Mountain fire flame contour recognition method and device based on improved Mask-RCNN architecture
By improving the Mask-RCNN architecture and combining multimodal feature fusion and boundary refinement, accurate extraction and height prediction of flame contours in complex environments are achieved, solving the problem of insufficient accuracy in flame recognition in existing technologies and improving the real-time performance and accuracy of wildfire monitoring.
Patent Information
- Application Number
- CN202510995692.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-18
- Publication Date
- 2025-11-21
AI Technical Summary
Existing wildfire flame recognition technologies lack the accuracy for extracting dynamic flame contours in complex environments, especially under conditions of smoke interference and changes in lighting, making it difficult to meet the needs of early warning.
An improved Mask-RCNN architecture is adopted, which combines residual network, feature pyramid network, multimodal feature fusion, region proposal network, boundary refinement and flame height prediction modules. Through multi-scale feature maps and optical flow estimation, the flame outline is accurately extracted and the height is predicted.
In complex scenarios, the flame contour segmentation accuracy is improved by more than 18%, the response latency is reduced to within 70ms, the flame height prediction error is less than 10%, and the adaptability and accuracy are significantly improved.
Smart Images
Figure CN120997524A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of computer vision, and particularly relates to a forest fire flame contour identification method and device based on an improved Mask-RCNN architecture, a storage medium and an electronic device. BACKGROUND
[0002] Forest fire monitoring is a core technology in the field of forest protection. Traditional methods rely on manual patrols or single sensors, which have the defects of poor real-time performance and high false alarm rate. With the development of computer vision technology, image-based forest fire detection has become a research hotspot. However, the existing technology still has bottlenecks in accurately extracting the dynamic contour of the flame, especially in complex environments (such as smoke interference and light changes), the coordination between flame height prediction and contour recognition is insufficient, which makes it difficult for the monitoring system to meet the early warning needs.
[0003] Existing forest fire flame identification technologies mainly include: traditional computer vision-based methods, such as color threshold segmentation combined with morphological operations, which rely on artificial features and have poor robustness; deep learning-based target detection methods, such as YOLO, Faster R-CNN, etc., which detect the flame as an ordinary target, but cannot provide accurate pixel-level contours; improved instance segmentation methods, such as the basic version of Mask-RCNN, which can generate target masks, but are not optimized for flame characteristics and lack height prediction capabilities. SUMMARY
[0004] The embodiments of the present application provide a forest fire flame contour identification method and device based on an improved Mask-RCNN architecture, a storage medium and an electronic device, which can improve the accuracy of forest fire flame contour extraction, improve environmental adaptability, and accurately predict the height of forest fire flames.
[0005] The embodiments of the present application provide a forest fire flame contour identification method based on an improved Mask-RCNN architecture, which comprises: Obtaining a forest fire RGB image and preprocessing the forest fire RGB image; Inputting the preprocessed forest fire RGB image into a trained forest fire flame contour identification model to obtain a flame contour mask and a forest fire flame height; wherein the forest fire flame contour identification model adopts an improved Mask-RCNN architecture.
[0006] Further, according to the forest fire flame contour identification method based on the improved Mask-RCNN architecture, the forest fire flame contour identification model comprises a backbone network, a multi-modal feature fusion module, a region proposal network, a boundary refinement module and a flame height prediction module; wherein the backbone network comprises a residual network and a feature pyramid network; the processing process of the forest fire flame contour identification model comprises: inputting the mountain fire RGB image into the residual network to obtain an initial feature map; inputting the mountain fire RGB image into the multi-modal feature fusion module to obtain a color feature map and a motion feature map; inputting the initial feature map, the HSV color space conversion result and the motion feature map into a feature pyramid network to obtain a multi-scale feature map; inputting the multi-scale feature map into the region proposal network to obtain a proposal region; aligning the proposal region and the multi-scale feature map to obtain a candidate flame region; inputting the candidate flame region into the boundary refinement module to obtain a refined flame contour mask; inputting the refined flame contour mask into the flame height prediction module to obtain a predicted mountain fire flame height.
