodiments of the present application, the ordinary convolution in CSPNet is replaced by region attention-based convolution (RAC). RAC uses an attention mechanism to accurately evaluate the importance of each feature point within the receptive field, thereby reducing the information sparsity caused by parameter sharing. In other words, the CSPNet improved by RAC enhances the extraction of weak features of smoke to improve the detection accuracy.

[0066] Optionally, for the specific composition of the improved CSPNet model, please refer to Figure 4 , and the following is a complete description of the process: sending the first image image of the area to be detected into the improved CSPNet model, first applying group convolution to the first image image to extract the feature representation of each feature point within the receptive field, so as to obtain features with dimensions , where H, W, and C represent height, width, and number of channels respectively.

[0067] This process is followed by batch normalization and the sigmoid activation function. After reshaping, the receptive field spatial features with a size of are obtained.

[0068] Next, the spatial features extracted from the receptive field are used for spatial attention calculation. First, max pooling and average pooling are used, and additionally using image to obtain a feature map with a size of , and then convolutional operations and the sigmoid activation function are applied. At the same time, the Squeeze-and-Excitation method is used as the channel attention mechanism, which consists of global average pooling, a fully connected layer, and the sigmoid activation function. Finally, the spatial and channel attention maps are combined (by directly adding) and then sent into the detection head.

[0069] Optionally, please refer to Figure 5 , the detection method for early fire provided by the embodiments of the present application may further include: steps 501 to 503.

[0070] Step 501: Obtain all cameras in the coverage area.

[0071] Among them, each camera captures a detection area to be detected.

[0072] Among them, the covered area can specifically refer to specific monitoring scenarios, such as companies, factories, shopping malls, etc. Multiple cameras need to be installed in the covered area to achieve monitoring of the covered area. For example, when the covered area is a company, multiple cameras need to be installed, such as in the office, in the toilet, in the corridor, and on the stairs. In this example, the company corresponds to the covered area, and the office, toilet, corridor, and stairs correspond to the detection areas to be detected.

[0073] Step 502: Construct a fire warning map based on all the cameras in the covered area.

[0074] Among them, each camera in the fire warning map is used as a node, and an edge is established between two nodes where the areas are connected.

[0075] Exemplarily, if camera A monitors the company entrance and camera B monitors the front desk, then an edge is established between the two nodes of camera A and camera B.

[0076] Step 503: Predict the fire spread direction based on the fire warning map.

[0077] It should be noted that by constructing a fire alarm map, fire spread warning can be achieved, and the fire spread direction and trend can be known.

[0078] Optionally, in one embodiment, the edges in the fire warning map are set with weights; the weights of the edges are determined by the connection type between two nodes, the physical distance between two nodes, and / or the flammability attribute of the materials between two nodes.

[0079] The connection type between two nodes can include the following examples: open space = 3 (high risk), ordinary door = 2 (medium risk), fire door = 1 (low risk), ventilation opening (lower risk).

[0080] The longer the physical distance between two nodes, the longer the fire spread time.

[0081] The flammability attribute of the materials between two nodes can include: paper material = 3, wood = 2, metal material = 0.5, etc.

[0082] Among them, one calculation formula for the weight E can be: E = physical distance / spread speed × A + connection type × B + flammability attribute of materials × C; Among them, A, B, and C respectively represent the contribution ratios.

[0083] The finally formed fire warning map can be expressed as G(N, E), where N represents nodes and E represents edges.

[0084] Record the time when each camera detects a fire. If node N1 detects a fire earlier than adjacent node N2 and it conforms to the fire spread direction (obtained from the displacement of the detection box), then infer that the fire spreads from 1→2.

[0085] Optionally, in one embodiment, the above steps predict the fire spread direction based on a fire warning map, including: when a target node with a detected fire is identified, determine the fire spread direction based on the weights of the edges between other nodes adjacent to the target node in the fire warning map.

[0086] That is, if there are multiple adjacent nodes, select the one with the largest edge weight value as the potential area. As the camera nodes gradually detect a fire, dynamically update the fire status in the topology map. After a fire is detected, the current fire point location (the area monitored by the camera) and the fire spread location (the location inferred from the graph) can be output simultaneously to facilitate personnel to take corresponding countermeasures in a timely manner.

