Automatic detection method of tobacco smoldering temperature and CO carbon intelligent perception

Abstract

An automatic detection method for tobacco smoldering temperature and CO carbon intelligent perception, which includes the following steps: first, when the temperature sensor is group A, a plurality of temperature sensors are formed into a triangular waveform network node shape at the geometric symmetry of the diagonal line of the cigarette pack, and a temperature difference AI detection algorithm is designed; second, when the CO carbon sensor is group B, it is installed between two cigarette packs; when there is smoldering under or in the middle of the cigarette pack, the generated CO rises to the upper part of the space, and the CO carbon sensor detects CO; third, before determining the temperature difference value and CO value, first obtain the big data range of the normal temperature of the cigarette pack, the initial sensitive temperature of smoldering and CO through repeated experiments; fourth, optimize the YOLOv7 neural network structure and parameters, establish a network model suitable for smoldering temperature and CO, and then use AI classification to find out the data relationship model of smoldering temperature and CO; fifth, use the improved evidence theory to make a decision on the characteristic data fusion of temperature and CO to inform the user of the smoldering state type.

Description

Technical Field

[0001] The invention relates to an automatic detection method for tobacco smoldering temperature and CO carbon intelligent perception. Background Art

[0002] At present, smoke detection mainly focuses on flame detection and infrared detection, and no in-depth research has been conducted on the extraction method of early suspected smoke areas (Jia Yang. Research on video fire detection method based on saliency detection and smoke spatiotemporal characteristics [D]. University of Science and Technology of China, 2016). There is a lack of visual methods for the visualization of the smoldering smoke of small targets without flames in the early stage and the edge features of objects after smoldering, and AI intelligent monitoring methods that fuse vision and CO sensor data. To this end, we developed an automatic detection method for tobacco smoldering temperature and CO carbon intelligent perception. Summary of the invention

[0003] The technical problem to be solved by the purpose of the present invention is to provide an automatic detection method for tobacco smoldering temperature and CO carbon intelligent perception, which uses the analysis results of YOLOV7 of deep learning to verify the correctness of the evidence theory analysis, which is equivalent to integrating the results of AI and data analysis, combining the dual-scale analysis of the cigarette package status, and drawing the final conclusion, making the entire monitoring process more robust and the results more accurate.

[0004] In order to achieve the above object, the technical solution adopted by the present invention is to provide an automatic detection method for tobacco smoldering temperature and CO carbon intelligent perception, which comprises the following steps:

[0005] First, when the temperature sensor is installed in group A, multiple temperature sensors are formed into a triangular waveform network node at the geometric symmetry of the diagonal line of the cigarette pack, and a temperature difference AI detection algorithm is designed;

[0006] 2. When the CO carbon sensor is installed as group B, it is between two cigarette packs. When there is smoldering under or in the middle of the cigarette pack, the generated CO rises to the upper part of the space, and the CO carbon sensor detects CO.

[0007] 3. Before determining the temperature difference and CO value, first conduct repeated experiments to obtain the large data range of cigarette pack normal temperature, smoldering initial sensitive temperature, smoldering combustion temperature and CO;

[0008] Fourth, optimize the YOLOv7 neural network structure and parameters, establish a network model suitable for smoldering temperature and CO, and then use AI classification to find the data relationship model between smoldering temperature and CO;

[0009] Fifth, the deep learning data is finally integrated and the improved evidence theory is used to make feature data fusion decisions of temperature and CO, including the synthesis of conflicting information and data to inform the user of the type of smoldering state.

[0010] In one or more embodiments of the present invention, the data relationship model in step 4 includes CO values ​​at room temperature of the cigarette package, at the initial smoldering sensitive temperature, and at the combustion temperature.

[0011] In one or more embodiments of the present invention, in step 1, the temperature difference AI detection algorithm includes: designing a constant temperature sensor A0 at the inner door of the warehouse, obtaining the warehouse environment temperature T0 under normal conditions, and comparing it with the temperature measured by the temperature sensor in the cigarette pack to obtain the temperature difference. The following is a mathematical model for temperature difference design.

