CN209433517U - A Fire Recognition and Alarm Device Based on Multiple Flame Images and Combination Criterion - Google Patents

Abstract

The utility model discloses a kind of based on more flame images and the fire identification warning device for combining criterion, including image acquisition units, conventional fire detector cells, image-signal processing system (image layer RBF recognition unit), control unit MCU, central control system (global layer RBF recognition unit), alarm output unit.The utility model is judged using more flame images with criterion is combined, and conventional fire detector is carried out criterion with image-type fire detector and is merged.The feature of doubtful fire image is identified using RBF neural in image layer, then output and the output of conventional fire sensor are judged together, global RBF neural is carried out to be identified, avoid that conventional fire detector sensitivity is not high, be not suitable for large space environment and image-type fire detector criterion is single, the excessively high disadvantage of wrong report rate of failing to report.

Description

A Fire Recognition and Alarm Device Based on Multiple Flame Images and Combination Criterion

technical field

The utility model belongs to the technical field of fire safety, and in particular relates to a fire recognition and alarm device based on multi-flame images and combined criteria.

Background technique

When a fire occurs, if it can be detected and called in time, the loss can be minimized. Traditional fire detectors mainly include temperature-sensing, smoke-sensing, light-sensing, gas-sensing, and compound-sensing types. The detection range is limited by space and height. When a fire occurs in the monitoring area, it cannot respond immediately. Only when the detection value reaches a certain level time to respond. Therefore, the use of traditional fire detectors is limited by conditions and is suitable for fire identification in small spaces, but its accuracy is difficult to meet the requirements in complex environments with large spaces.

Utility model content

The purpose of this utility model is to provide a fire identification and alarm device based on multi-flame images and combination criteria, which solves the problem that the traditional fire detection technology in the prior art is not suitable for large spaces, and the image-based fire detection technology has a single criterion , it is prone to the problem of false positives, omissions and false positives.

The technical scheme adopted by the utility model is that a fire identification and alarm device based on multi-flame images and combined criteria includes a plurality of cameras, the cameras are connected with a video acquisition card unit, and the video acquisition card unit is connected with an image signal processing unit; it also includes Sensor a, sensor b, sensor c, sensor a, sensor b, sensor c are connected to the control unit MCU as the input end; the image signal processing unit and the control unit MCU are connected to the central control system unit as the input end; the output end of the central control system unit Connected with sound alarm and fire alarm lights;

The utility model is also characterized in that:

Wherein said camera comprises two kinds of cameras of some infrared CCD cameras and ordinary CCD cameras;

Wherein said sensor a is a temperature sensor, using NTC thermistor sensor; sensor b is a gas sensor, mainly including MQ2, MQ7; sensor c is a flame sensor, composed of ultraviolet photosensitive tube R2868 and supporting circuit board C3704;

Wherein the image signal processing unit (3) and the central control system unit (8) are RBF neural network identification units.

Description of drawings

Fig. 1 is a kind of structural representation of the fire recognition alarm device based on double-layer RBF neural network multi-criteria of the present utility model;

In the figure, 1. camera unit, 2. video capture card unit, 3. image signal processing unit, 4. sensor a, 5. sensor b, 6. sensor c, 7. control unit MCU, 8. central control system unit, 9. Sound alarm, 10. Fire alarm lights.

Detailed ways

The utility model discloses a fire identification and alarm device based on multi-flame images and combined criteria, as shown in Figure 1: it includes a plurality of cameras 1, and the cameras 1 include two types of cameras, a plurality of infrared CCD cameras and ordinary CCD cameras, and the cameras 1 are connected There is a video capture card unit 2; the signal output of the video capture card unit 2 is to the image signal processing unit 3; the image signal processing unit 3 outputs signals to the sensor a4 respectively, the sensor a4 is a temperature sensor, the sensor b5, the sensor b5 is a gas sensor, and the sensor c6 , sensor c6 is a flame sensor: sensor a4, sensor b5, and sensor c6 are connected to the control unit MCU7 as input terminals; flame sensor c6 is composed of ultraviolet photosensitive tube R2868 and supporting circuit board C3704; gas sensor b5 mainly includes MQ2, MQ7; temperature sensor a4 adopts NTC thermistor sensor. The image signal processing unit 3 and the control unit MCU7 output signals to the central control system unit 8; the central control system unit 8 is connected with a sound alarm 9 and a fire alarm lamp 10.