[0007] Further, according to the above mountain fire flame contour identification method based on the improved Mask-RCNN architecture, wherein the multi-mountain fire RGB image is input into the multi-modal feature fusion module to obtain a color feature map and a motion feature map, comprising: convert the multi-mountain fire RGB image to an HSV color space and enhance it to obtain an HSV color space conversion result, and obtain a color feature map based on the HSV color space conversion result; perform optical flow estimation on the multi-mountain fire RGB image of adjacent frames, calculate the motion vector of each pixel between adjacent frames to obtain a motion feature map.
[0008] Further, according to the above mountain fire flame contour identification method based on the improved Mask-RCNN architecture, wherein the boundary refinement module includes a fully convolutional network, and the candidate flame region is input into the boundary refinement module to obtain a refined flame contour mask, comprising: input the candidate flame region into the fully convolutional network for multi-level convolution operation to obtain a refined flame contour mask.
[0009] Further, according to the above mountain fire flame contour identification method based on the improved Mask-RCNN architecture, wherein the refined flame contour mask is input into the flame height prediction module to obtain a predicted mountain fire flame height, comprising: based on an edge tracking algorithm, extract the connected region of the bright curve in the refined flame contour mask to obtain a curve contour; based on the curve contour, calculate the curve length to obtain the mountain fire flame height.
[0010] Further, according to the mountain fire flame contour recognition method based on the improved Mask-RCNN architecture, the curve length is calculated based on the curve contour, the mountain fire flame height is obtained, and the following formula is used for calculation:
[0011] wherein L is the flame length, (x i , y i ) and (x i+1 , y i+1 ) are two consecutive points on the curve, and N is the total number of points on the curve.
[0012] Further, according to the mountain fire flame contour recognition method based on the improved Mask-RCNN architecture, the method further comprises: training the mountain fire flame contour recognition model based on the total loss function to obtain a trained mountain fire flame contour recognition model; the total loss function is:
[0013] wherein,
[0014]
[0015]
[0016]
[0017]
[0018]
[0019] wherein, is the total loss function, is a balance coefficient, is a sub-loss, L cls is a classification loss, N cls is the number of classification samples, is a true label, is a model predicted flame probability, L box is a bounding box regression loss, N box is the number of bounding box samples, is a model predicted bounding box parameter, is a true bounding box parameter, L mask is a mask loss, HxW is the mask size is a real mask pixel value, is a model predicted mask probability, L hN is the number of samples, P is a model predicted flame probability, H is a real height measured by a laser rangefinder.
[0020] The embodiment of the present application also provides a forest fire flame contour identification device based on an improved Mask-RCNN architecture, comprising: An acquisition and preprocessing module is configured to acquire a forest fire RGB image and perform preprocessing on the forest fire RGB image. A forest fire flame contour identification module is configured to input the preprocessed forest fire RGB image into a trained forest fire flame contour identification model to obtain a flame contour mask and a forest fire flame height, wherein the forest fire flame contour identification model adopts an improved Mask-RCNN architecture.
[0021] The embodiment of the present application also provides a computer readable storage medium, wherein a plurality of instructions are stored in the computer readable storage medium, and the instructions are suitable for being loaded by a processor to execute any one of the forest fire flame contour identification methods based on the improved Mask-RCNN architecture.
[0022] The embodiment of the present application also provides an electronic device comprising a processor and a memory, wherein the processor is electrically connected with the memory, the memory is configured to store instructions and data, and the processor is configured to perform the steps in any one of the forest fire flame contour identification methods based on the improved Mask-RCNN architecture.
[0023] The forest fire flame contour identification method, device, storage medium and electronic device based on the improved Mask-RCNN architecture provided by the present application, the forest fire flame contour identification model is constructed through the improved Mask-RCNN architecture, the flame contour mask and the forest fire flame height are obtained based on the model, and the method has the following beneficial effects: (1) accurate contour extraction: through a boundary thinning module and multi-modal feature fusion, the flame contour segmentation precision (average IOU) in a complex scene is improved by more than 18% than that of a standard Mask-RCNN; (2) dynamic adaptability enhancement: a motion feature branch captures a flame boundary motion mode, and the response delay of a flame diffusion process is reduced to within 70 ms; (3) high prediction ability: non-contact estimation of the flame height is realized, the prediction error is less than 10%, and a quantitative basis is provided for fire grade evaluation; (4) improved generalization performance: a special data set covers multi-scene data, and the model maintains an average IOU greater than or equal to 0.85 in cross-season and cross-region tests. BRIEF DESCRIPTION OF DRAWINGS
[0024] The technical solutions and other beneficial effects of the present application will become apparent through the following detailed description of the specific embodiments of the present application in combination with the accompanying drawings.
