Infrared smoke detection method and early warning system based on convolutional neural network

Abstract

The application discloses an infrared smoke detection method and early warning system based on a convolutional neural network. Two infrared light emitting units are arranged at different positions in a smoke sensor labyrinth. The two infrared light emitting units emit infrared light in sequence after receiving the emission instruction of a receiving control unit. After the infrared light signal, the infrared light signal is converted into an available electric signal according to the intensity of the infrared light signal. The electric signal is amplified to an amplified electric signal which can be directly used for analysis. The amplified electric signal is received and signal length accumulation calculation is performed. Whether the signal length accumulation reaches L is judged. The signal with the signal length accumulation reaching L is obtained and smoke signal judgment is performed. The result is the number of accumulated smoke signals and comparison with a warning threshold. An alarm unit receives an alarm signal to trigger an alarm. The application can distinguish smoke and water vapor, and provides a more reliable, accurate and effective fire detection and early warning system.

Description

Technical Field

[0001] This invention relates to the field of digital signal processing and artificial intelligence, and more particularly to an infrared smoke detection method and early warning system based on convolutional neural networks. Background Technology

[0002] Optical reflection is the most common method used in smoke detectors. Optical smoke detectors use a light beam to pass through the air; when smoke particles enter the beam, the light is scattered and detected by the warning system. Changes in this scattered signal are used to determine the presence of smoke. Optical smoke detectors are generally classified into two types: ionization and light scattering.

[0003] While traditional smoke detection technology plays a crucial role in fire detection, it also has some drawbacks: 1. High false alarm rate: Smoke detectors can sometimes trigger false alarms due to non-fire-related causes such as cooking, steam, or dust, leading to unnecessary trouble and interference. 2. Missed alarms: Certain types of fires, especially those with light or no smoke, may fail to be detected by the smoke detector in time, thus delaying fire discovery and response. 3. Inapplicability to certain environments: Smoke detectors may not be suitable for certain environments, such as extreme temperatures, humidity levels, or chemical gas environments, leading to performance degradation. For example, in common cold storage facilities, optical signals often experience unstable changes, affecting normal alarm operation.

[0004] To improve the smoke detection capabilities of smoke detectors, the current mainstream approach is to use red and blue dual-light technology in the hardware. When smoke enters the detection area, the smoke causes light scattering, interfering with the receiving beam and weakening the signal. When the received light signal weakens to a certain level, combined with the difference in red and blue light, the smoke detector will determine that a fire may have occurred and trigger an alarm. This design can effectively improve the accuracy and stability of smoke detection, reducing the risk of false alarms and missed alarms.

[0005] However, this type of device also has certain drawbacks: 1. Complexity: The design and installation of this type of smoke detection system is usually quite complex, requiring precise installation and calibration to ensure accurate beam transmission and detection. This may require more time and technical expertise. 2. False Alarms: Although red-blue dual-light smoke detectors are relatively good at reducing false alarms, they still cannot filter water vapor with high precision. This may cause unnecessary inconvenience to users. 3. Difficulty in Sensitivity Adjustment: This type of smoke detector may require sensitivity adjustment in different environments to adapt to different smoke concentrations. Adjusting the sensitivity may require some professional knowledge and experience; otherwise, it may lead to false alarms or missed alarms. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to provide an infrared smoke detection method and early warning system based on convolutional neural networks that can distinguish between smoke and water vapor.

[0007] To address the aforementioned technical problems, an infrared smoke detection method based on convolutional neural networks includes the following steps:

[0008] S1: Two infrared light emitting units are set at two different positions in the smoke detector maze, and emit infrared light after receiving the emission command from the control unit;

[0009] S2: After receiving the infrared light signal, the infrared-to-electrical conversion unit converts the infrared light signal into a usable electrical signal according to the intensity of the infrared light signal;

[0010] S3: The infrared signal amplification unit receives the electrical signal and amplifies it to an amplified electrical signal that can be directly used for analysis;

[0011] S4: The signal length accumulation unit receives the amplified electrical signal and performs signal length accumulation calculation to determine whether the signal length accumulation has reached L;

[0012] S5: The smoke probability analysis unit acquires signals whose cumulative signal length reaches L and judges the smoke signal. The result is the cumulative number of warnings for the smoke signal and compared with the warning threshold. If the number of warnings is greater than the threshold, an alarm signal is issued.