The functions and working principles of the main components in a fire identification and alarm device based on multi-flame images and combination criteria of the present utility model are as follows:

Camera 1 monitors the target area through video, and obtains image frames according to a certain frequency sampling. Camera 1 uses an ordinary CCD camera combined with an infrared CCD camera to monitor the scene. The optical signal entering the lens is converted into a video electrical signal, and the video is connected through an optical fiber The server uses the IP network to realize the network connection with the monitoring center.

The video capture card unit 2 can monitor multiple targets at the same time, and conduct inspections according to certain rules. Through the video acquisition card unit, real-time video signals from different monitoring points can be received and stored.

The image signal processing unit 3 (image layer RBF identification unit) performs pre-judgment processing on the result output by the image acquisition unit. The sampled image is compared with the reference image, and the change of the direct mutual information value between the sampled frame and the reference frame is calculated to predict whether there is a fire. Then the suspected image is segmented and feature extracted, and the extracted feature is input into the image layer RBF neural network, and its output is used as the input of the global neural network.

The control unit MCU7 can be realized by an 8-32-bit microprocessor chip, and the microprocessor chip used must have a timer unit and a user I/O port, and the working main clock of the microprocessor should be above 24MHz. Available chips such as MCS51 series, PIC series, AVR series and ARM series all have more products to satisfy, specifically MCS51 series chips, which can fully meet the requirements of the utility model.

The central control system unit 8 (the global layer RBF identification unit) performs combined identification and judgment on the output of the temperature sensor, gas sensor, flame sensor and image layer RBF neural network, and if the result meets the criteria and exceeds the threshold, the alarm unit is driven to work.

The alarm unit mainly includes a sound alarm 9 and a fire alarm lamp 10 to form. When the central control unit sends a light alarm command or a sound alarm command, the fire alarm light 10 and the sound alarm 9 respond and drive the fire flashing light alarm and the fire alarm sound respectively.

In the fire identification and alarm device based on double-layer RBF neural network and multi-criteria of the utility model, the RBF (Radial Basis Function) neural network is also called radial basis neural network, which is a kind of device with strong input-output mapping function. The optimal network has very obvious advantages in terms of learning speed, classification ability, and approximation ability. It has a very wide range of applications in the field of pattern and function recognition, especially in fire recognition. This algorithm uses image processing technology to extract the feature information of fire images, and uses RBF neural network as the carrier to carry out the final recognition and judgment of fire images.

The RBF neural network is mainly composed of an input layer, a hidden layer and an output layer, and the input node only transmits the input signal to the hidden layer. The hidden layer nodes are composed of radial basis functions (such as Gaussian functions), and the action function in the hidden layer nodes has the ability of local approximation, and responds locally to the input signal. The output layer nodes generally use linear functions. For a given input vector x, the unit output of the RBF network output layer is:

Its output serves as the global network input

In the formula, R _i (x) is the hidden layer output, _{wi ik} is the network weight, ci is the center vector of the basis function, m is the number of hidden layer nodes, p is the number of output nodes, ||xc _i || represents x to the distance between c _i . R _i (x) has a unique maximum at c _i , and the increase of ||xc _i || will cause R _i (x) to attenuate. From formula (1) and formula (2), it can be seen that the input layer realizes the nonlinear mapping from x to R _i (x), and the output layer realizes the nonlinear mapping from R _i (x) to y _k (x), The composition of the hidden layer is a group of radial basis functions, and the nonlinear mapping relationship can be realized through the radial basis functions. The parameter vectors associated with each hidden layer node are _{ci (ie center) and σ i (} ie width). The learning process of the RBF network is mainly divided into three stages:

(1) Determine the center c _i of each RBF unit: Generally, c _i is determined by the k-means clustering analysis technique. Selecting representative data as the center of the RBF unit can reduce the number of RBF units in the hidden layer and reduce the network cost. level of complexity.

(2) Determine the radius σ _i of each RBF unit: the radius σ _i determines the size of the RBF unit's acceptance domain, which has a great influence on the network accuracy.

(3) Adjustment weight matrix w: Here w refers to the weight between the hidden layer and the output layer, and the adjustment weight matrix can use the gradient method.

The central control system unit 8 (global layer RBF recognition unit) in the utility model recognizes and judges the temperature sensor, gas sensor, flame sensor and image layer RBF neural network output, and the number of design output neurons is 1, and the value is 0 or 1. 0 means no fire occurred, 1 means fire occurred. The result drives the alarm unit to work.