[0025] Figure 1 A flowchart of a forest fire flame contour recognition method based on an improved Mask-RCNN architecture is provided for the embodiments of the present application.
[0026] Figure 2 A structural schematic diagram of a forest fire flame contour recognition model is provided for the embodiments of the present application.
[0027] Figure 3 The flame contour masks identified are provided for the embodiments of the present application, wherein (a)-(h) are respectively different candidate flame regions or corresponding flame contour masks identified by the forest fire flame.
[0028] Figure 4 A structural schematic diagram of a forest fire flame contour recognition device based on an improved Mask-RCNN architecture is provided for the embodiments of the present application.
[0029] Figure 5 A structural schematic diagram of an electronic device is provided for the embodiments of the present application. DETAILED DESCRIPTION
[0030] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.
[0031] The embodiments of the present application provide a forest fire flame contour recognition method, device, storage medium and electronic device based on an improved Mask-RCNN architecture. The forest fire flame contour recognition device based on an improved Mask-RCNN architecture provided by the embodiments of the present application can be integrated in an electronic device, which can be a terminal, a server or the like, wherein the terminal can include a tablet computer, a notebook computer, a personal computer (PC), a micro processing box or other devices, etc.
[0032] Please refer to Figure 1 , Figure 1 A flowchart of a forest fire flame contour recognition method based on an improved Mask-RCNN architecture is provided for the embodiments of the present application, which is applied in an electronic device. The forest fire flame contour recognition method based on an improved Mask-RCNN architecture includes the following steps: S1, acquiring a forest fire RGB image and pre-processing the forest fire RGB image; S2, input the pre-processed wildfire RGB image into the trained wild fire flame contour recognition model to obtain a flame contour mask and a wild fire flame height; wherein the wild fire flame contour recognition model adopts an improved Mask-RCNN architecture.
[0033] Figure 2 A structural schematic diagram of the wild fire flame contour recognition model provided by the embodiments of the present application is shown in FIG. 1, which includes a backbone network, a multi-modal feature fusion module, a region proposal network, a boundary refinement module and a flame height prediction module; wherein the backbone network includes a residual network and a feature pyramid network (i.e. ResNet-50 / 101-FPN). In an embodiment, step S2 includes: Figure 2 S21, input the wild fire RGB image into the residual network to obtain an initial feature map.
[0034] In the residual network, the wild fire RGB image is passed through a residual block and a convolutional layer to obtain an initial feature map.
[0035] S22, input the wild fire RGB image into the multi-modal feature fusion module to obtain a color feature map and a motion feature map.
[0036] In an embodiment, step S22 includes the following steps: S221, convert the multi-wild fire RGB image to an HSV color space and perform enhancement to obtain an HSV color space conversion result, and obtain a color feature map based on the HSV color space conversion result.
[0037] The input RGB image is converted to an HSV color space, and the specific hue range (e.g. H∈[0, 30] or [160, 180], S∈[50, 255], V∈[50, 255]) of the flame in the HSV space is used to enhance the distinguishability of the flame and the background, thereby highlighting the typical color features of the flame such as orange-red and yellow, and taking the three channels of HSV as color features.
[0038] S222, perform optical flow estimation on the multi-wild fire RGB images of adjacent frames, calculate the motion vector of each pixel between adjacent frames, and obtain a motion feature map.
[0039] The motion vector (including direction and speed) of each pixel between adjacent frames is calculated to generate a motion feature map. The optical flow estimation can use a traditional algorithm (e.g. Lucas-Kanade) or a deep learning model (e.g. FlowNet) to capture dynamic characteristics such as flame diffusion and shape fluctuation.
[0040] S23, input the initial feature map, the HSV color space conversion result and the motion feature map into a feature pyramid network to obtain a multi-scale feature map.
[0041] Through the top-down path and the lateral connection of the FPN, the semantic information in the color feature map (such as “flame color region”), the dynamic information of the motion feature map (such as “region boundary motion direction”) and the initial feature map are fused at different scales to form a composite feature containing “color-motion” double clues.
[0042] S24, input the multi-scale feature map into a region proposal network to obtain a proposal region.