[0013] S6: The alarm unit receives an alarm signal and triggers the alarm.

[0014] In a further optimized scheme, the smoke probability analysis unit is a smoke convolutional neural network trained to convergence.

[0015] To further optimize the solution, step S5 specifically includes:

[0016] S51: Obtain the cumulative signal length that has reached L;

[0017] S52: Compare whether the signal with a length of L is a smoke signal;

[0018] S53: If the comparison result is a smoke signal, then the number of warnings is n+1. Calculate whether the number of warnings has reached the threshold t. If it has, then trigger an alarm. Otherwise, delete the accumulated signal data and continue to execute step S51 for cumulative comparison. If the result is a non-smoke signal, then the number of warnings is n+0. Delete the accumulated signal data and continue to execute step S51 for cumulative comparison.

[0019] Further optimization of the scheme reveals that the basic module of the smoke convolutional neural network mainly consists of two blocks, used to calculate the probability of smoke, including:

[0020] S531: The acquired signal is input into the network in a size of 1×L×2;

[0021] S532: Signal processing is performed on the network backbone;

[0022] S533: The final output of the network is a 1×3 vector.

[0023] In a further optimized scheme, the network backbone in step S532 consists of blocks repeated m times, and each block is composed of a basic block and a shrinkage unit.

[0024] Further optimization of the scheme: the basic block includes Conv, Batch Normalization and leakyrelu modules, which are used to obtain signal data feature maps.

[0025] The further optimized scheme is that the shrinkage is composed of convolutional layers, including a GAP module, a CSP module, and a residual module.

[0026] Further optimization of the scheme, the corresponding formula for soft threading in the shrinkage unit is as follows:

[0027]

[0028] Here, x represents the input vector, and f(x) represents the mapped features. t is a positive number, which we can express as:

[0029] t = α·average|x h，w，c |

[0030]

[0031] Where h, w, and c represent height, width, and number of channels, respectively, and z is the output feature vector after the fully connected layer.

[0032] This invention also discloses an early warning system for infrared smoke detection based on a convolutional neural network, which further includes:

[0033] Infrared emitting units are positioned at different locations in the smoke detector maze, facing forward and backward.

[0034] The control unit is used to send commands to the infrared emitting unit to emit infrared light;

[0035] An infrared photoelectric conversion unit is used to receive the electrical signal and amplify it to an amplified electrical signal that can be directly used for analysis;

[0036] The signal amplification unit is used to receive amplified electrical signals, accumulate signal length, and make judgments.

[0037] The signal length accumulation unit is used to receive amplified electrical signals, accumulate the signal length, and transmit a signal whose accumulated signal length reaches L.

[0038] The smoke probability analysis unit acquires signals whose cumulative signal length reaches L and compares them with smoke signals. It calculates the cumulative number of warnings for smoke signals and compares it with the warning threshold. If the number of warnings exceeds the threshold, an alarm signal is issued.

[0039] Alarm unit: Used to receive alarm signals that reach a threshold and trigger the alarm.

[0040] Furthermore, the smoke probability analysis unit includes:

[0041] Signal acquisition module: Acquires the cumulative signal length;

[0042] Smoke signal recognition module: used to compare whether a signal with a length of L is a smoke signal;

[0043] Alarm Analysis Module: If the comparison result is a smoke signal, then the number of warnings is n+1. Calculate whether the number of warnings has reached the threshold t. If it has, then trigger an alarm. Otherwise, delete the accumulated signal data and continue the comparison of accumulated receipt length. If the result is a non-smoke signal, then the number of warnings is n+0. Delete the accumulated signal data and continue the comparison of accumulated receipt length.