[0012] Assume the temperature difference variable parameter is T ij , the temperature sensor in the cigarette pack is A i , the temperature measured by the temperature sensor inside the cigarette pack is recorded as T i , where (i=n-1), this case takes 5 sensors as an example, then:

[0013] T i1 =T1-T0

[0014] T i2 =T2-T0

[0015] T i3 =T3-T O

[0016] T i4 =T4-T0 ......

[0018] T i(n-1) =T n-1 -T0

[0019] Average temperature difference:

[0020]

[0021] When T ij -Δ T >C, or when T i When the temperature is equal to or greater than the smoldering start temperature T, smoldering may occur within the detection range of the temperature sensor to alert the user.

[0022] In one or more embodiments of the present invention, when reminding the user, a fusion decision is made with the sensor data of group B.

[0023] In one or more embodiments of the present invention, when reminding the user, a fusion decision is made with the sensor data of group B.

[0024] In one or more embodiments of the present invention, in step three, through repeated tobacco smoldering experiments, the obtained temperature and CO concentration data of each situation are screened and the temperature is labeled and classified into normal initial temperature, smoldering critical temperature, and smoldering combustion temperature; and the CO concentration is also labeled and classified into three categories, normal initial concentration, smoldering critical temperature, and smoldering combustion concentration. The temperature and CO concentration image label data are pushed into the improved YOLOV7 neural network for training, and the trained data is analyzed.

[0025] In one or more embodiments of the present invention, in step four, the YOLOV7 model is improved, the Focus module is modified to perform targeted training on the data, an output is traversed separately to improve efficiency, and the slice module is sliced ​​horizontally to convert the CO concentration and temperature arrays into a set of two-dimensional arrays, strengthen data analysis and extraction, improve the connection between the two, compress and smooth the function data, reduce data redundancy, reduce the oscillation of the probability line, make the data more credible, and then use the training results to obtain the temperature and CO concentration probability distribution chart, combine the two charts to fit the regression curve of temperature and CO concentration, and obtain the situation between CO concentration and temperature under the condition of tobacco smoldering.

[0026] In one or more embodiments of the present invention, the user is notified via a network server or a mobile phone to obtain an early warning prompt.

[0027] Compared with the background technology, the effect of the present invention is: adopting the above-mentioned technical solution, using the analysis results of YOLOV7 of deep learning to verify the correctness of the evidence theory analysis, which is equivalent to fusing the results of AI and data analysis, combining the dual-scale analysis of the cigarette package status, and drawing the final conclusion, making the entire monitoring process more robust and the results more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 A schematic diagram of the positions of a temperature sensor and a CO sensor in one embodiment of the present invention;

[0029] Figure 2 A system operation block diagram in one embodiment of the present invention;

[0030] Figure 3 A schematic diagram of the improved YOLOv7 system structure in one embodiment of the present invention;

[0031] Those skilled in the art can understand the shapes, structures and schemes obtained from the figures shown in the drawings. The various components of the drawings are not necessarily to scale, and the sizes of the various components and elements of the drawings may be enlarged or reduced to more clearly illustrate the embodiments of the present invention described herein. DETAILED DESCRIPTION

[0032] The following references Figure 1-3 To describe specific embodiments, however, those skilled in the art will easily understand that the detailed description given herein with respect to these drawings is for illustrative purposes only and should not be understood as limiting. This embodiment provides a method for automatically detecting tobacco smoldering temperature and CO carbon intelligent perception in the present invention, which includes the following steps:

[0033] First, when the temperature sensor is installed in group A, multiple temperature sensors are formed into a triangular waveform network node at the geometric symmetry of the diagonal line of the cigarette pack, and a temperature difference AI detection algorithm is designed;

[0034] 2. When the CO carbon sensor is installed as group B, it is between two cigarette packs. When there is smoldering under or in the middle of the cigarette pack, the generated CO rises to the upper part of the space, and the CO carbon sensor detects CO.