The main eigenvalues of the RBF neural network in the utility model include the number of flame sharp angles, the area change rate, the density, the elongation, and the offset distance of the centroid point, and the extracted eigenvalues are input into the RBF neural network of the image layer; due to the 5 input layers The calculated values of the features are too different. In order to make the input data in the [0,1] interval, normalization processing is required. The advantage of this processing is that the RBF neural network is easier to learn and train the processed data, and the input node only transmits the input signal to the hidden layer. The hidden layer nodes are composed of radial basis functions (such as Gaussian functions), and the action function in the hidden layer nodes has the ability of local approximation, and responds locally to the input signal. The initial weight is assigned by a random function, and the error between the target output and the actual output is calculated; the linear weight of the output layer is adjusted with an appropriate learning rate by the least square method, and the training ends when the network reaches the specified accuracy. When forecasting, if the gap with the actual result exceeds the predetermined limit, the network will automatically adjust, and this forecasting model can adapt to the needs of fire variability.

The utility model is a fire identification and alarm device based on multi-flame images and combined criteria. Its working principle is as follows: obtain the sampling image of the monitoring area through the camera 1 infrared CCD camera and the video acquisition card unit 2, and calculate the sampling image and the reference image or The similarity measure of the background image, when the similarity of the sampled images of consecutive frames is higher than the set threshold, further judgment is made. First, preprocessing, segmentation, and feature extraction are performed on the suspected fire sampling image, and then the feature parameters are input into the pre-trained image signal processing system unit 3 (image layer RBF neural network), but at this time, the image signal processing system unit 3 The output cannot completely and accurately judge whether it is a fire. The output of sensor a4, sensor b5 and sensor c6 is input into the control unit MCU7 through analog-to-digital conversion in parallel, and then the signal output of the control unit MCU7 is fed back to the central control system unit 8 through the serial bus, and the central control system unit 8 processes the image signal The input of the unit 3 and the control unit MCU7 is normalized, and the trained RBF neural network is input, and the output of the global neural network is used to judge whether it is a fire. If a fire occurs, the fire alarm lamp and the sound alarm are driven to respond.

Illustrate the utility model advantage in principle:

The fire recognition and alarm device based on multi-flame images and combined criteria of the utility model is a fire detection technology based on digital image processing, which intercepts suspected fire images from monitoring video for analysis, and then extracts characteristic parameters of suspected areas and compares them with preset Feature thresholds are compared to determine whether a fire has occurred. Multi-layer image-based fire detection technology can effectively solve the problem of fire safety in large spaces. However, due to the complex background of fire images in large spaces, it is difficult to accurately segment the flame area, and the feature criterion is single, so there is a high rate of false positives and negative negatives. Combining traditional fire detection technology and image-based fire detection technology with multi-criteria, the RBF neural network is used to establish a fire recognition model, and the extracted multi-layer image criterion features are used as input to classify and recognize fire images. The results of a series of fire experiments show that the fire recognition of different scenes has a high accuracy rate, which can effectively reduce the false alarm rate of fire and improve the accuracy of fire alarm;

On the other hand, the combination of multi-layer image-based fire detection and traditional fire detection technology is used as the input of RBF neural network, which overcomes the shortcomings of traditional fire detectors that are not suitable for large spaces and image-based fire detection technology. The entire system does not require manual intervention in related operations, and the fire alarm response speed is fast and has high accuracy.

Claims (4)

1. a kind of based on more flame images and the fire identification warning device for combining criterion, which is characterized in that including several camera shootings Machine (1), video camera (1) are connected with video frequency collection card unit (2), and video frequency collection card unit (2) connects image signal processing unit (3)；It further include sensor a (4), sensor b (5), sensor c (6), sensor a (4), sensor b (5), sensor c (6) make Control unit MCU (7) are connected with for input terminal；Image signal processing unit (3) and control unit MCU (7) connect as input terminal Connect central control system unit (8)；The output end of central control system unit (8) is connected with voice guard (9) and fire report Warning lamp (10).

2. it is according to claim 1 a kind of based on more flame images and the fire identification warning device for combining criterion, it is special Sign is that video camera (1) includes several infrared C CD video cameras and two kinds of video cameras of common CCD camera.

3. it is according to claim 1 a kind of based on more flame images and the fire identification warning device for combining criterion, it is special Sign is that sensor a (4) is temperature sensor, using NTC thermistor sensor；Sensor b (5) is gas sensor, main It to include MQ2 and MQ7；Sensor c (6) is flame sensor, is made of ultraviolet light sensing tube R2868 and support circuit plate C3704.

4. it is according to claim 1 a kind of based on more flame images and the fire identification warning device for combining criterion, it is special Sign is that described image signal processing unit (3) and central control system unit (8) are RBF neural recognition unit.