[0044] The technical effects of this invention are as follows:

[0045] 1. This invention relates to an infrared smoke detection method and early warning system based on a convolutional neural network. The detection circuit of the early warning system transmits and receives infrared signals and sends the received infrared signals to a signal length accumulation unit. When the signal length accumulation unit receives the infrared signal, it judges the signal, thus achieving the beneficial effect of distinguishing between smoke and water vapor. This provides a more reliable, accurate and effective fire detection and early warning system, helps reduce false alarms, improves early warning speed, and better protects people's lives and property in emergency situations.

[0046] 2. This invention relates to an infrared smoke detection method and early warning system based on convolutional neural networks. It introduces convolutional neural network methods into smoke detection and smoke recognition, utilizing the ability of neural networks to automatically extract complex features. By learning the characteristics and patterns of smoke through a large amount of data, it improves the accuracy of fire detection by smoke detectors. In particular, it addresses the common problem of false alarms caused by water vapor in smoke detectors. This method can adapt well to different environments and conditions while accurately distinguishing different types of interference factors, thereby reducing the risk of false alarms. It can also capture weak smoke signals, reducing the possibility of missed alarms. Attached Figure Description

[0047] Figure 1This is a signal flow diagram of the infrared smoke detection method based on convolutional neural networks of the present invention;

[0048] Figure 2 This is a flowchart of the probability analysis of the infrared smoke detection method based on convolutional neural networks of the present invention;

[0049] Figure 3 This is a structural diagram of the infrared smoke detection and early warning system based on convolutional neural networks of the present invention;

[0050] Figure 4 This is a schematic diagram of the network module structure of the infrared smoke detection method based on convolutional neural networks of the present invention;

[0051] Figure 5 This is a schematic diagram of the basic block structure of the infrared smoke detection method based on convolutional neural networks in this invention;

[0052] Figure 6 This is a schematic diagram of the neural network shrinkage unit structure of the infrared smoke detection method based on convolutional neural networks of the present invention. Detailed Implementation

[0053] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand and implement the present invention. However, the embodiments are not intended to limit the invention.

[0054] In some of the processes described in the specification, claims, and accompanying drawings of this invention, multiple operations appearing in a specific order are included. However, it should be clearly understood that these operations may not be executed in the order they appear herein, or may be executed in parallel. The operation numbers, such as 101, 102, etc., are merely used to distinguish different operations and do not represent any execution order. Furthermore, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the descriptions such as "first," "second," etc., in this document are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to different types.

[0055] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0056] like Figures 1 to 6As shown, this invention provides an infrared smoke detection method based on a convolutional neural network, comprising the following steps:

[0057] S1: Two infrared light emitting units are set at two different positions in the smoke detector maze. The control unit of the warning system issues a command; the two infrared light emitting units, which are facing forward and backward, emit infrared light in sequence after receiving the command.

[0058] S2: After receiving the infrared light signal, the infrared-to-electrical conversion unit converts it into a usable electrical signal according to the intensity of the infrared light signal.

[0059] S3: The infrared signal amplification unit will receive the above-mentioned electrical signal and amplify it into an amplified electrical signal that can be directly used for analysis;

[0060] S4: The signal length accumulation unit receives the amplified electrical signal and performs signal length accumulation calculation, and transmits a signal whose signal length has accumulated to L.

[0061] S5: The smoke probability analysis unit acquires signals whose cumulative signal length reaches L and compares them with smoke signals. It calculates the cumulative number of warnings for smoke signals and compares them with the warning threshold. If the number of warnings is greater than the threshold, an alarm signal is issued.

[0062] S6: The alarm unit receives an alarm signal and triggers the alarm device to make a judgment.

[0063] like Figure 2 As shown, step S5 includes:

[0064] S51: Obtain the cumulative signal length that has reached L;

[0065] S52: Compare whether the signal with a length of L is a smoke signal;

[0066] S53: If the comparison result is a smoke signal, then the number of warnings is n+1. Calculate whether the number of warnings has reached the threshold t. If it has, then trigger an alarm. Otherwise, delete the accumulated signal data and continue to execute step S51 for cumulative comparison. If the result is a non-smoke signal, then the number of warnings is n+0. Delete the accumulated signal data and continue to execute step S51 for cumulative comparison.