[0035] 3. Before determining the temperature difference and CO value, first conduct repeated experiments to obtain the large data range of cigarette pack normal temperature, smoldering initial sensitive temperature, smoldering combustion temperature and CO;

[0036] Fourth, optimize the YOLOv7 neural network structure and parameters, establish a network model suitable for smoldering temperature and CO, and then use AI classification to find the data relationship model between smoldering temperature and CO;

[0037] Fifth, the deep learning data is finally integrated and the improved evidence theory is used to make feature data fusion decisions of temperature and CO, including the synthesis of conflicting information and data to inform the user of the type of smoldering state.

[0038] In one or more embodiments of the present invention, the data relationship model in step 4 includes CO values ​​at room temperature of the cigarette package, at the initial smoldering sensitive temperature, and at the combustion temperature.

[0039] In one or more embodiments of the present invention, in step 1, the temperature difference AI detection algorithm includes: designing a constant temperature sensor A0 at the inner door of the warehouse, obtaining the warehouse environment temperature T0 under normal conditions, and comparing it with the temperature measured by the temperature sensor in the cigarette pack to obtain the temperature difference. The following is a mathematical model for temperature difference design.

[0040] Assume the temperature difference variable parameter is T ij , the temperature sensor in the cigarette pack is A i , the temperature measured by the temperature sensor inside the cigarette pack is recorded as T i , where (i=n-1), this case takes 5 sensors as an example, then:

[0041] T i1 =T1-T0

[0042] T i2 =T2-T o

[0043] T i3 =T3-T0

[0044] T i4 =T4-T o ......

[0046] T i(n-1) =T n-1 -T0

[0047] Average temperature difference:

[0048]

[0049] When T ij -Δ T >C, or when T i When the temperature is equal to or greater than the smoldering start temperature T, smoldering may occur within the detection range of the temperature sensor to alert the user.

[0050] When reminding the user, it makes a fusion decision with the sensor data of Group B.

[0051] When reminding the user, it makes a fusion decision with the sensor data of Group B.

[0052] In step three, through repeated tobacco smoldering experiments, the temperature and CO concentration data of each situation were screened and classified by label, and the temperature was classified into normal initial temperature, smoldering critical temperature, and smoldering combustion temperature; CO concentration was also classified by label, and divided into three categories: normal initial concentration, smoldering critical temperature, and smoldering combustion concentration. The temperature and CO concentration image label data were pushed into the improved YOLOV7 neural network for training, and the trained data was analyzed.

[0053] In step 4, the YOLOV7 model is improved, the Focus module is modified to train the data in a targeted manner, and an output is traversed separately to improve efficiency. In addition, the slice module is sliced ​​horizontally, and the CO concentration and temperature arrays are converted into a set of two-dimensional arrays to strengthen data analysis and extraction, improve the connection between the two, compress and smooth the function data, reduce data redundancy, reduce the oscillation of the probability line, and make the data more credible. The results of the training are used to obtain the temperature and CO concentration probability distribution chart, and the two charts are combined to fit the regression curve of temperature and CO concentration, and the CO concentration and temperature are obtained under the condition of tobacco smoldering. The user is notified by the network server or mobile phone to get an early warning prompt.

[0054] Based on the above content and all the figures, the installation and operation model is further implemented:

[0055] Sensor installation: When the temperature sensor (set as group A) is installed and designed, it is geometrically symmetrical at the diagonal of the cigarette pack. Multiple temperature sensors form a triangular waveform network node. A small number of temperature sensors are used to detect the largest possible range, reducing material waste and costs while ensuring the detection quality. See the figure. The CO sensor (set as group B) is installed between two cigarette packs. When there is smoldering under or in the middle of the cigarette pack, CO will be generated. Since the air density is greater than the CO density, CO will rise to the upper part of the space and is easy to detect. See Figure 1 .

[0056] Mathematical calculation model of temperature difference sensor detection, firstly, a set of temperature difference sensor detection methods are designed, a constant temperature sensor A0 is designed for the inner door of the warehouse, the warehouse ambient temperature T0 under normal conditions is obtained, and the temperature measured by the temperature sensor in the cigarette pack is compared to obtain the temperature difference. The following is the mathematical model of temperature difference design.

