Fire smoke early recognition system and method based on deep learning

Abstract

The invention discloses a fire smoke early recognition system and method based on deep learning, and relates to the technical field of fire smoke early recognition, and the method comprises the steps: obtaining image or video data in real time through a front-end data collection module, and transmitting the image or video data to a rear-end processing unit; the method comprises the following steps: marking various fire smoke samples through a deep learning training model module, training a deep learning model, learning basic morphological characteristics of smoke, and constructing a fire smoke detection model; the real-time smoke recognition module integrates a fire smoke detection model, analyzes image or video data acquired in real time, quickly recognizes early fire smoke, and gives an alarm through the alarm feedback module to guide related personnel to take actions. Therefore, by adopting the fire smoke early recognition method based on deep learning, the fire smoke can be monitored in real time, the recognition accuracy of the model is improved, the adaptability of the model to different environmental conditions is enhanced, and the recognition of the fire smoke under a complex background is met.

Description

Technical Field

[0001] The present invention relates to the technical field of early fire smoke identification, and in particular to a system and method for early fire smoke identification based on deep learning. Background Art

[0002] Fire is a serious safety accident. Early detection and timely measures can greatly reduce casualties and property losses. Traditional fire detection technology mainly relies on hardware devices such as temperature sensors and smoke sensors to detect fires. Although it can play a warning role to a certain extent, the response speed and accuracy in complex environments still need to be improved. Especially in the case of low smoke concentration in the initial stage, traditional sensors may have false alarms or missed alarms.

[0003] In recent years, with the development of artificial intelligence technology, especially the advancement of deep learning algorithms, new possibilities have been provided for early identification of fires. Existing fire detection systems based on image processing and video surveillance have begun to try to use deep learning technology to analyze smoke features in images or video data captured by cameras for early warning of fires. However, the existing solutions still have some shortcomings, such as poor adaptability to different environmental conditions, sensitivity to changes in lighting, and insufficiently comprehensive data sets for training models.

[0004] In addition, some existing deep learning models also have a contradiction between real-time performance and computing resource requirements. In other words, in order to achieve high recognition accuracy, the model is often more complex and requires a large amount of computing, which may be difficult to deploy and run in some embedded devices or resource-constrained scenarios. The current image early fire alarm system uploads the video stream to the cloud through the network, performs image analysis through the server, and combines machine learning algorithms with various deep learning algorithms to improve the accuracy of the model in detecting flames and smoke. However, the simultaneous processing of multiple algorithms will make the algorithm larger in size, and the hardware requirements for the processing equipment are higher. For terminal devices with poor network conditions, small network bandwidth, and large video stream transmission delay, it is difficult to transfer the model to the edge computing camera for real-time recognition.

[0005] Therefore, the present invention constructs a smoke detection model based on the latest YOLO series target detection algorithm, and develops a deep neural network architecture suitable for complex and low-light environments by adopting a more lightweight network structure and efficient reasoning technology to improve the accuracy and real-time performance of fire warning. Summary of the invention

[0006] The purpose of the present invention is to provide a fire smoke early identification system and method based on deep learning, so as to realize real-time detection of smoke in monitoring video streams, solve the problems of low recognition rate, slow alarm response speed and susceptibility to interference from environmental factors in the prior art fire detection means in complex environments, and improve the effect of fire smoke early identification technology.

[0007] To achieve the above object, the present invention provides a method for early identification of fire smoke based on deep learning, comprising the following steps: S1, obtain smoke image or video data in real time through the front-end data acquisition module and transmit it to the back-end processing unit; S2. Label various fire smoke samples, expand the data set through data enhancement technology, introduce the improved bidirectional feature pyramid network BIMAFPN module and SimAM attention mechanism, build an improved YOLO V8 deep learning network model, and then use the expanded data set for model training; S3. After the model is verified, it is integrated into the fire smoke detection model as a real-time smoke recognition module to analyze the real-time acquired image or video data and quickly identify early fire smoke; S4. When fire smoke is identified, the system automatically triggers the alarm mechanism, issues an alarm through the alarm feedback module, and guides relevant personnel to take action.

[0008] Preferably, labeling various types of fire smoke samples includes collecting smoke image data under various scenes and environmental conditions, and labeling the smoke morphology of various fire types.

[0009] Preferably, the data enhancement technology includes image rotation, random cropping, elastic deformation, perspective transformation and application of Gaussian noise, so as to obtain fire smoke characteristics in various scenes.

[0010] Preferably, using the expanded data set for model training includes: Firstly, the expanded dataset is divided into training set, test set and validation set. Then, the divided smoke image training set is used to train the deep learning network model based on the improved YOLO V8 to obtain the trained smoke detection model based on the improved YOLO V8. Next, the divided smoke image test set is used to test the trained smoke detection model based on the improved YOLO V8.

[0011] Preferably, the fire smoke detection model comprises: Using YOLO V8 as the basic deep learning algorithm model, the SimAM attention mechanism is added to the Backbone structure to adjust the weights of feature maps of different scales and extract the key features of smoke images; An improved bidirectional feature pyramid network (BIMAFPN) module is added to the feature fusion part of the Neck structure to fuse and extract feature information at different levels.

