Medical place fire monitoring method and device and storage medium

Abstract

The invention discloses a medical place fire monitoring method and device and a storage medium, and relates to the technical field of image recognition, and the method comprises the steps: obtaining a monitoring image in a medical place, carrying out the color simplification of the monitoring image through a local CPU service cluster, and obtaining a feature image corresponding to the monitoring image; uploading the feature picture to a GPU service cluster deployed at a cloud end, and identifying the feature picture through a pre-training neural network model configured in the GPU service cluster; and under the condition that the pre-trained neural network model recognizes that the fire features exist in the feature picture, executing a preset fire alarm service. According to the method and the device, load balancing is carried out on the GPU cluster through the low-cost CPU cluster, the overall data processing complexity of the system is reduced, and the fire behavior identification efficiency of the medical place is improved.

Description

Technical Field

[0001] The present application relates to the field of image recognition technology, and in particular to a fire monitoring method, device and storage medium for medical facilities. Background Art

[0002] Fire monitoring systems in medical facilities usually use deep learning technology to identify color surveillance images, especially using the YOLO (You Only Look Once, a target detection algorithm) series of algorithms, such as YOLOv3 and YOLOv8, to perform efficient flame detection. These algorithms can quickly and accurately detect flames from images, thereby achieving timely identification and response to fires. For example, the YOLOv3 algorithm is widely used in fire monitoring systems. The algorithm has excellent performance and can quickly and accurately detect fires in images, improving the efficiency and accuracy of fire detection. In addition, by using a large number of fire images to train the model and verifying it in different environments, it is shown that these models have good generalization capabilities and can adapt to various complex scenarios.

[0003] As the main computing resource for deep learning algorithms, GPUs (Graphics Processing Units) are relatively expensive. For fire monitoring systems in medical facilities, the amount of data that needs to be processed is usually very large, which leads to very high costs for purchasing and maintaining GPU hardware. If you choose to use cloud GPU services, although you can reduce hardware investment, it will take up a lot of bandwidth in the process of transmitting a large number of color images to the cloud for recognition. This not only increases the cost of data transmission, but also makes it difficult to meet the high-speed and real-time requirements of medical facilities for fire monitoring, and is likely to be affected by network stability and speed, thereby affecting the overall recognition efficiency of the fire monitoring system in medical facilities.

[0004] Therefore, how to achieve fire monitoring that takes into account both low computing cost and high recognition efficiency is a problem that needs to be solved urgently. Summary of the invention

[0005] The main purpose of this application is to provide an invention name, aiming to solve the technical problems corresponding to the background technology.

[0006] To achieve the above-mentioned purpose, the present application proposes a medical site fire monitoring method, which is applied to a fire monitoring system, wherein the fire monitoring system includes a central processing unit CPU service cluster deployed locally and an image processing unit GPU service cluster deployed in the cloud, and the medical site fire monitoring method includes: Acquire a monitoring picture in a medical setting, and perform color simplification processing on the monitoring picture through the CPU service cluster to obtain a feature picture corresponding to the monitoring picture; Uploading the feature picture to the GPU service cluster, and identifying the feature picture through a pre-trained neural network model configured in the GPU service cluster; When the pre-trained neural network model recognizes that there are fire features in the feature image, a preset fire alarm service is executed.

[0007] In one embodiment, the step of performing color simplification processing on the monitoring screen by the CPU service cluster to obtain a feature screen corresponding to the monitoring screen includes: For any pixel in the monitoring screen, the CPU service cluster performs color inversion processing on the pixel, wherein, during the color inversion processing, the pixel is converted into black or white based on a preset color threshold; After traversing each pixel, a feature picture corresponding to the monitoring picture is obtained.

[0008] In one embodiment, the step of performing color inversion processing on the pixel by the CPU service cluster includes: Comparing the color value of the pixel with a preset color threshold by the CPU service cluster; When the color value is greater than the preset color threshold, converting the color value into a white color value; When the color value is less than or equal to a preset color threshold, the color value is converted to a black color value.

[0009] In one embodiment, the step of performing color simplification processing on the monitoring screen by the CPU service cluster to obtain a feature screen corresponding to the monitoring screen includes: For any pixel in the monitoring screen, the CPU service cluster performs color inversion processing on the pixel, wherein, during the color inversion processing, the pixel is converted into black or white based on a preset color threshold; After traversing each pixel, a feature picture corresponding to the monitoring picture is obtained.

[0010] In one embodiment, the step of identifying the feature picture by using the pre-trained neural network model configured in the GPU service cluster includes: Recognize the color and shape of the feature picture through the pre-trained neural network model configured in the GPU service cluster; If the color value of any pixel group in the feature image is within a preset flame color value interval, and the shape of the pixel group conforms to the preset flame shape, the pixel group is determined to be a fire feature.

[0011] In one embodiment, the medical site fire monitoring method further includes: Acquire a collection of historical monitoring images in a medical setting, and synthesize a preset flame image into each historical monitoring image in the collection of historical monitoring images to obtain a candidate training data set; Performing color simplification processing on each candidate training picture in the candidate training data set according to a preset color threshold to obtain a training data set; The preset neural network model is trained based on the training data set to obtain the pre-trained neural network model.

[0012] In one embodiment, the step of synthesizing the preset flame image into each historical monitoring screen of the historical monitoring screen collection to obtain a candidate training data set includes: The preset flame image is synthesized into each historical monitoring screen in the collection of historical monitoring screens, and data enhancement is performed on each historical monitoring screen synthesized with the preset flame image to obtain a candidate training data set.

[0013] In one embodiment, the medical site fire monitoring method further includes: Performing edge detection on each historical monitoring screen in the candidate training data set to obtain an edge detection result corresponding to each historical monitoring screen; Determine the distribution characteristics of each edge detection result in the histogram of each historical monitoring screen, and select the peak area of ​​the preset flame feature in each distribution feature as the target color threshold range, wherein the preset flame feature is determined based on the preset flame image; The overlapping range of each target color threshold range is determined, and the middle value of the overlapping range is used as the threshold color threshold.