[0057] Assume the temperature difference variable parameter is T ij , the temperature sensor in the cigarette pack is A i , the temperature measured by the temperature sensor inside the cigarette pack is recorded as T i , where (i=n-1), this case takes 5 sensors as an example, then:

[0058] T i1 =T1-T0

[0059] T i2 =T2-T0

[0060] T i3 =T3-T0

[0061] T i4 =T4-T0 ......

[0063] T i(n-1) =T n-1 -T0

[0064] Average temperature difference:

[0065]

[0066] When T ij -Δ T >C, or when T iWhen it is equal to or greater than the smoldering starting temperature T (for example: 100 degrees), it means that smoldering may occur within the detection range of the temperature sensor, reminding the user to make a fusion decision with the sensor data of group B. Specifically, after determining the temperature sensor with the highest sensing temperature, the improved evidence theory module is connected to make a fusion decision with the CO sensor data.

[0067] Improved YOLOv7 model based on smoldering temperature and CO features: Through a large number of repeated tobacco smoldering experiments, the temperature and CO concentration data obtained in each case were screened and classified by label, and the temperature was classified into normal initial temperature, smoldering critical temperature, and smoldering combustion temperature. The CO concentration was also labeled and classified into three categories: normal initial concentration, smoldering critical temperature, and smoldering combustion concentration. The temperature and CO concentration image label data were pushed into the improved YOLOV7 neural network for training, and the trained data was analyzed.

[0068] The improvement of the YOLOV7 model specifically modifies the Focus module to conduct targeted training on the data, traverses an output separately to improve efficiency, and slices the slice module horizontally, converting the CO concentration and temperature arrays into a set of two-dimensional arrays, strengthening data analysis and extraction, improving the connection between the two, compressing and smoothing the function data, reducing data redundancy, reducing the oscillation of the probability line, and making the data more credible. The results after training are then used to obtain the temperature and CO concentration probability distribution chart, and the two charts are combined to fit the regression curve of temperature and CO concentration to obtain the situation between CO concentration and temperature under the condition of tobacco smoldering. According to the curve, the specific data of temperature and CO concentration are preliminarily obtained, and the range of the cigarette package from normal temperature to the critical temperature of smoldering, and the range of the critical temperature of smoldering to smoldering combustion are obtained from the interval of the two sets of data. Then, the smoldering state judgment of the cigarette package is carried out at the AI ​​scale, so as to realize verification and fusion with the state judgment of evidence theory decision-making.

[0069] Using improved evidence theory to make decision on temperature and CO characteristic data fusion:

[0070] 1. Data collection and processing

[0071] First, it is necessary to conduct experiments on temperature sensors and carbon monoxide sensors (hereinafter collectively referred to as CO sensors) and collect statistical data. The number of experimental groups is greater than or equal to 1000 groups, and then process the data of the two sensors separately.

[0072] For temperature data, we first count the temperature ranges of the sensor in four situations: room temperature, normal temperature inside the cigarette pack, critical smoldering temperature, and formal smoldering temperature. The purposes of setting these four situations are as follows:

[0073] ① Room temperature: This indicates that the temperature sensor may not be placed correctly. Because there must be a temperature difference between the temperature inside the tobacco pack in the tobacco storage warehouse and the ambient temperature of the warehouse, it is necessary to measure and count the room temperature under a certain environment.

[0074] ② Normal temperature inside the cigarette pack: This situation indicates that the temperature is normal and there is no danger of tobacco leaves smoldering.

[0075] ③ Smoldering critical temperature: This situation does not necessarily mean that smoldering has occurred. The temperature of the cigarette package may rise to a temperature close to smoldering due to other factors, such as environmental changes, sensor errors, movement during manual inspection, etc. Therefore, in this case, an early warning prompt of the critical state is issued, and it is necessary to manually determine the condition of the cigarette package.

[0076] ④ Formal smoldering temperature: This indicates that smoldering has occurred, and a smoldering combustion status warning prompt is issued. The smoldering alarm should be directly connected and manually inspected and handled.

[0077] All experimental data are processed according to the above situation, and four temperature intervals are obtained: A (room temperature), B (normal temperature inside the cigarette package), C (critical smoldering temperature), and D (formal smoldering temperature). These four temperature intervals are allowed to have overlapping areas. The method of judging the overlapping areas is to calculate the probability by weighting the analysis of the quantity of the two situations.