[0067] The AI ​​smoke detection calculation module of this invention includes a smoke probability analysis unit and a signal length accumulation unit. The AI ​​smoke detection calculation module is based on a smoke convolutional neural network trained to convergence. Convolutional neural networks have excellent generalization ability in visual models, and weight sharing greatly reduces the number of free parameters, thereby reducing memory requirements and facilitating the training of more complex and powerful networks. Therefore, the AI ​​smoke detection calculation module of this embodiment mainly relies on a convolutional neural network for its construction.

[0068] like Figure 4 As shown, the smoke convolutional neural network of this invention mainly refers to a convolutional neural network constructed based on CSP and shrinkage. The basic module mainly consists of two blocks used to calculate the probability of smoke, including the following:

[0069] S531: Input terminal (S301), the acquired signal is input into the network in a size of 1×L×2;

[0070] S532: Network backbone (S302), the network backbone performs signal processing;

[0071] S533: Output terminal (S303), the final output of the network is a 1×3 vector.

[0072] The network backbone consists of blocks repeated m times, each block being composed of a basic block and a shrinkage unit.

[0073] like Figure 6 As shown, the shrinkage unit is composed of convolutional layers, including a GAP module, a CSP module, and a residual module. The shrinkage unit is an improved version of Deep Residual Shrinkage Networks for Fault Diagnosis (IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL.16, NO.7, JULY 2020). The main improvement is that the original GAP input is processed by a Cross Stage Partial Network (CVPR 2020 CSPNet: A New Backbone that can Enhance Learning Capability of CNN). This processing can save 20% of computation without changing accuracy. The final output of this structure is a 1*3 vector used to determine whether it is water vapor.

[0074] The shrinkage unit structure has a strong suppression effect on noise such as water vapor, especially the softthreading, and the corresponding formula is as follows:

[0075]

[0076] Here, x represents the input vector, and f(x) represents the mapped features. t is a positive number, which we can express as:

[0077] t = α·average|x h，w，c|

[0078]

[0079] Where h, w, and c represent height, width, and number of channels, respectively, and z is the output feature vector after the fully connected layer.

[0080] like Figure 5 The diagram shown is the basic block structure S401, which includes Conv, Batch Normalization, and leakyrelu modules to obtain signal data feature maps.

[0081] like Figure 3 As shown, the infrared smoke detection and early warning system based on convolutional neural networks of the present invention further includes:

[0082] Infrared emitting units are positioned at different locations in the smoke detector maze, facing forward and backward.

[0083] The control unit is used to send commands to the infrared emitting unit to emit infrared light;

[0084] An infrared photoelectric conversion unit is used to receive the aforementioned electrical signal and amplify it to an amplified electrical signal that can be directly used for analysis;

[0085] The signal amplification unit is used to receive amplified electrical signals, accumulate signal length, and make judgments.

[0086] The signal length accumulation unit is used to receive amplified electrical signals, accumulate the signal length, and transmit a signal whose accumulated signal length reaches L.

[0087] The smoke probability analysis unit acquires signals with a cumulative signal length of L and compares them with smoke signals. It calculates the cumulative number of warnings for the smoke signals and compares this number with a warning threshold. If the number of warnings exceeds the threshold, an alarm signal is issued. This unit includes:

[0088] Signal acquisition module: Acquires the cumulative signal length;

[0089] Smoke signal recognition module: used to compare whether a signal with a length of L is a smoke signal;

[0090] Alarm Analysis Module: If the comparison result is a smoke signal, then the number of warnings is n+1. Calculate whether the number of warnings has reached the threshold t. If it has, then trigger an alarm. Otherwise, delete the accumulated signal data and continue to perform length accumulation comparison. If the result is not a smoke signal, then the number of warnings is n+0. Delete the accumulated signal data and continue to perform length accumulation comparison.

[0091] The above-described embodiments are merely preferred embodiments provided to fully illustrate the present invention, and the scope of protection of the present invention is not limited thereto. Equivalent substitutions or modifications made by those skilled in the art based on the present invention are all within the scope of protection of the present invention. The scope of protection of the present invention is defined by the claims.