[0014] In one embodiment, the step of training a preset neural network model based on the training data set includes: Inputting the training data set into a preset neural network model, and monitoring the recognition result of the preset neural network model on the training data set; The model parameters in the preset neural network model are adjusted according to the recognition result.

[0015] In addition, to achieve the above-mentioned purpose, the present application also proposes an electronic device, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the medical facility fire monitoring method as described above.

[0016] In addition, to achieve the above-mentioned purpose, the present application also proposes a storage medium, which is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by the processor, the steps of the medical facility fire monitoring method as described above are implemented.

[0017] In addition, to achieve the above-mentioned purpose, the present application also provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, it implements the steps of the medical facility fire monitoring method as described above.

[0018] One or more technical solutions proposed in this application have at least the following technical effects: The present application first obtains the monitoring screen in the medical place, and performs color simplification processing on the monitoring screen through the CPU service cluster deployed locally to obtain the feature screen corresponding to the monitoring screen, so as to pre-execute the color simplification task of the screen through the low-cost CPU, and transfer the color simplification task from the GPU to the CPU, which effectively reduces the data volume of the image and reduces the computational complexity of subsequent image processing and analysis, significantly reduces the computational burden of the GPU, and reduces the network bandwidth occupied by subsequent data transmission; uploads the feature screen to the GPU service cluster deployed in the cloud, and recognizes the feature screen through the pre-trained neural network model configured in the GPU service cluster, so that on the basis of reducing hardware investment through cloud GPU service, the GPU can focus on efficiently processing the task it is best at processing: image recognition, and at the same time, due to the small data volume of the feature screen, the image recognition processing efficiency of the GPU is further improved; when the pre-trained neural network model recognizes that there are fire features in the feature screen, the preset fire alarm service is executed to ensure that the medical place can quickly take countermeasures by issuing a fire alarm in a timely manner, thereby improving the emergency response capability of fire accidents.

[0019] In summary, the present application pre-simplifies the colors of the surveillance images in a low-cost local CPU cluster, and then uploads the processed feature images to a high-performance cloud GPU cluster for fire identification and detection, thereby avoiding the high computing cost, excessive bandwidth usage, and untimely processing problems brought about by directly using the GPU to process large amounts of color image data. The low-cost CPU cluster is used to load balance the high-cost GPU cluster, thereby reducing the overall data processing complexity of the system, improving the efficiency of identifying fires in medical facilities, and thereby ensuring the timeliness of fire emergency response. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0021] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0022] Figure 1 A flow chart of the first embodiment of the fire monitoring method for medical facilities provided in the present application; Figure 2 A flow chart of Embodiment 2 of the fire monitoring method for medical facilities provided in this application; Figure 3 A schematic diagram of a brief process of a fire monitoring method for a medical facility provided in Example 2 of the present application; Figure 4 A schematic diagram of the model training process of the medical facility fire monitoring method provided in Example 2 of the present application; Figure 5 Schematic diagram of the equipment structure of the hardware operating environment involved in the medical facility fire monitoring method in the embodiment of the present application.

[0023] The purpose, features and advantages of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0024] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application and are not used to limit the present application.

[0025] In order to better understand the technical solution of the present application, a detailed description will be given below in conjunction with the accompanying drawings and specific implementation methods.

[0026] The main solution of the embodiment of the present application is: obtaining the monitoring screen in the medical field, and performing color simplification processing on the monitoring screen through the CPU service cluster deployed locally to obtain the feature screen corresponding to the monitoring screen; uploading the feature screen to the GPU service cluster deployed in the cloud, and identifying the feature screen through the pre-trained neural network model configured in the GPU service cluster; when the pre-trained neural network model recognizes the presence of fire features in the feature screen, executing the preset fire alarm service.

[0027] Since GPUs are the main computing resources for deep learning algorithms, they are relatively expensive. For fire monitoring systems in medical facilities, the amount of data that needs to be processed is usually very large, which leads to very high costs for purchasing and maintaining GPU hardware. If you choose to use cloud GPU services, although you can reduce hardware investment, a large amount of bandwidth will be occupied in the process of transmitting a large number of color pictures to the cloud for recognition. This not only increases the cost of data transmission, but also makes it difficult to meet the high-speed and real-time requirements of fire detection in medical facilities, and is likely to be affected by network stability and speed, thereby affecting the overall recognition efficiency of the fire monitoring system. Therefore, how to achieve fire monitoring that can take into account both low computing cost and high recognition efficiency is a problem that needs to be solved urgently.

[0028] The present application provides a solution, which pre-simplifies the color of the monitoring image in a low-cost local CPU cluster, and then uploads the processed feature image to a high-performance cloud GPU cluster for fire identification and detection, thereby avoiding the high computing cost, excessive bandwidth usage and untimely processing problems caused by directly using GPU to process a large amount of color image data, and realizes load balancing for the high-cost GPU cluster through the low-cost CPU cluster, thereby reducing the overall data processing complexity of the system, improving the efficiency of identifying fires in medical facilities, and ensuring the timeliness of fire emergency response.

[0029] It should be noted that the execution subject of this embodiment can be a computing service device with data processing, network communication and program running functions, such as a tablet computer, a personal computer, a mobile phone, etc., or an electronic device capable of realizing the above functions, a fire monitoring system, etc. The following takes the fire monitoring system as an example to illustrate this embodiment and the following embodiments.

[0030] Based on this, the present application embodiment provides a fire monitoring method for a medical facility, referring to Figure 1 , Figure 1 This is a flow chart of the first embodiment of the fire monitoring method for medical facilities of the present application.

[0031] In this embodiment, the fire monitoring method for medical sites is applied to a fire monitoring system, and the fire monitoring system includes a central processing unit CPU service cluster deployed locally and an image processing unit GPU service cluster deployed in the cloud. The fire monitoring method for medical sites includes steps S10 to S30: Step S10, obtaining a monitoring screen in a medical setting, and performing color simplification processing on the monitoring screen through the CPU service cluster to obtain a feature screen corresponding to the monitoring screen; It should be noted that color simplification processing refers to the process of converting a color image into a grayscale image or using other methods to reduce color information in order to reduce the complexity of the image; the feature image refers to an image after color simplification processing, which retains the key information in the original monitoring image that can be used for fire monitoring, but the amount of data is smaller than the original monitoring image.