[0087] Please refer to Figure 6 , based on the same inventive concept, an embodiment of the present application further provides a detection system 600 for early fire, including: A first acquisition module 601, configured to acquire an image group captured by a camera for an area to be detected and form a target tensor; wherein, the image group includes multiple frames of images, and each frame of image includes an identified image background and a detection object.

[0088] A second acquisition module 602, configured to acquire first mapping data and second mapping data; wherein, the first mapping data is the mapping of all image backgrounds in the image group in the target tensor, and the second mapping data is the mapping of all detection objects in the image group in the target tensor.

[0089] A first processing module 603, configured to perform Tucker decomposition on the first mapping data to obtain a core tensor, an orthogonal first factor matrix, and a second factor matrix.

[0090] A second processing module 604, configured to input the first mapping data and the second mapping data into a convolutional network respectively, and fuse the output results with the core tensor, the first factor matrix, and the second factor matrix to obtain a target fusion feature.

[0091] A detection module 605, configured to use a detection head to detect the target fusion feature to detect and locate the fire point.

[0092] Please refer to Figure 7, based on the same inventive concept, an embodiment of the present application provides a module housing of an electronic device 700 that applies the above method. The electronic device 700 includes: at least one processor 701 ( Figure 7 only one is shown in the figure), a memory 702, and a computer program 703 stored in the memory 702 and executable on at least one processor 701. When the processor 701 executes the computer program 703, it implements the steps of the method in any of the foregoing embodiments.

[0093] The electronic device 700 may be a server, a personal computer, a laptop computer, and so on.

[0094] Those skilled in the art can understand that Figure 7 these are merely examples of the electronic device 700 and do not constitute a limitation on the electronic device 700. It may include more or fewer components than shown in the figure, or combine some components, or different components.

[0095] The so-called processor 701 may be a central processing unit (CPU), and the processor 701 may also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0096] In some embodiments, the memory 702 may be an internal storage unit of the electronic device 700, such as the hard disk or memory of the electronic device 700. In other embodiments, the memory 702 may also be an external storage device of the electronic device 700, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the electronic device 700. Further, the memory 702 may also include both the internal storage unit and the external storage device of the electronic device 700.

[0097] It should be noted that for the above systems, devices, etc., since they are based on the same concept as the method embodiments of the present application, the modules designed by the system and the steps executed by the device and the technical effects brought about can be referred to the method embodiment part, and will not be elaborated here.

[0098] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the above-mentioned division of each functional unit and module is used as an example. In actual applications, the above functions can be allocated to different functional units and modules according to needs, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. Each functional unit and module in the embodiments can be integrated into a processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of a software functional unit. In addition, the specific names of each functional unit and module are only for the convenience of mutual distinction and do not limit the protection scope of this application. The specific working processes of the units and modules in the above system can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0099] An embodiment of this application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps in the foregoing method embodiments can be implemented.

[0100] An embodiment of this application provides a computer program product. When the computer program product runs on a mobile terminal, the mobile terminal can implement the steps in the foregoing method embodiments when executed.

[0101] If the above-mentioned integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, to implement all or part of the processes in the method embodiments of this application, a computer program can be used to instruct relevant hardware to complete. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, the steps in the foregoing method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, executable file or some intermediate form, etc. The computer-readable medium can at least include: any entity or device that can carry the computer program code to the photographing device / electronic device, recording medium, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium. For example, a USB flash drive, a mobile hard disk, a magnetic disk, or an optical disc, etc.

[0102] In the above embodiments, the descriptions of each embodiment have their own emphases. For the parts not detailed or recorded in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0103] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.

[0104] In the embodiments provided in this application, it should be understood that the disclosed apparatus / network device and method can be implemented in other ways. For example, the apparatus / network device embodiments described above are merely illustrative. For example, the division of the modules or units is only a logical function division, and there may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some interfaces. The indirect couplings or communication connections of the devices or units can be in electrical, mechanical or other forms.

[0105] The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they can be located in one place, or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0106] 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 of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included in the protection scope of this application.