[0043] Specifically, the multi-scale feature map is sequentially input into a 3x3 convolution layer and a 1x1 convolution layer to obtain a convolution feature, and the convolution feature is subjected to an activation function and a bounding box regression operation to obtain the proposal region.
[0044] S25, align the proposal region and the multi-scale feature map to obtain a candidate flame region. S26, input the candidate flame region into a boundary refinement module to obtain a refined flame contour mask.
[0045] Specifically, the candidate flame region is input into a fully convolutional network for multi-level convolution operation to obtain the refined flame contour mask. Figure 3 The flame contour mask identified by the embodiments of the present application, wherein (a)-(h) are respectively a candidate flame region or a corresponding flame contour mask identified by the identified flame contour mask of the different forest fire flames.
[0046] Meanwhile, the candidate flame region is input into a fully connected layer and subjected to a bounding box regression to obtain the bounding box coordinates of the candidate flame region, and the candidate flame region is input into a fully connected layer and subjected to an activation function to obtain the predicted category corresponding to the candidate flame region.
[0047] S27, input the refined flame contour mask into a flame height prediction module to obtain a predicted height of the forest fire flame.
[0048] In an embodiment, step S27 comprises the following steps: S271, based on an edge tracking algorithm, extract the connected region of the bright curve in the refined flame contour mask to obtain a curve contour.
[0049] Through the edge tracking algorithm, the connected region of the bright curve is extracted. A connected region marking algorithm can be used to identify the curve region, and the specific curve contour is extracted according to the marking.
[0050] S272, calculate the curve length based on the curve contour to obtain the height of the forest fire flame.
[0051] On the extracted curve profile, the length of the curve is calculated by digital integration method (such as accumulating the pixel distance of the edge points). The following formula can be used:
[0052] Where L is the length of the flame, (x i , y i ) and (x i+1 , y i+1 ) are two consecutive points on the curve, and N is the total number of points on the curve.
[0053] Given that the length of a certain reference in the picture is x m, and after image processing calculation, it is known that the gap length occupies N pixel points, so the actual length occupied by one pixel point is about (x / N) m.
[0054] The training process of the forest fire flame profile recognition model is introduced as follows: (1) Obtain the forest fire flame image dataset and the corresponding label.
[0055] (2) Input the forest fire flame image in the forest fire flame image dataset into the forest fire flame profile recognition model to obtain the flame profile mask.
[0056] (3) Based on the flame profile mask and the corresponding label, a total loss function is constructed, and the forest fire flame profile recognition model is iteratively trained based on the total loss function.
[0057] Where the total loss function is:
[0058] Where,
[0059]
[0060]
[0061]
[0062]
[0063]
[0064] Where, is the total loss function, is the balance coefficient, is the sub-loss, L cls is the classification loss, which is used to judge the classification error of whether the candidate region is a flame target, and the cross-entropy loss function is used, N cls is the number of classification samples, is the true label (1 represents flame, 0 represents background), is the model predicted flame probability, L box is the bounding box regression loss, used to optimize the position and size of the flame target bounding box, adopts Smooth L1 Loss, reduces the influence of outliers on training, N box is the number of bounding box samples, is the model predicted bounding box parameter (offset and scale), is the true bounding box parameter, L mask is the mask loss, used to generate the segmentation error of the flame pixel-level contour mask, adopts binary cross-entropy (BCE) loss, judges whether each pixel belongs to the flame area or not, HxW is the mask size, is the true mask pixel value, is the model predicted mask probability, L h is the height prediction loss, N is the number of samples, is the model predicted flame probability, is the true height measured by the laser range finder.
[0065] According to the method described in the above embodiment, this embodiment will be further described from the perspective of the wildfire flame contour recognition device based on the improved Mask-RCNN architecture. The wildfire flame contour recognition device based on the improved Mask-RCNN architecture can be implemented as an independent entity, or can be integrated in an electronic device, which can be a terminal, a server, or the like. The terminal can include a tablet computer, a notebook computer, a personal computer (PC), a micro processing box, or other devices, etc.
[0066] Please refer to Figure 4 , Figure 4 The wildfire flame contour recognition device based on the improved Mask-RCNN architecture provided by the embodiments of the present application is specifically described, which is applied in an electronic device. The wildfire flame contour recognition device based on the improved Mask-RCNN architecture can include: An acquisition and preprocessing module is configured to acquire a wildfire RGB image and pre-process the wildfire RGB image. A wildfire flame contour recognition module is configured to input the pre-processed wildfire RGB image into a trained wildfire flame contour recognition model to obtain a flame contour mask and a wildfire flame height. The wildfire flame contour recognition model adopts an improved Mask-RCNN architecture.