[0032] In addition, it should be noted that the central processing unit (CPU) is a conventional server processing unit. It is inexpensive and can effectively perform some computing-intensive tasks, such as simple color simplification of images, and can support flexible expansion without large-scale changes to the entire system architecture. The image processing unit (GPU) has better performance than the CPU in image processing (such as image recognition and rendering), but is expensive.

[0033] It is understandable that due to the large amount of data in the color monitoring screen, the computational burden of GPU processing is high. When choosing to use cloud GPU services, a large number of color monitoring screens will also occupy a large amount of bandwidth resources, which not only increases the cost of data transmission, but also makes it difficult to meet the high-speed real-time requirements of the fire monitoring system. Therefore, performing step S10 can avoid the problem of using high-cost GPUs to directly process the color monitoring screen. By pre-simplifying the color of the monitoring screen in a low-cost local CPU cluster, the effect of load balancing for the GPU cluster through the CPU cluster is achieved, thereby reducing the data processing complexity and data cost of the subsequent system.

[0034] For example, high-definition surveillance cameras are installed at various key locations in medical facilities to capture live images in real time. These surveillance images are transmitted to the CPU service cluster deployed locally through the gateway in the form of digital signals. In the CPU service cluster, by executing image processing algorithms such as grayscale conversion or color quantization, the color surveillance images are converted into grayscale images or images with limited colors, thereby achieving color simplification processing. The processed image is the feature image, which retains the shape and dynamic information of the flame, but the data volume is smaller, which is convenient for subsequent processing.

[0035] In a feasible implementation manner, the step of performing color simplification processing on the monitoring screen by the CPU service cluster in step S10 to obtain a feature screen corresponding to the monitoring screen may include steps S11-S12: Step S11, for any pixel in the monitoring image, performing color inversion processing on the pixel through the CPU service cluster, wherein, during the color inversion processing, the pixel is converted into black or white based on a preset color threshold; It should be noted that the color inversion process refers to reversing the color value of each pixel in the image, usually by subtracting the current value of each color channel from the maximum value to obtain the inversion. In this step, the pixel color is converted to black or white based on the preset color threshold, that is, if the color value of the pixel is higher than the color threshold, the pixel will be converted to white, otherwise it will be black.

[0036] It is understandable that, since the color information in the color monitoring screen is usually not necessary for flame recognition and will increase the processing complexity, step S11 is performed to invert the color of each pixel, which can avoid processing high-dimensional color data, thereby reducing the consumption of computing resources in the subsequent recognition processing process, thereby simplifying the image data and improving the processing speed.

[0037] Exemplarily, each processor core in the CPU service cluster is assigned to process one or more pixels. For each pixel, the system reads its RGB value and calculates its comparison result with the preset color threshold. If the color value of the pixel, such as brightness (which can be obtained by calculating the weighted average of the RGB values), or the maximum value of the RGB value (i.e., the brightest color channel), is higher than the preset color threshold, the RGB value of the pixel is set to (255, 255, 255) white color value; if the color value is lower than the threshold, the RGB value of the pixel is set to (0, 0, 0) black color value. This process is implemented through parallel computing to ensure that the entire monitoring screen can complete the inversion processing quickly.

[0038] In a feasible implementation manner, the step of performing color inversion processing on the pixel by the CPU service cluster in step S11 may include steps S111 to S113: Step S111, comparing the color value of the pixel with a preset color threshold through the CPU service cluster; It can be understood that since it is necessary to simplify the complex color image data into binary image data that is easier to process so that the neural network model can learn and identify fire characteristics more quickly, step S111 is performed to avoid the problems of large computing resource consumption, slow processing speed and high model training complexity caused by processing high-dimensional color image data. By simplifying the image data into black and white binary images, the complexity of data processing is reduced, the training speed and recognition efficiency of the neural network are improved, and it also helps to improve the accuracy and real-time performance of the model in fire monitoring.

[0039] For example, a color threshold is first set, which is predetermined based on the general characteristics of flame color and the lighting conditions of the monitoring scene. During the processing, the system scans each pixel in the monitoring screen one by one and evaluates its RGB color value. For each pixel, the system compares its color value with the preset color threshold; conversely, if the maximum color value is less than or equal to the preset color threshold, the system sets the color value of the pixel to black (0, 0, 0).

[0040] Step S112, when the color value is greater than the preset color threshold, converting the color value into a white color value; Exemplarily, if the color value of a pixel, such as brightness (which can be obtained by calculating the weighted average of the RGB values), or the maximum value of the RGB values ​​(i.e., the brightest color channel), is greater than a preset color threshold, the system sets the color value of the pixel to a white color value (255, 255, 255).

[0041] Step S113: when the color value is less than or equal to a preset color threshold, convert the color value into a black color value.

[0042] Exemplarily, if the color value of a pixel, such as brightness (which can be obtained by calculating the weighted average of the RGB values), or the maximum value of the RGB values ​​(i.e., the brightest color channel), is less than or equal to a preset color threshold, the system sets the color value of the pixel to a black value (0, 0, 0).

[0043] In this embodiment, by adopting a binarization processing method based on a preset color threshold, the high computational cost and complexity of processing high-resolution color images and the impact of lighting changes on the accuracy of fire detection are avoided, thereby achieving the effect of simplifying image data, improving processing speed and reducing false alarm rate, so that the neural network model can identify fire characteristics more efficiently and accurately.

[0044] Step S12, after traversing each pixel, obtaining a feature picture corresponding to the monitoring picture.