Claims

1. A detection method for early fire, characterized in that Including: Obtain an image group captured by a camera for the area to be detected and form a target tensor; wherein, the image group includes multiple frames of images, and each frame of image includes an identified image background and a detection object; Obtain first mapping data and second mapping data; wherein, the first mapping data is the mapping of all image backgrounds in the image group in the target tensor, and the second mapping data is the mapping of all detection objects in the image group in the target tensor; Perform Tucker decomposition on the first mapping data to obtain a core tensor, an orthogonal first factor matrix, and a second factor matrix; Input the first mapping data and the second mapping data into a convolutional network respectively, and fuse the output results with the core tensor, the first factor matrix, and the second factor matrix to obtain a target fusion feature; Use a detection head to detect the target fusion feature to detect and locate a fire point.

2. The early fire detection method according to claim 1, wherein The step of inputting the first mapping data and the second mapping data into a convolutional network respectively, and fusing the output results with the core tensor, the first factor matrix, and the second factor matrix to obtain a target fusion feature includes: Input the first mapping data into the convolutional network to obtain a first feature; Input the second mapping data into the convolutional network to obtain a second feature; wherein, the output results include the first feature and the second feature; Fuse the second feature with the core tensor, the first factor matrix, and the second factor matrix to obtain a first fusion feature; Add the first fusion feature to the first feature to obtain the target fusion feature.

3. The detection method for early fire according to claim 2, wherein When the scales of the first fusion feature and the first feature are not unified, the method further includes: Perform a zero-padding operation on the edge of the feature with a smaller scale among the first fusion feature and the first feature.

4. The early fire detection method according to claim 1, characterized in that, The method further includes: If a fire point is detected, determine the target fire-fighting equipment closest to the fire point based on the positioning data of the fire point, and upload the detection data to the control terminal; Wherein, the target fire-fighting equipment is the fire-fighting equipment closest to the fire point; Wherein, the detection data includes an approved control instruction for the target fire-fighting equipment and the positioning data of the fire point.

5. The detection method for early fire situation according to claim 1, wherein The convolutional network includes three convolutional layers.

6. The detection method for early fire as claimed in claim 1, wherein The method further includes: Obtain a first image of the area to be detected; Input the first image into an improved CSPNet model to obtain a first feature map; wherein, the improved CSPNet model uses a convolutional module based on regional attention; Use the detection head to detect the first feature map to detect smoke.

7. The detection method for early fire conditions according to claim 1, wherein The method further includes: Obtain all cameras in the coverage area; wherein, each camera captures an area to be detected; Construct a fire warning map based on all cameras in the coverage area; wherein, each camera in the fire warning map is used as a node, and an edge is established between two nodes with regional connectivity; Predict the fire spread direction based on the fire warning map.

8. The detection method for early fire situation according to claim 7, wherein Edges in the fire warning map are set with weights; The weight of an edge is determined by the connection type between two nodes, the physical distance between two nodes, and / or the flammability property of the material between two nodes.

9. The detection method for early fire situation according to claim 8, characterized in that, Predicting the fire spread direction based on the fire warning map includes: When a target node with a fire is detected, the fire spread direction is determined based on the weights of the edges between other nodes adjacent to the target node in the fire warning map.

10. An early fire detection system, characterized in that, Including: A first acquisition module, configured to acquire an image group captured by a camera for the area to be detected and form a target tensor; wherein, the image group includes multiple frames of images, and each frame of image includes an identified image background and a detection object; A second acquisition module, configured to acquire first mapping data and second mapping data; wherein, the first mapping data is the mapping of all image backgrounds in the image group in the target tensor, and the second mapping data is the mapping of all detection objects in the image group in the target tensor; A first processing module, configured to perform Tucker decomposition on the first mapping data to obtain a core tensor, an orthogonal first factor matrix, and a second factor matrix; A second processing module, configured to input the first mapping data and the second mapping data into a convolutional network respectively, and fuse the output results with the core tensor, the first factor matrix, and the second factor matrix to obtain a target fusion feature; A detection module, configured to detect the target fusion feature by using a detection head to detect and locate the fire point.

Images