[0067] In specific implementation, the above various modules and / or units can be implemented as independent entities, or can be combined as the same or several entities, and the specific implementation of the above various modules and / or units can refer to the preceding method embodiments, and the beneficial effects that can be achieved can also refer to the beneficial effects in the preceding method embodiments, which will not be described here again.
[0068] In addition, the embodiments of the present application further provide an electronic device, which can be a computer, a tablet computer, or the like. The electronic device can implement the steps in any of the embodiments of the method for identifying a flame outline of a forest fire based on an improved Mask-RCNN architecture provided by the embodiments of the present application, and thus can achieve the beneficial effects of any of the methods for identifying a flame outline of a forest fire based on an improved Mask-RCNN architecture provided by the embodiments of the present application. Details are described in the preceding embodiments, which will not be described here again.
[0069] Figure 5 A specific structural block diagram of an electronic device provided by the embodiments of the present application is shown, which can be used to implement the method for identifying a flame outline of a forest fire based on an improved Mask-RCNN architecture provided in the above embodiments. The electronic device 500 can be a terminal, a server, or the like, wherein the terminal can include a tablet computer, a notebook computer, a personal computer (PC), a micro processing box, or other devices, and the like.
[0070] RF circuit 510 is used to receive and transmit electromagnetic waves, converting electromagnetic waves into electrical signals and vice versa, thereby enabling communication with communication networks or other devices. RF circuit 510 may include various existing circuit elements used to perform these functions, such as antennas, radio frequency transceivers, digital signal processors, encryption / decryption chips, subscriber identity modules (SIM cards), memory, etc. RF circuit 510 can communicate with various networks such as the Internet, corporate intranets, and wireless networks, or communicate with other devices via wireless networks. The aforementioned wireless networks may include cellular telephone networks, wireless local area networks (WLANs), or metropolitan area networks (MANs). The aforementioned wireless networks may use various communication standards, protocols, and technologies, including but not limited to Global System for Mobile Communication (GSM), Enhanced Data GSM Environment (EDGE), Wideband Code Division Multiple Access (WCDMA), Code Division Multiple Access (CDMA), Time Division Multiple Access (TDMA), Wireless Fidelity (Wi-Fi) (such as IEEE 802.11a, IEEE 802.11b, IEEE 802.11g, and / or IEEE 802.11n), Voice over Internet Protocol (VoIP), Worldwide Interoperability for Microwave Access (Wi-Max), other protocols for email, instant messaging, and short messages, and any other suitable communication protocols, including those that have not yet been developed.
[0071] The memory 520 can be used to store software programs and modules, such as the corresponding program instructions / modules in the above-described embodiments, and the processor 580 can execute various functions and data processing by running the software programs and modules stored in the memory 520, i.e., realize functions such as front camera shooting, processing of the shot image, and switching of display colors of the display content on the display screen. The memory 520 can include a high-speed random access memory, and can further include a non-volatile memory such as one or more magnetic storage devices, flash memories, or other non-volatile solid-state memories. In some examples, the memory 520 can further include memories disposed remotely with respect to the processor 580, which can be connected to the electronic device 500 through a network. Examples of the network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof.
[0072] The input unit 530 can be used to receive inputted digital or character information, and generate a keyboard, a mouse, and the like related to user settings and function control. The display unit 540 can be used to display information inputted by the user or provided to the user, and various graphical user interfaces which can be constituted by graphics, texts, icons, videos, and any combination thereof. The display unit 540 can include a display panel 541, which can be configured in the form of an LCD (Liquid Crystal Display), an OLED (Organic Light-Emitting Diode), or the like.
[0073] The audio circuit 560, the speaker 561, and the microphone 562 can provide an audio interface between the user and the electronic device 500. The audio circuit 560 can convert received audio data into an electrical signal, transmit the electrical signal to the speaker 561, and convert the electrical signal into a sound signal outputted by the speaker 561; on the other hand, the microphone 562 can convert a sound signal collected into an electrical signal, and the audio circuit 560 can convert the electrical signal into audio data, output the audio data to the processor 580 for processing, and then transmit the audio data to another terminal through the RF circuit 510, or output the audio data to the memory 520 for further processing. The audio circuit 560 can further include an earphone jack to provide communication between an external earphone and the electronic device 500.