[0045] It can be understood that after traversing all the pixels in the monitoring screen and inverting them, a new image, namely the feature screen, is obtained. The screen only contains black and white pixels and the color information in the original image is removed. Since it is necessary to extract key information from the original monitoring screen so that the neural network model can perform fire detection more effectively, step S12 is performed. By obtaining the feature screen, the problem of reduced detection accuracy due to interference from irrelevant color information in the image can be avoided, thereby achieving the effect of improving the accuracy of fire monitoring and reducing the false alarm rate.

[0046] Exemplarily, after completing the inversion processing of all pixels, the CPU service cluster reassembles the processed pixels into a new image, namely the feature picture. This feature picture is a binary image composed of black and white pixels, which retains the key shape and texture information of the flames and high temperature areas in the original monitoring picture, but removes other unnecessary color details. This feature picture is then stored in the memory and is ready to be uploaded to the GPU service cluster for further fire detection analysis. In this way, the system realizes the conversion from the original color monitoring picture to the simplified feature picture, providing image data that is more suitable for processing for the subsequent deep learning model.

[0047] In this implementation, by using a local CPU service cluster to perform parallel processing of inversion processing, the high computing cost and delay caused by the GPU directly processing complex color images are avoided, and the effects of simplifying image data, improving processing efficiency and reducing system resource consumption are achieved, thereby providing a more efficient and accurate processing basis for subsequent fire detection.

[0048] Step S20, uploading the feature picture to the GPU service cluster, and identifying the feature picture through a pre-trained neural network model configured in the GPU service cluster; It should be noted that the pre-trained neural network model refers to a deep learning model that has been trained on a large-scale image dataset, such as YOLOv3, Faster R-CNN, etc., which is specifically used to identify specific objects or patterns in images, such as flames.

[0049] It can be understood that since the GPU has a high degree of parallel processing capabilities and is suitable for executing computationally intensive tasks of deep learning neural network models, step S20 is performed. By uploading the feature image to the GPU service cluster in the cloud, the efficiency of image recognition is effectively guaranteed while effectively reducing the local hardware configuration cost, thereby achieving the effect of fast and accurate fire monitoring.

[0050] For example, the color-simplified feature image is transmitted to the GPU service cluster through a message queue or an http request. In the GPU service cluster, a pre-trained neural network model (such as YOLOv3) has been loaded and is ready for image recognition. When the feature image is input into the model, the model uses its trained weights and algorithms to quickly analyze the feature image in real time and identify the fire characteristics therein.

[0051] In a feasible implementation manner, the step of identifying the feature picture by using the pre-trained neural network model configured in the GPU service cluster in step S20 may include steps S21-S22: Step S21, identifying the color and shape of the feature picture through the pre-trained neural network model configured in the GPU service cluster; It can be understood that in order to utilize the advanced feature recognition capabilities of the neural network model to analyze the fire characteristics in the image, step S21 is performed, which can avoid the inaccuracy and limitations of traditional image processing methods in identifying fires in complex scenes, thereby achieving the effect of accurate identification and classification of fire characteristics.

[0052] Step S22: if the color value of any pixel group in the feature image is within a preset flame color value interval, and the shape of the pixel group conforms to a preset flame shape, then the pixel group is determined to be a fire feature.

[0053] It should be noted that the color value of a pixel group refers to the common color attribute of a group of adjacent pixels in a feature image, usually expressed as an RGB value; the shape of a pixel group refers to the geometric form or outline formed by these pixels in spatial arrangement.

[0054] It is understandable that since it is necessary to confirm whether the pixel group represents a fire based on the two dimensions of color and shape in order to reduce false alarms, step S22 is performed. The color value of the pixel group is in the preset flame color value range, which means that the color of these pixels matches the color characteristics of the flame, and the shape of the pixel group conforms to the preset flame shape, which indicates that the arrangement of these pixels forms a typical form of the flame. This can avoid the problem of misidentification caused by single feature judgment, such as misjudging lights or other bright objects as flames, thereby improving the accuracy and reliability of fire detection.

[0055] Exemplarily, in the recognition results output by the neural network model, the system will screen pixel groups based on preset flame color value intervals (such as a single color value (255, 255, 255) or a specific grayscale range) and flame shape features (such as the tip of the flame, the fluctuating edge, etc.). Specifically, the system will judge the color value of each pixel group, and if the color value falls within the preset flame color value interval, the shape of the pixel group will be further analyzed. This analysis can be achieved by calculating the geometric parameters of the pixel group, such as the outline, area, and perimeter, and matching them with the preset flame shape template. If both the color and shape meet the conditions, the system will mark the pixel group as a fire feature and prepare to trigger further alarm actions.

[0056] In this implementation, by using the pre-trained neural network model configured in the GPU service cluster to recognize the color and shape of the feature image, the inaccuracy and limitations of traditional image processing methods in identifying fire conditions in complex scenes are avoided, and false alarms and missed alarms caused by factors such as changes in ambient light, object occlusion, and background interference are reduced, thereby achieving accurate recognition of fire characteristics and quickly and accurately detecting fires.

[0057] Step S30, when the pre-trained neural network model recognizes that there are fire features in the feature image, a preset fire alarm service is executed.

[0058] It should be noted that fire characteristics refer to the visual features related to flames that can be recognized by the neural network model, such as the color, shape, flickering, etc. of the flame; the preset fire alarm service refers to the alarm and notification mechanism automatically triggered by the system once the model detects the fire characteristics.

[0059] It is understandable that in order to respond to the detected fire in a timely manner, step S30 is performed. By automatically executing the preset fire alarm service when the conditions are met, the risk of the fire not being handled in a timely manner can be avoided, thereby achieving a rapid emergency response and reducing the damage caused by the fire.

[0060] For example, once the pre-trained neural network model detects fire features in the feature image, the system immediately triggers the preset fire alarm service. This usually includes sending an alarm to the security control center of the medical site, activating the fire protection system, notifying personnel to evacuate through the voice broadcast system, and notifying relevant managers and emergency service agencies through SMS or email. The execution of these alarm services is automated, ensuring that quick action can be taken when a fire is detected to minimize the damage that the fire may cause.