[0078] For CO data, the number of experimental groups is the same as that of temperature data, but when dealing with the situation, it is divided into three situations: normal concentration, critical smoldering concentration, and formal smoldering concentration. The situation descriptions are as follows:

[0079] ①Normal concentration: This concentration is the normal concentration of CO in the tobacco storage environment.

[0080] ② Smoldering critical concentration: This concentration indicates that local initial smoldering may have occurred, but it may also be caused by other factors, such as environmental changes, CO sensor errors, and movement during manual inspection. Therefore, in this case, it is necessary to manually determine the condition of the cigarette pack.

[0081] ③ Formal smoldering concentration: This concentration has reached the point where the tobacco leaves are completely smoldering, and should be directly connected to the smoldering alarm and manually inspected and handled.

[0082] The CO concentration is divided into the above situations and three concentration intervals are obtained: E (normal concentration), F (critical smoldering concentration), and G (formal smoldering concentration). These three concentration intervals are allowed to have overlapping areas. The method for judging the overlapping areas is to calculate the probability by weighting the analysis of the quantity of the two situations.

[0083] Based on the decision of temperature and CO concentration, based on the improved system:

[0084] In the actual application of monitoring, taking the number of sensors designed in the diagram as an example, the data of five sensors and three CO sensors are first monitored in real time, and the sensor data that maintains the maximum value in temperature and CO concentration for a certain period of time is selected for analysis. Three temperature data are collected for this sensor, and the probability that these three data fall into the above four temperature conditions (ABCD) is obtained, that is, three groups of BPA functions (evidence). Then, the improved evidence theory is used to make decisions to judge the temperature of the cigarette package. The same is true for the treatment of CO concentration. Three groups of evidence are also used to make decisions to judge the CO concentration (EFG). Then A and B in the temperature are collectively referred to as normal, that is, the temperature is divided into the same three conditions as the CO concentration (normal, critical, and smoldering). Then, the evidence of temperature and CO concentration is integrated and analyzed to draw conclusions and feedback specific information about the cigarette package (normal, critical, and smoldering). The specific improvements to the evidence theory are as follows:

[0085] ① The normalization factor of the traditional DS evidence combination rule is cancelled, and the traditional factor multiplication is changed into factor addition through factor decomposition to perform a form transformation. In this way, when there is a high degree of conflict between the evidences, extreme errors will not occur.

[0086] ② The large data sets of temperature and CO concentration were integrated, and sufficient evidence sources were added to reduce the conflicts between evidences, making the decision results more reasonable and accurate.

[0087] ③ Improve and construct new synthesis rules:

[0088] m(Z)=g(Z)+kλf(z)(Z≠φ,x)

[0089] m(x)=Δf(x)

[0090] m(φ)=0

[0091] Fusion deep learning results decision

[0092] After the conclusion based on the fusion of temperature and CO concentration information is drawn in the above steps, it is combined with the judgment conclusion of deep learning. Deep learning and evidence theory analysis are also divided into three situations: normal, critical, and smoldering. The analysis results of YOLOV7 based on deep learning are used to verify the correctness of the evidence theory analysis, which is equivalent to fusing the results of AI and data analysis, combining the dual-scale analysis of the cigarette package status, and drawing the final conclusion, making the entire monitoring process more robust and the results more accurate.

[0093] Experimental results

[0094] Take temperature as an example

[0095] Set three sets of temperature data evidence, namely m1 (150℃), m2 (50℃), and m3 (30℃)

[0096] Temperature conditions:

[0097] A(Room temperature)

[0098] B (normal temperature inside cigarette pack)

[0099] C(smoldering critical temperature)

[0100] D (official smoldering temperature)

[0101] In order to have the same number as the CO concentration situation in the later stage, situations A and B are combined and collectively referred to as normal, which are divided into the following three situations:

[0102] Q1(Normal)

[0103] Q2(critical)

[0104] Q3(smoldering)

[0105] ① Evidence of the correspondence between the three groups of temperatures

[0106]

[0107] ② Improved evidence theory fusion judgment results

[0108]

[0109] For the purpose of illustration, the foregoing description has been thoroughly understood by those skilled in the art of the embodiments. However, it will be apparent to those skilled in the art that specific details are not required in order to practice the embodiments. Those skilled in the art should understand that the present invention is not limited to the specific embodiments described above, and improvements and substitutions based on the present invention using the known techniques in the art fall within the scope of the present invention and should be defined by the claims.