[0074] The electronic device 500 can help the user to receive requests, send information, and the like through the transmission module 570 (e.g., a Wi-Fi module), which provides the user with wireless broadband Internet access. Although the transmission module 570 is shown, it can be understood that it does not belong to the essential components of the electronic device 500, and can be omitted as needed without changing the essence of the application.
[0075] The processor 580 is a control center of the electronic device 500, which connects various parts of the entire mobile phone through various interfaces and lines, and performs various functions of the electronic device 500 and processes data by running or executing software programs and / or modules stored in the memory 520 and calling data stored in the memory 520, thereby monitoring the entire electronic device. Optionally, the processor 580 can include one or more processing cores; in some embodiments, the processor 580 can integrate an application processor and a modem processor, wherein the application processor mainly processes the operating system, user interface, and application programs, and the modem processor mainly processes wireless communication. It can be understood that the above-mentioned modem processor can also not be integrated into the processor 580.
[0076] The electronic device 500 further includes a power supply 590 (such as a battery) for supplying power to various components, and in some embodiments, the power supply can be logically connected to the processor 580 through a power management system, so as to realize functions such as management of charging, discharging, and power consumption management through the power management system. The power supply 590 can also include one or more direct or alternating power supplies, recharging systems, power failure detection circuits, power converters or inverters, power state indicators, and any other components.
[0077] Although not shown, the electronic device 500 also includes a camera (such as a front camera, a rear camera), a Bluetooth module, and the like, which are not described here in detail. In particular, in the present embodiment, the display unit of the electronic device is a touch screen display, and the mobile terminal further includes a memory and one or more programs, wherein one or more programs are stored in the memory and configured to be executed by one or more processors, and the one or more programs include instructions for performing the following operations: Obtaining a forest fire RGB image and pre-processing the forest fire RGB image; Inputting the pre-processed forest fire RGB image into a trained forest fire flame contour recognition model to obtain a flame contour mask and a forest fire flame height; wherein the forest fire flame contour recognition model uses an improved Mask-RCNN architecture.
[0078] In specific implementation, each of the above modules can be implemented as an independent entity, or can be combined as the same or one or more entities, and the specific implementation of each of the above modules can be referred to the method embodiments described above, which will not be described here in detail.
[0079] Those skilled in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructions or by controlling relevant hardware by the instructions, which can be stored in a computer readable storage medium and loaded and executed by a processor. To this end, the embodiments of the present application provide a storage medium, which stores a plurality of instructions capable of being loaded by a processor to execute the steps of any one of the embodiments of the method for identifying a wild fire flame contour based on an improved Mask-RCNN architecture provided by the embodiments of the present application.
[0080] The computer readable storage medium can include a read only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.
[0081] Due to the instructions stored in the storage medium, the steps in any one of the embodiments of the method for identifying a wild fire flame contour based on an improved Mask-RCNN architecture provided by the embodiments of the present application can be executed, and thus the beneficial effects of any one of the embodiments of the method for identifying a wild fire flame contour based on an improved Mask-RCNN architecture provided by the embodiments of the present application can be achieved. Details are shown in the above embodiments, which will not be described here.
[0082] The method for identifying a wild fire flame contour based on an improved Mask-RCNN architecture, the device, the storage medium and the electronic device provided by the embodiments of the present application are described in detail above, and the principles and implementation manners of the present application are described by applying specific examples. The above embodiment is only used to help understand the method of the present application and its core idea; meanwhile, for those skilled in the art, according to the idea of the present application, the specific implementation manner and application range will be changed, and the above description should not be understood as a limitation of the present application.
Claims
1. A method for identifying a wild fire flame profile based on an improved Mask-RCNN architecture, characterized in that, The method comprises: obtaining a forest fire RGB image, and preprocessing the forest fire RGB image; inputting the preprocessed forest fire RGB image into a trained forest fire flame contour recognition model to obtain a flame contour mask and a forest fire flame height; wherein the forest fire flame contour recognition model adopts an improved Mask-RCNN architecture.