[0061] This embodiment provides a method for monitoring fire in medical facilities. By pre-simplifying the color of the monitoring image in a low-cost local CPU cluster, and then uploading the processed feature image to a high-performance cloud GPU cluster for fire identification and detection, the high computing cost, excessive bandwidth usage, and untimely processing problems caused by directly using GPU to process a large amount of color image data are avoided. The low-cost CPU cluster is used to load balance the high-cost GPU cluster, thereby reducing the overall data processing complexity of the system, improving the efficiency of identifying fires in medical facilities, and ensuring the timeliness of fire emergency response.

[0062] Based on the first embodiment of the present application, in the second embodiment of the present application, the same or similar contents as those in the above-mentioned embodiment 1 can be referred to the above introduction, and will not be repeated in the following. Figure 2The medical site fire monitoring method further includes steps S01 to S03: Step S01, obtaining a collection of historical monitoring images in a medical setting, and synthesizing a preset flame image into each historical monitoring image in the collection of historical monitoring images to obtain a candidate training data set; It should be noted that the preset flame images refer to a series of images with representative flame shapes and colors, which are used to simulate real fire scenes; the candidate training data set refers to a set of image data used to train the neural network model after synthesizing the preset flame images into historical monitoring images.

[0063] It is understandable that, since real fire conditions are difficult to predict and simulate, but it is necessary to create a training data set containing fire scenes in order to train the neural network model to recognize fire conditions, performing step S01 can avoid the problem of insufficient model training due to lack of sufficient fire scene data, thereby providing the neural network model with diverse and real fire scene training data.

[0064] For example, first, historical monitoring images within a certain time range are extracted from the monitoring system database of the medical site to form a collection. Then, a series of preset flame images are synthesized into these historical monitoring images according to different positions, sizes and angles using image processing software to simulate various possible fire scenes. In this way, a candidate training data set containing real background and synthetic flames is created, providing a basis for the subsequent training process.

[0065] Step S02, performing color simplification processing on each candidate training picture in the candidate training data set according to a preset color threshold to obtain a training data set; It should be noted that the candidate training picture refers to each image in the candidate training data set, and the training data set refers to the final data set after color simplification processing, which is used for training the neural network model.

[0066] It is understandable that since it is necessary to reduce the complexity of the image data so that the neural network model can learn features more effectively, performing step S02 can avoid the problems of low training efficiency and overfitting caused by overly complex image data, thereby achieving the effect of improving model training efficiency and generalization ability.

[0067] For example, for each candidate training image in the candidate training data set, a programming language such as Python is used in combination with an image processing library (such as Pillow, OpenCV) to process the image pixel by pixel. The RGB value of each pixel is converted to a grayscale value, or the color value of the pixel is mapped to a specific color range according to the color characteristics of the flame, for example, the pixel close to the flame color is marked as black or white. In this way, all pixels in the entire data set are traversed and processed, and finally a color-simplified training data set is obtained.

[0068] Step S03: training a preset neural network model based on the training data set to obtain the pre-trained neural network model.

[0069] It is understandable that since the neural network model needs to be trained to be able to recognize and distinguish fire characteristics, performing step S03 can avoid the problem of using an untrained model to accurately identify fire in real scenarios, thereby achieving the effect of efficient recognition and accurate classification of fire characteristics through the neural network model.

[0070] Exemplarily, a deep learning framework (such as TensorFlow or PyTorch) is used to build and train the neural network model. The training data set is input into the neural network, and the model parameters such as the network weights are adjusted through multiple iterations to minimize the classification error. In this process, the back propagation algorithm and optimizer (such as Adam or SGD) are used to optimize the model parameters. After the training is completed, the obtained pre-trained neural network model can recognize and distinguish the characteristics of flames and other non-flame scenes, so as to be used in actual fire monitoring applications.

[0071] In a feasible implementation manner, the step of training the preset neural network model based on the training data set in step S03 may include steps S031-S032: Step S031, inputting the training data set into a preset neural network model, and monitoring the recognition results of the preset neural network model on the training data set, which generally include the classification label (e.g., whether it is a fire situation) and / or confidence score of the model for each input sample; It should be noted that the recognition result refers to the output of the neural network model after classification or regression prediction of the input training data set, that is, the model's judgment on whether each training sample is a fire.

[0072] It is understandable that in order to evaluate and verify the neural network model's ability to identify fire characteristics and to ensure that the model can accurately detect fires in practical applications, step S031 is performed. This can prevent the model from being deployed and used without sufficient training or verification, thereby reducing problems such as missed alerts or false alerts due to poor model performance, thereby ensuring that the model can achieve the expected recognition accuracy when processing actual monitoring images.

[0073] For example, the preprocessed and labeled training data set is first loaded into the input layer of the neural network model. The training data set contains a large number of fire images, and each image is marked with a corresponding tag. Then, the training process of the model is started, and the neural network calculates the output through forward propagation and updates the weights through the back-propagation algorithm. In this process, cross-validation and other techniques are used to monitor the performance of the model on the validation set, and key indicators such as the recognition accuracy of the model for the training data set and the loss function value are recorded to evaluate the training progress and performance of the model.

[0074] Step S032: adjusting the model parameters in the preset neural network model according to the recognition result.

[0075] It should be noted that model parameters refer to internal variables learned during the network training process, which determine how the network processes input data and the output results, including but not limited to: weights, which are adjustable values ​​connecting two adjacent layers of neurons in a neural network, which determine the intensity of information transmission; and biases, which are fixed values ​​added to the activation function of each neuron in the neural network, which determine the threshold of the activation function.

[0076] In addition, it should be noted that the user can also adjust the hyperparameters in the preset neural network model according to the recognition results, wherein the hyperparameters are a set of parameters set before training the neural network model, including learning rate, number of iterations, batch size, etc.

[0077] It is understandable that since the performance of the neural network model is often affected by the hyperparameter setting, it is necessary to optimize the model performance by adjusting the hyperparameters. Therefore, performing step S032 can avoid problems such as model overfitting or underfitting caused by improper hyperparameter settings, thereby improving the generalization ability of the model and making the model have higher accuracy and lower false alarm rate in fire monitoring.