Claims

1. An automatic detection method for tobacco smoldering temperature and CO carbon intelligent perception, characterized in that: It includes the following steps:

1. When the temperature sensor is installed as group A, a plurality of temperature sensors are formed into a triangular waveform network node at the geometric symmetry of the diagonal line of the cigarette package, and a temperature difference AI detection algorithm is designed; the temperature difference AI detection algorithm includes: designing a constant temperature sensor A0 at the inner door of the warehouse, obtaining the warehouse environment temperature T0 under normal conditions, and comparing it with the temperature measured by the temperature sensor in the cigarette package to obtain the temperature difference. The following is the mathematical model of the temperature difference design; Assume the temperature difference variable parameter is T ij , the temperature sensor in the cigarette pack is A i , the temperature measured by the temperature sensor inside the cigarette pack is recorded as T i , where i = n-1, and the number of temperature sensors is 5, then: T i1 =T1-T0 T i2 =T2-T0 T i3 =T3-T0 T i4 =T4-T0 ...... T i(n-1) =T n-1 -T0 Average temperature difference: When T ij -Δ T >C, or when T i When it is equal to or greater than the smoldering start temperature T, smoldering may occur within the detection range of the temperature sensor to alert the user; 2. When the CO carbon sensor is installed as group B, it is between two cigarette packs. When there is smoldering under or in the middle of the cigarette pack, the generated CO rises to the upper part of the space, and the CO carbon sensor detects CO.

3. Before determining the temperature difference and CO value, first obtain the big data range of normal temperature of cigarette packs, smoldering initial sensitive temperature, smoldering combustion temperature and CO through repeated experiments; through repeated tobacco smoldering experiments, screen the obtained temperature and CO concentration data of each situation and classify the temperature by label, which is divided into normal initial temperature, smoldering critical temperature and smoldering combustion temperature; then classify the CO concentration by label, which is divided into three categories, normal initial concentration, smoldering critical concentration and smoldering combustion concentration; push the temperature and CO concentration image label data into the improved YOLOV7 neural network for training, and analyze the trained data; Fourth, optimize the YOLOv7 neural network structure and parameters, establish a network model suitable for smoldering temperature and CO, and then use AI classification to find the data relationship model between smoldering temperature and CO; including improvements to the YOLOV7 model, modify the Focus module to conduct targeted training on the data, traverse an output separately to improve efficiency, and slice the slice module horizontally, convert the CO concentration and temperature arrays into a set of two-dimensional arrays, strengthen data analysis and extraction, improve the connection between the two, compress and smooth the function data, reduce data redundancy, reduce the probability line oscillation, make the data more credible, and then use the training results to obtain the temperature and CO concentration probability distribution chart, and combine the two charts to fit the regression curve of temperature and CO concentration. Get the CO concentration and temperature under the condition of tobacco smoldering; Fifth, the deep learning data is finally integrated and the improved evidence theory is used to make feature data fusion decisions of temperature and CO, including the synthesis of conflicting information and data to inform the user of the type of smoldering state.

2. The automatic detection method of tobacco smoldering temperature and CO carbon intelligent perception according to claim 1 is characterized in that: The data relationship model in step 4 includes the CO value at room temperature of the cigarette package, the CO value at the initial smoldering sensitive temperature, and the CO value at the combustion temperature.

3. The automatic detection method of tobacco smoldering temperature and CO carbon intelligent perception according to claim 2 is characterized in that: When reminding the user, it makes a fusion decision with the sensor data of Group B.

4. The automatic detection method of tobacco smoldering temperature and CO carbon intelligent perception according to claim 3 is characterized in that: The user is notified via a network server or mobile phone to receive an early warning prompt.

Images