[0012] A fire smoke early recognition system based on deep learning, comprising: Front-end data acquisition module, used to collect image or video data; Deep learning training model module, used to design and build smoke detection models and perform model training; Real-time smoke recognition module, used to identify fire smoke from collected image or video data; The alarm feedback module issues early warning for fire smoke identified in real time and responds promptly.

[0013] Therefore, the present invention adopts the above-mentioned fire smoke early recognition method based on deep learning, which has the following technical effects: (1) Improve recognition accuracy: Due to the use of deep learning technology, the system can learn complex smoke feature patterns from a large amount of training data. Therefore, in practical applications, it can significantly improve the recognition accuracy of fire smoke and reduce the occurrence of false alarms and missed alarms.

[0014] (2) Real-time monitoring: By optimizing the model structure and algorithm, the goal of real-time operation on resource-constrained devices is achieved. The smoke identification process can be completed within a few milliseconds, greatly improving the timeliness of fire warnings and buying precious time for rapid response in emergency situations.

[0015] (3) Strong adaptability and robustness: On the one hand, a variety of data enhancement techniques are used in the processing of training data sets. For example, the bounding box position data is enriched through operations such as image rotation, random cropping, elastic deformation, and perspective transformation; various scene features in real life are simulated by applying Gaussian noise, compressing images, changing brightness and contrast, and generating fog, rain, and snow effects. Data enhancement technology not only enriches the training data set, but also enables the model to adapt to different light changes and shooting angles, and learn more diverse and generalized feature representations of smoke. This allows the model to maintain its ability to detect smoke and the stability of its performance in the face of various scene changes in the real world.

[0016] On the other hand, in terms of model improvement, the SimAM attention mechanism is added to the Backbone structure, and similar information is used to adjust the attention weights of different channels, reduce the focus on secondary features, and enhance the network's robustness to noise interference. An improved bidirectional feature pyramid network BIMAFPN module is added to the feature fusion part of the Neck structure, which extracts feature information at different levels through fusion to improve the detection performance of the fire smoke detection model.

[0017] (4) Easy to deploy and maintain: Based on the existing surveillance camera output video stream, the requirements for hardware configuration are reduced, allowing the system to be quickly deployed in various occasions. At the same time, it is both green and environmentally friendly, and economical and practical.

[0018] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a flowchart of a method for early identification of fire smoke based on deep learning; Figure 2 This is a schematic diagram of a fire smoke early identification system based on deep learning. DETAILED DESCRIPTION

[0020] The present invention can be explained in more detail by the following examples. The purpose of disclosing the present invention is to protect all changes and improvements within the scope of the present invention. The present invention is not limited to the following examples.

[0021] The present invention provides a method for early identification of fire smoke based on deep learning, comprising a front-end data acquisition module, a deep learning model training module, a real-time smoke identification module and an alarm feedback module.

[0022] S1. The front-end data acquisition module includes devices such as monitoring cameras or image sensors, which are used to capture images or video data in the monitoring area in real time and transmit them to the back-end processing unit.

[0023] S2, such as Figure 1 As shown in the figure, the deep learning model training module mainly includes four stages: data preparation and preprocessing, algorithm model design, model training and optimization, and model evaluation and verification.

[0024] In the data preparation and preprocessing stage, we first need to collect a large amount of smoke image data from various scenes and environmental conditions. In order to make the model have better generalization ability, it should cover as many fire types and smoke forms as possible. The second step is to annotate the collected data, especially to accurately mark the location of the smoke and fire.

[0025] Next, the fire smoke samples are randomly cropped, perspective transformed, elastically deformed, rotated, and enhanced by applying Gaussian noise, compressing images, changing brightness and contrast, etc. to enhance the expanded data set, simulating fire smoke in various scenes such as fog, rain and snow in real life, providing rich data support for the subsequent feature extraction process, so that it contains smoke features under various conditions, thereby improving the applicability of the algorithm in different scenarios. The enhanced expanded data set is divided into training set, test set and validation set for the subsequent algorithm training and testing process.

[0026] During the algorithm model design stage, a smoke detection model was constructed based on the latest YOLO series target detection algorithm. By training on a large number of labeled fire smoke samples, the basic morphological characteristics of smoke were learned, and the ability to recognize smoke under different lighting conditions and backgrounds was improved, resulting in a model that can effectively identify flame smoke characteristics.

[0027] In this embodiment, YOLO V8 is selected as the basic algorithm model for deep learning, and the characteristics of fire smoke in different scenarios are understood. In addition, the network model structure of YOLO V8 is improved as follows: Add the SimAM attention mechanism to the Backbone structure. Through 3-D weights, the SimAM attention module can assign a uniform weight to the feature information from each channel and globally adjust the attention weights of different channels. For non-fixed morphological features, it is necessary to consider global smoke, which can improve the accuracy of network model detection and have better robustness under complex environmental interference. In the feature fusion part of the Neck structure, the improved bidirectional feature pyramid network BIMAFPN module is added to quickly perform multi-scale feature fusion. The output of the Backbone network is bidirectionally connected to the Neck network, retaining an appropriate amount of shallow information to promote subsequent learning of image features, and enhancing the network's ability to detect small targets; at the same time, in the Neck network, richer gradient information will be passed to the output layer. Fusion extracts feature information at different levels and with different weights to improve the detection performance of the fire smoke detection model.