[0078] For example, the training data set is input into the neural network model, and the recognition result is obtained through forward propagation; then, the loss function value between the recognition result and the true label is calculated, such as cross entropy loss or mean square error; then, the gradient of the loss function value with respect to the model parameters is calculated through the back propagation algorithm; finally, the model parameters are updated using the gradient descent algorithm for the calculated gradient to reduce the loss function value and improve the prediction accuracy of the model. This process will be iterated many times until the model performance reaches the preset standard or the training reaches a certain number of iterations.

[0079] In this embodiment, by performing neural network training and hyperparameter optimization, the problems of overfitting, underfitting and inefficient learning in the model training process are avoided, thereby improving the accuracy of the neural network model in the fire identification task, reducing the false alarm rate, and improving the generalization ability of the model.

[0080] In this embodiment, by using synthetic flame images to create candidate training data sets, color simplification processing is performed on training pixels to generate data sets suitable for training, and these data sets are used to train neural network models, the problems of insufficient model training due to lack of real fire scene data and the impact of high complexity of image data on training efficiency and model generalization ability are avoided, thereby achieving the effect of improving the recognition accuracy, training efficiency and actual application performance of the neural network model in fire monitoring.

[0081] In a feasible implementation manner, after the step of synthesizing the preset flame image into each historical monitoring screen of the historical monitoring screen collection in step S01, step S011 may also be included: Step S011 , synthesizing a preset flame image into each historical monitoring screen in the collection of historical monitoring screens, and performing data enhancement on each historical monitoring screen synthesized with the preset flame image to obtain a candidate training data set.

[0082] It should be noted that data enhancement refers to increasing the diversity and quantity of training data by performing a series of transformations on historical surveillance images, such as lateral shift, central axis flip, mirroring, rotation, cropping, color adjustment, etc., thereby improving the generalization ability of the neural network model.

[0083] It is understandable that since the original monitoring images usually have limited data volume or insufficient scene diversity, this will cause the neural network model to be unable to fully learn various possible fire characteristics during training. Therefore, step S011 is performed. Through data enhancement, the problem of model overfitting caused by the uniformity of training data can be avoided, that is, the model performs well on the training data but performs poorly in unseen real scenes, thereby improving the robustness and accuracy of the model in the fire identification task, because the model can better identify and generalize to different fire scenes by learning more diverse data, thereby more effectively detecting fires in practical applications.

[0084] For example, first, any historical monitoring picture is selected from the historical monitoring picture library of synthesized preset flame images; then, a series of data enhancement operations are performed on the historical monitoring picture, including but not limited to the following steps: 1) randomly rotate the image to simulate flame observation at different angles; 2) flip the image horizontally and vertically to increase the diversity of the flame; 3) adjust the brightness, contrast and saturation of the image to simulate flames under different lighting conditions; 4) add random noise to the image to simulate image quality problems that may occur in monitoring equipment; 5) crop and scale the image to simulate flames in different sizes and positions. Through these data enhancement operations, each historical monitoring picture can generate multiple variants. After traversing all historical monitoring pictures, these enhanced image sets are used as candidate training data sets for subsequent training of neural network models.

[0085] In this implementation, by adopting data enhancement methods, the problems of model overfitting and poor generalization ability caused by insufficient training data or a single scenario are avoided, and the robustness and accuracy of the neural network model used for fire identification are improved, ensuring that the model can effectively identify fires under different monitoring environments and conditions.

[0086] In a feasible implementation manner, the medical site fire monitoring method may further include steps S100 to S300: Step S100, performing edge detection on each historical monitoring screen in the candidate training data set to obtain an edge detection result corresponding to each historical monitoring screen; It should be noted that edge detection refers to the use of image processing technology to identify points in an image where brightness changes significantly. These points usually mark the outline of an object or the boundary of a shape. The edge detection result refers to the image after edge detection processing, which only contains the edge information of the object and is usually represented in the form of a binary image, in which the edge part is white and the non-edge part is black.

[0087] It is understandable that in order to extract the contour information of the flame from the candidate training data set, step S100 is performed to avoid problems such as misjudgment of the shape during flame recognition due to the changeable and irregular flame shape. Through edge detection, accurate recognition of the flame contour can be achieved, thereby improving the accuracy of flame detection.

[0088] Exemplarily, first, a historical surveillance picture is selected from the candidate training data set; then, the Canny edge detection algorithm is applied for processing, and the algorithm includes the following steps: 1) Use a Gaussian filter to smooth the image to reduce the influence of noise; 2) Calculate the gradient amplitude and direction of each pixel in the image; 3) Apply non-maximum suppression to refine the edge; 4) Use a dual threshold algorithm to detect and connect the edges, and finally obtain a binary image containing the image edge information, that is, the edge detection result.

[0089] Step S200, determining the distribution characteristics of each edge detection result in the histogram of each historical monitoring screen, and selecting the peak area of ​​the preset flame feature in each distribution feature as the target color threshold range, wherein the preset flame feature is determined based on the preset flame image; It should be noted that the distribution feature refers to the pixel intensity distribution of the edge detection result in the image histogram. This feature describes the distribution of the number of pixels of different grayscale levels in the edge detection result image; the peak area refers to the area in the image histogram where the grayscale level representing the color intensity of the flame feature appears more frequently. This area appears as a significant "peak" in the histogram, indicating the main distribution range of the flame color in the image; the target color threshold range refers to the color threshold range determined according to the peak area of ​​the flame feature in the histogram distribution feature. This range is used for subsequent color simplification processing to distinguish between flame pixels and non-flame pixels.

[0090] It can be understood that in order to avoid the problem of inaccurate color threshold setting due to the similarity between the flame color and the background color, step S200 is performed. By analyzing the histogram and edge detection results, a suitable color threshold range can be determined, thereby achieving the effect of accurately distinguishing the flame color from the background color, reducing false alarms and missed alarms, and thus improving the reliability of the fire monitoring system.