Claims

1. An infrared smoke detection method based on convolutional neural networks, characterized in that, It includes the following steps: S1: Two infrared light emitting units are set at two different positions in the smoke detector maze, and emit infrared light after receiving the emission command from the control unit; S2: After receiving the infrared light signal, the infrared-to-electrical conversion unit converts the infrared light signal into a usable electrical signal according to the intensity of the infrared light signal; S3: The infrared signal amplification unit receives the electrical signal and amplifies it to an amplified electrical signal that can be directly used for analysis; S4: The signal length accumulation unit receives the amplified electrical signal and performs signal length accumulation calculation to determine whether the signal length accumulation has reached L; S5: The smoke probability analysis unit acquires signals whose cumulative signal length reaches L and judges the smoke signal. The result is the cumulative number of warnings for the smoke signal and compared with the warning threshold. If the number of warnings is greater than the threshold, an alarm signal is issued. Step S5 specifically includes: S51: Obtain the cumulative signal length that has reached L; S52: Compare whether the signal with a length of L is a smoke signal; S53: If the comparison result is a smoke signal, the number of warnings is increased by n+1. It is then calculated whether the number of warnings has reached the threshold t. If it has, an alarm is triggered; otherwise, the accumulated signal data is deleted, and step S51 is continued for cumulative comparison. If the result is a non-smoke signal, the number of warnings is increased by n+0, the accumulated signal data is deleted, and step S51 is continued for cumulative comparison. The basic module of the smoke convolutional neural network mainly consists of two blocks, used to calculate the probability of smoke, including: S531: The acquired signal is input into the network in a size of 1×L×2; S532: Signal processing is performed on the network backbone; S533: The final output of the network is a 1×3 vector; in step S532, the backbone of the network consists of blocks repeated m times, each block being composed of a basic block and a shrinkage unit; the basic block includes Conv, Batch Normalization, and leaky ReLU modules to obtain signal data feature maps; the shrinkage is composed of convolutional layers, including a GAP module, a CSP module, and a residual module; the corresponding formula for soft threading in the shrinkage unit is as follows: ； Here, x represents the input vector, f(x) represents the mapped features, and t is a positive number, expressed as: ； ； Where h, w, and c represent height, width, and number of channels, respectively, and z is the output feature vector after the fully connected layer; S6: The alarm unit receives an alarm signal and triggers the alarm.

2. The infrared smoke detection method based on convolutional neural networks as described in claim 1, characterized in that, The smoke probability analysis unit is a smoke convolutional neural network trained to convergence.

3. A system for an infrared smoke detection and early warning method based on a convolutional neural network as described in claim 1 or 2, characterized in that, It also includes: Infrared emitting units are positioned at different locations in the smoke detector maze, facing forward and backward. The control unit is used to send commands to the infrared emitting unit to emit infrared light; An infrared photoelectric conversion unit is used to receive the electrical signal and amplify it to an amplified electrical signal that can be directly used for analysis; The signal amplification unit is used to receive amplified electrical signals, accumulate signal length, and make judgments. The signal length accumulation unit is used to receive amplified electrical signals, accumulate the signal length, and transmit a signal whose accumulated signal length reaches L. The smoke probability analysis unit acquires signals whose cumulative signal length reaches L and compares them with smoke signals. It calculates the cumulative number of warnings for smoke signals and compares it with the warning threshold. If the number of warnings exceeds the threshold, an alarm signal is issued. Alarm unit: Used to receive alarm signals that reach a threshold and trigger the alarm.

4. The infrared smoke detection and early warning system based on convolutional neural networks according to claim 3, characterized in that, The smoke probability analysis unit includes: Signal acquisition module: Acquires the cumulative signal length; Smoke signal recognition module: used to compare whether a signal with a length of L is a smoke signal; Alarm Analysis Module: If the comparison result is a smoke signal, then the number of warnings is n+1. Calculate whether the number of warnings has reached the threshold t. If it has, then trigger an alarm. Otherwise, delete the accumulated signal data and continue the comparison of accumulated receipt length. If the result is a non-smoke signal, then the number of warnings is n+0. Delete the accumulated signal data and continue the comparison of accumulated receipt length.

Images