2. The forest fire flame contour identification method based on the improved Mask-RCNN architecture according to claim 1, characterized in that, The forest fire flame contour recognition model comprises a backbone network, a multi-modal feature fusion module, a region proposal network, a boundary refinement module, and a flame height prediction module; wherein the backbone network comprises a residual network and a feature pyramid network; The processing process of the forest fire flame contour recognition model comprises: inputting the forest fire RGB image into the residual network to obtain an initial feature map; inputting the forest fire RGB image into the multi-modal feature fusion module to obtain a color feature map and a motion feature map; inputting the initial feature map, the HSV color space conversion result, and the motion feature map into the feature pyramid network to obtain a multi-scale feature map; inputting the multi-scale feature map into the region proposal network to obtain a proposal region; aligning the proposal region and the multi-scale feature map to obtain a candidate flame region; inputting the candidate flame region into the boundary refinement module to obtain a refined flame contour mask; inputting the refined flame contour mask into the flame height prediction module to obtain a predicted forest fire flame height.
3. The forest fire flame contour identification method based on the improved Mask-RCNN architecture according to claim 2, characterized in that, Inputting the multi-forest fire RGB image into the multi-modal feature fusion module to obtain a color feature map and a motion feature map, comprising: converting the multi-forest fire RGB image to an HSV color space and enhancing it to obtain an HSV color space conversion result, and obtaining a color feature map based on the HSV color space conversion result; performing optical flow estimation on the multi-forest fire RGB images of adjacent frames to calculate the motion vector of each pixel between adjacent frames, and obtaining a motion feature map.
4. The forest fire flame contour identification method based on the improved Mask-RCNN architecture according to claim 2, characterized in that, The boundary refinement module comprises a fully convolutional network, and inputting the candidate flame region into the boundary refinement module to obtain a refined flame contour mask, comprising: inputting the candidate flame region into the fully convolutional network for multi-level convolution operation to obtain a refined flame contour mask.
5. The forest fire flame contour identification method based on the improved Mask-RCNN architecture according to claim 2, characterized in that, Inputting the refined flame contour mask into the flame height prediction module to obtain a predicted forest fire flame height, comprising: based on an edge tracking algorithm, extracting the connected region of the bright curve in the refined flame contour mask to obtain a curve contour; based on the curve contour, calculating the curve length to obtain the forest fire flame height.
6. The forest fire flame contour identification method based on the improved Mask-RCNN architecture according to claim 5, characterized in that, Based on the curve contour, the curve length is calculated to obtain the forest fire flame height, which is calculated by the following formula: wherein L is the flame length, x i , y i ) and (r, r) are two consecutive points on the curve, x i+1 , y i+1 ) are two consecutive points on the curve, N is the total number of points on the curve.
7. The forest fire flame contour identification method based on the improved Mask-RCNN architecture according to claim 1, characterized in that, The method further comprises: training the forest fire flame contour recognition model based on a total loss function to obtain a trained forest fire flame contour recognition model; the total loss function is: wherein, wherein, is a total loss function, is a balance coefficient, is a partial loss, L cls is a classification loss, N cls is a number of classification samples, is a true label, is a model predicted flame probability, L box is a bounding box regression loss, N box is a number of bounding box samples, is a model predicted bounding box parameter, is a true bounding box parameter, L mask is a mask loss, H × W is a mask size is a true mask pixel value, is a model predicted mask probability, L h is a height prediction loss, N is a number of samples, is a model predicted flame probability, is a true height measured by a laser range finder.
8. A wildfire flame perimeter identification device based on an improved Mask-RCNN architecture, characterized in that, comprises: an acquisition and preprocessing module for acquiring a forest fire RGB image and preprocessing the forest fire RGB image; The mountain fire flame contour identification module is configured to input the preprocessed mountain fire RGB image into a trained mountain fire flame contour identification model to obtain a flame contour mask and a mountain fire flame height.
9. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a plurality of instructions adapted to be loaded by a processor to perform the mountain fire flame contour identification method based on the improved Mask-RCNN architecture according to any one of claims 1 to 7.
10. An electronic device, comprising: The device comprises a processor and a memory, the processor is electrically connected with the memory, the memory is used for storing instructions and data, and the processor is used for executing the steps in the mountain fire flame contour identification method based on the improved Mask-RCNN architecture according to any one of claims 1 to 7.