[0091] Exemplarily, the color histogram of each surveillance image processed by edge detection is first calculated, which involves counting the distribution of pixel values ​​in each color channel of the image; then, these histograms are analyzed to identify peak areas that match preset flame features. The preset flame features are determined by analyzing a set of images known to contain flames (i.e., preset flame images), which are used to calibrate the general range and intensity of flame color. By comparing the peaks in the histogram with these preset flame features, the peak area in the histogram that best matches the flame color features is selected as the target color threshold range.

[0092] Step S300, determining the overlap range of each target color threshold range, and taking the middle value of the overlap range as the threshold color threshold.

[0093] It can be understood that in order to find a universal color threshold that can be widely applied to different scenes and lighting conditions, step S300 is performed. By calculating the middle value of the threshold overlap range as the final color threshold, the problem of inaccurate threshold setting due to scene diversity and changes in lighting conditions can be avoided, thereby improving the adaptability and accuracy of the fire monitoring system in different environments.

[0094] Exemplarily, the target color threshold ranges determined from different historical monitoring images are compared to find the overlapped parts in these ranges. This process can be achieved by calculating the minimum and maximum values ​​of all target color threshold ranges, thereby determining a numerical interval common to all ranges. Then, the numerical median of this overlapping interval is determined, and the median represents a comprehensive index of the color threshold, which can better reflect the general characteristics of the flame color. Finally, this median is set as the threshold color threshold for the subsequent color simplification process.

[0095] In this implementation, the color threshold for color simplification processing is determined by combining edge detection algorithm and histogram analysis, thereby avoiding the problem of inaccurate flame recognition due to similarity between flame and background color or image noise, thereby improving the accuracy and robustness of the fire monitoring system, ensuring that the system can effectively identify flames in complex and changeable monitoring environments, thereby achieving the effect of reducing false alarms and missed alarms and improving fire warning efficiency.

[0096] For example, in order to help understand the implementation process of the medical place fire monitoring method obtained by combining this embodiment with the above-mentioned embodiment 1, please refer to Figure 3 , Figure 3 A brief flow chart of a fire monitoring method for medical facilities is provided, specifically: First, the monitoring images transmitted by multiple cameras are obtained through the gateway, and then the monitoring images are pre-simplified in color through the CPU service cluster deployed locally to obtain the feature images corresponding to the monitoring images. The feature images are then uploaded to the GPU service cluster deployed in the cloud via the Internet, so that the uploaded feature images can be recognized through the pre-trained neural network model configured in the GPU service cluster. When the fire characteristics are recognized in the feature images, the preset alarm service is executed.

[0097] For further information, please refer to Figure 4 , Figure 4 A model training flow diagram of a fire monitoring method for medical facilities is provided, specifically: First, obtain hospital image samples (such as a collection of historical surveillance images in medical facilities), synthesize the preset flame image into the sample, and perform data enhancement on the synthesized sample to obtain a candidate training data set. Then, manually adjust or analyze the edge detection results and histograms of the candidate training data set to determine a relatively appropriate color threshold, and then invert the candidate training data set according to the color threshold to obtain an output sample set (i.e., training data set). After inverting, the color threshold can be adjusted again by inspecting the data set. Finally, the preset neural network model is trained with the output sample set to obtain a pre-trained neural network model.

[0098] It should be noted that the above examples are only used to understand the present application and do not constitute a limitation on the fire monitoring method for medical facilities of the present application. More simple transformations based on this technical concept are all within the scope of protection of the present application.

[0099] The present application provides an electronic device, which includes: at least one processor; and a memory that is communicatively connected to the at least one processor; wherein the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the medical facility fire monitoring method in the above-mentioned embodiment 1.

[0100] Reference below Figure 5, which shows a schematic diagram of the structure of an electronic device suitable for implementing the embodiments of the present application. The electronic devices in the embodiments of the present application may include but are not limited to mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions: tablet computers), PMPs (Portable Media Players: portable multimedia players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 5 The electronic device shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0101] like Figure 5 As shown, the electronic device may include a processing device 1001 (e.g., a central processing unit, a graphics processor, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM: Read Only Memory) 1002 or a program loaded from a storage device 1003 to a random access memory (RAM: Random Access Memory) 1004. In RAM1004, various programs and data required for the operation of the electronic device are also stored. The processing device 1001, ROM1002, and RAM1004 are connected to each other through a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Generally, the following systems can be connected to the I / O interface 1006: an input device 1007 including, for example, a touch screen, a touch pad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; an output device 1008 including, for example, a liquid crystal display (LCD: Liquid Crystal Display), a speaker, a vibrator, etc.; a storage device 1003 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 1009. The communication device 1009 can allow the electronic device to communicate with other devices wirelessly or by wire to exchange data. Although the figure shows an electronic device with various systems, it should be understood that it is not required to implement or have all the systems shown. More or fewer systems can be implemented or have alternatively.

[0102] In particular, according to the embodiments disclosed in the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device, or installed from a storage device 1003, or installed from a ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment disclosed in the present application are executed.

[0103] The electronic device provided by the present application adopts the medical site fire monitoring method in the above embodiment, which can solve the technical problem of how to achieve fire monitoring that can take into account both low computing cost and high recognition efficiency. Compared with the prior art, the beneficial effects of the electronic device provided by the present application are the same as the beneficial effects of the medical site fire monitoring method provided by the above embodiment, and the other technical features in the electronic device are the same as the features disclosed in the method of the previous embodiment, which will not be repeated here.

[0104] It should be understood that the various parts disclosed in this application can be implemented by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0105] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0106] The present application provides a computer-readable storage medium having computer-readable program instructions (ie, computer programs) stored thereon, and the computer-readable program instructions are used to execute the medical facility fire monitoring method in the above-mentioned embodiment.

[0107] The computer-readable storage medium provided in the present application may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems or devices, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM: Random Access Memory), a read-only memory (ROM: Read Only Memory), an erasable programmable read-only memory (EPROM: Erasable Programmable Read Only Memory or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM: CD-Read Only Memory), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system or device. The program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (Radio Frequency: Radio Frequency), etc., or any suitable combination of the above.