[0028] In the model training and optimization stage, the divided smoke image training set is used to train the deep learning network model based on the improved YOLO V8, and the hyperparameters are continuously adjusted to improve the prediction accuracy of the model, and the trained smoke detection model based on the improved YOLO V8 is obtained. The performance of the model is evaluated using the validation set, and the model with slow detection speed and low detection accuracy is returned to the second stage to modify the network structure of the model.

[0029] In the model evaluation and verification stage, the smoke detection model based on the improved YOLOV8 after training is tested using the divided smoke image test set to analyze the performance of the model in different scenarios. By comparing the difference between the true label and the predicted result, the performance indicators of the model, such as accuracy, recall rate and F1 score, are quantified to comprehensively measure the advantages and disadvantages of the model in different scenarios.

[0030] In this embodiment, the OpenCV library is used in the data preparation and preprocessing stage to realize the collection and annotation of image data; the pytorch framework is used to build a deep learning model in the algorithm model design stage; and the model training and optimization stage uses a large-scale labeled data set for training.

[0031] S3. The real-time smoke recognition module integrates the deep learning model trained by the deep learning model training module to perform real-time analysis on the video stream data collected by the front end to quickly determine whether fire smoke occurs; when the real-time smoke recognition module determines that there is a fire risk, the alarm feedback module is activated.

[0032] S4. In the alarm feedback module, the system will automatically trigger the alarm mechanism and send an alarm to relevant personnel through various means (such as sound and light alarms, mobile phone message notifications, etc.) so that timely action can be taken.

[0033] In another embodiment, a fire smoke early recognition system based on deep learning is also provided. Figure 2 As shown, including: Front-end data acquisition module, used to collect image or video data; Deep learning training model module, used to design and build smoke detection models and perform model training; Real-time smoke recognition module, used to identify fire smoke from collected image or video data; The alarm feedback module issues early warning for fire smoke identified in real time and responds promptly.

[0034] Therefore, the present invention adopts the above-mentioned deep learning-based fire smoke early identification system and method, which can monitor fire smoke in real time, improve the recognition accuracy of the model, enhance the model's adaptability to different environmental conditions, and meet the recognition of fire smoke in complex backgrounds.

[0035] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A fire smoke early identification method based on deep learning, characterized in that: The following steps are involved: S1, obtain smoke image or video data in real time through the front-end data acquisition module and transmit it to the back-end processing unit; S2. Label various fire smoke samples, expand the data set through data enhancement technology, introduce the improved bidirectional feature pyramid network BIMAFPN module and SimAM attention mechanism, build an improved YOLO V8 deep learning network model, and then use the expanded data set for model training; S3. After the model is verified, it is integrated into the fire smoke detection model as a real-time smoke recognition module to analyze the real-time acquired image or video data and quickly identify early fire smoke; S4. When fire smoke is identified, the system automatically triggers the alarm mechanism, issues an alarm through the alarm feedback module, and guides relevant personnel to take action.

2. The method for early identification of fire smoke based on deep learning according to claim 1 is characterized in that: Labeling various types of fire smoke samples includes collecting smoke image data in various scenes and environmental conditions, and labeling the smoke morphology of various fire types.

3. The method for early identification of fire smoke based on deep learning according to claim 1 is characterized in that: Data enhancement techniques include image rotation, random cropping, elastic deformation, perspective transformation, and application of Gaussian noise to obtain fire smoke characteristics in various scenarios.

4. The method for early identification of fire smoke based on deep learning according to claim 1 is characterized in that: Model training using the expanded dataset includes: Firstly, the expanded dataset is divided into training set, test set and validation set. Then, the divided smoke image training set is used to train the deep learning network model based on the improved YOLO V8 to obtain the trained smoke detection model based on the improved YOLO V8. Next, the divided smoke image test set is used to test the trained smoke detection model based on the improved YOLO V8.

5. The method for early identification of fire smoke based on deep learning according to claim 1 is characterized in that: Fire and smoke detection model, including: Using YOLO V8 as the basic deep learning algorithm model, the SimAM attention mechanism is added to the Backbone structure to adjust the weights of feature maps of different scales and extract the key features of smoke images; An improved bidirectional feature pyramid network (BIMAFPN) module is added to the feature fusion part of the Neck structure to fuse and extract feature information at different levels.

6. A fire smoke early recognition system based on deep learning, comprising: Front-end data acquisition module, used to collect image or video data; Deep learning training model module, used to design and build smoke detection models and perform model training; Real-time smoke recognition module, used to identify fire smoke from collected image or video data; The alarm feedback module issues early warning for fire smoke identified in real time and responds promptly.

Images