[0108] The computer-readable storage medium may be included in the electronic device, or may exist independently without being installed in the electronic device.

[0109] The computer-readable storage medium carries one or more programs. When the one or more programs are executed by an electronic device, the electronic device: obtains the monitoring screen in the medical field, and performs color simplification processing on the monitoring screen through a CPU service cluster deployed locally to obtain a feature screen corresponding to the monitoring screen; uploads the feature screen to a GPU service cluster deployed in the cloud, and identifies the feature screen through a pre-trained neural network model configured in the GPU service cluster; and executes a preset fire alarm service when the pre-trained neural network model identifies the presence of fire features in the feature screen.

[0110] Computer program code for performing the operations of the present application may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0111] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present application. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0112] The modules involved in the embodiments described in this application may be implemented by software or hardware, wherein the name of the module does not constitute a limitation on the unit itself in some cases.

[0113] The readable storage medium provided in this application is a computer-readable storage medium, which stores computer-readable program instructions (i.e., computer programs) for executing the above-mentioned medical site fire monitoring method, and can solve the technical problem of how to achieve fire monitoring that can take into account both low computing cost and high recognition efficiency. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as the beneficial effects of the medical site fire monitoring method provided in the above-mentioned embodiment, and will not be repeated here.

[0114] The present application also provides a computer program product, including a computer program, which implements the steps of the medical facility fire monitoring method as described above when executed by a processor.

[0115] The computer program product provided by the present application can solve the technical problem of how to achieve fire monitoring that can take into account both low computing cost and high recognition efficiency. Compared with the prior art, the beneficial effects of the computer program product provided by the present application are the same as the beneficial effects of the medical place fire monitoring method provided by the above embodiment, which will not be repeated here.

[0116] The above descriptions are only some embodiments of the present application, and are not intended to limit the patent scope of the present application. All equivalent structural changes made using the contents of the present application specification and drawings under the technical concept of the present application, or direct / indirect applications in other related technical fields are included in the patent protection scope of the present application.

Claims

1. A fire monitoring method for a medical facility, characterized in that: Applied to a fire monitoring system, the fire monitoring system includes a central processing unit CPU service cluster deployed locally and an image processing unit GPU service cluster deployed in the cloud. The medical site fire monitoring method includes: Acquire a monitoring picture in a medical setting, and perform color simplification processing on the monitoring picture through the CPU service cluster to obtain a feature picture corresponding to the monitoring picture; Uploading the feature picture to the GPU service cluster, and identifying the feature picture through a pre-trained neural network model configured in the GPU service cluster; When the pre-trained neural network model recognizes that there are fire features in the feature image, a preset fire alarm service is executed.

2. The method for monitoring fire in a medical setting according to claim 1, wherein: The step of performing color simplification processing on the monitoring picture by the CPU service cluster to obtain a feature picture corresponding to the monitoring picture comprises: For any pixel in the monitoring screen, the CPU service cluster performs color inversion processing on the pixel, wherein, during the color inversion processing, the pixel is converted into black or white based on a preset color threshold; After traversing each pixel, a feature picture corresponding to the monitoring picture is obtained.

3. The method for monitoring fire in a medical setting according to claim 2, wherein: The step of performing color inversion processing on the pixel by the CPU service cluster comprises: Comparing the color value of the pixel with a preset color threshold by the CPU service cluster; When the color value is greater than the preset color threshold, converting the color value into a white color value; When the color value is less than or equal to a preset color threshold, the color value is converted to a black color value.

4. The method for monitoring fire in a medical setting according to claim 1, wherein: The step of identifying the feature picture by using the pre-trained neural network model configured in the GPU service cluster includes: Recognize the color and shape of the feature picture through the pre-trained neural network model configured in the GPU service cluster; If the color value of any pixel group in the feature image is within a preset flame color value interval, and the shape of the pixel group conforms to the preset flame shape, the pixel group is determined to be a fire feature.

5. The method for monitoring fire in a medical setting according to claim 1, wherein: The medical site fire monitoring method further comprises: Acquire a collection of historical monitoring images in a medical setting, and synthesize a preset flame image into each historical monitoring image in the collection of historical monitoring images to obtain a candidate training data set; Performing color simplification processing on each candidate training picture in the candidate training data set according to a preset color threshold to obtain a training data set; The preset neural network model is trained based on the training data set to obtain the pre-trained neural network model.

6. The method for monitoring fire in a medical setting according to claim 5, characterized in that: The step of synthesizing the preset flame image into each historical monitoring screen of the historical monitoring screen collection to obtain a candidate training data set includes: The preset flame image is synthesized into each historical monitoring screen in the collection of historical monitoring screens, and data enhancement is performed on each historical monitoring screen synthesized with the preset flame image to obtain a candidate training data set.

7. The method for monitoring fire in a medical setting according to claim 5, characterized in that: The medical site fire monitoring method further comprises: Performing edge detection on each historical monitoring screen in the candidate training data set to obtain an edge detection result corresponding to each historical monitoring screen; Determine the distribution characteristics of each edge detection result in the histogram of each historical monitoring screen, and select the peak area of ​​the preset flame feature in each distribution feature as the target color threshold range, wherein the preset flame feature is determined based on the preset flame image; The overlapping range of each target color threshold range is determined, and the middle value of the overlapping range is used as the threshold color threshold.

8. The method for monitoring fire in a medical setting according to claim 5, characterized in that: The step of training the preset neural network model based on the training data set includes: Inputting the training data set into a preset neural network model, and monitoring the recognition result of the preset neural network model on the training data set; The model parameters in the preset neural network model are adjusted according to the recognition result.

9. An electronic device, characterized in that: The device comprises: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the medical facility fire monitoring method according to any one of claims 1 to 8.

10. A storage medium, characterized in that: The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the medical facility fire monitoring method according to any one of claims 1 to 8 are implemented.

Images