Intelligent system based on multi-source sensor data analysis

Abstract

The present invention relates to the technical field of sensor data analysis, and in particular to an intelligent system based on multi-source sensor data analysis, comprising a data collection module, a temperature impact analysis and flame detection module, and a smoke and gas impact analysis fusion module, wherein: the data collection module acquires multi-source sensor data, wherein the multi-source sensor data comprises flame image data, temperature data, smoke data, and gas data; the temperature impact analysis and flame detection module is used to perform weight modification on flame image data according to temperature data to obtain flame weight image data; based on the flame weight image data, flame detection results of each area are calculated, wherein the flame detection results include confidence; the smoke and gas impact analysis fusion module performs abnormality detection on smoke data and gas data, and according to the detection result, increases the confidence in the flame detection results of the corresponding area in the flame detection results.

Description

Technical Field

[0001] The present invention relates to the technical field of sensor data analysis, and in particular to an intelligent system based on multi-source sensor data analysis. Background Art

[0002] In traditional fire monitoring and early warning systems, fire detection relies on a single type of sensor (such as a smoke sensor or temperature sensor), which is prone to false alarms and missed alarms in complex environments. Specifically, smoke sensors are very sensitive to dust, steam, and aerosols in the air. When these factors exist, the sensor may misjudge it as fire smoke, thereby triggering unnecessary alarms. For example, cooking steam in the kitchen, hot water vapor in the bathroom, or smoke generated by equipment in an industrial environment can all cause smoke sensors to falsely alarm.

[0003] In addition, smoke sensors frequently send out false alarms in conditions of poor ventilation or the presence of other aerosol sources (e.g., cigarettes, spray, etc.). Temperature sensors are prone to miss fire reports when the ambient temperature varies greatly. Normal temperature fluctuations in a hot environment, direct sunlight, or the influence of equipment heat sources may prevent temperature sensors from accurately distinguishing between normal high temperatures and high temperatures caused by fire, resulting in missed reports.

[0004] In addition, equipment failures of temperature sensors, such as power problems, line failures, or sensor aging, can also cause temperature data to become invalid, making it impossible to detect the temperature rise at the beginning of a fire in a timely manner. The limitation of these single sensors is that they lack comprehensive judgment capabilities and cannot accurately identify fires under the interference of multiple environmental factors. For example, relying solely on smoke sensors is prone to false alarms in environments with dust and steam, while relying solely on temperature sensors is prone to missed alarms when the ambient temperature fluctuates greatly, thus failing to provide comprehensive and accurate fire detection results.

[0005] Therefore, in a complex and changing environment, the traditional single sensor system is prone to false alarms or missed alarms due to interference from a single factor, which greatly reduces the accuracy and reliability of fire detection. Summary of the invention

[0006] The purpose of the present invention is to provide an intelligent system based on multi-source sensor data analysis to solve the problems raised in the above background technology.

[0007] To achieve the above object, the present invention provides the following technical solution: an intelligent system based on multi-source sensor data analysis, including a data collection module, a temperature impact analysis and flame detection module and a smoke gas impact analysis fusion module, wherein:

[0008] The data collection module acquires multi-source sensor data, wherein the multi-source sensor data includes flame image data, temperature data, smoke data and gas data;

[0009] The temperature impact analysis and flame detection module is used to modify the weight of the flame image data according to the temperature data, detect abnormally high temperature of the temperature data, and increase the weight of the corresponding area in the flame image data according to the detection result, as the flame weight image data; calculate the flame detection result of each area according to the flame weight image data, wherein the flame detection result includes the confidence level;

[0010] The smoke and gas impact analysis fusion module combines the smoke data and the gas data to verify the flame weight image data, performs anomaly detection on the smoke data and the gas data, and increases the confidence in the flame detection results of the corresponding area in the flame detection results based on the detection results.

[0011] As a further improvement of the technical solution, the multi-source sensor data in the data collection module includes a timestamp, which is used to synchronously process different data in the multi-source sensor data.

[0012] As a further improvement of the technical solution, the temperature impact analysis and flame detection module includes a temperature impact analysis unit, and the process of the temperature impact analysis unit generating flame weight image data specifically includes:

[0013] Set a temperature threshold; compare the acquired temperature data with the temperature threshold. If the temperature data exceeds the temperature threshold, it is marked as an abnormally high temperature area;

[0014] Obtain the location information of the temperature sensor in the abnormally high temperature area, and map the temperature reading of the temperature sensor to the corresponding pixel position of the flame image according to the location information , and calculate the corresponding position weight to generate flame weight image data.

[0015] As a further improvement of the present technical solution, the calculation formula of the flame detection weight is as follows:

[0016] ,in Indicates the pixel position The flame detection weight; represents the weight adjustment coefficient; Indicates the temperature reading of the temperature sensor; Indicates the temperature threshold.

[0017] As a further improvement of the technical solution, the temperature impact analysis and flame detection module includes a flame detection unit, and the process of the flame detection unit calculating the flame detection result specifically includes:

[0018] In the process of calculating the flame detection results, the flame weight image data is input into the pre-trained flame detection model for processing. The flame detection model uses a multi-scale convolutional neural network;

[0019] The flame weight image data passes through multiple convolutional layers, pooling layers, and upsampling layers in the flame detection model to gradually extract low-level and high-level features in the image as feature maps;

[0020] In each convolutional layer, the flame detection model uses activation functions to perform nonlinear transformations on feature maps to enhance the expressiveness of features. The feature maps are gradually restored to the resolution of the original image through upsampling layers to generate flame detection results.

[0021] Each pixel in the flame detection result corresponds to a confidence level , which indicates the probability that the pixel belongs to the flame area. For each confidence level in the flame detection result Determine the confidence threshold.

[0022] As a further improvement of the technical solution, the determination process of the confidence threshold is as follows:

[0023] If the flame detection confidence of a certain pixel in the flame detection result exceeds a preset confidence threshold, the area is marked as a flame area and a corresponding flame detection result is generated.

[0024] As a further improvement of the technical solution, the calculation of the confidence is completed by the last layer of the flame detection model, which converts the feature map output by the previous layer into a probability map, and in the case of the sigmoid activation function, the output value ranges from 0 to 1, indicating the probability that the pixel is a flame, where the probability value is the confidence .

[0025] As a further improvement of the technical solution, the process of increasing the confidence level by the smoke gas impact analysis fusion module specifically includes:

[0026] Use timestamps to synchronize flame image data, smoke data, and gas data to ensure that the three types of data are processed at the same time point; set smoke concentration thresholds and gas concentration thresholds to identify abnormal smoke and gas areas;

[0027] Compare the smoke data with the smoke concentration threshold, detect the area where the smoke concentration exceeds the threshold, and use it as the abnormal smoke area;

[0028] Compare the gas data with the gas concentration threshold, and detect the area where the gas concentration exceeds the threshold as the abnormal gas area;

[0029] Obtain the location information of the smoke sensor in the abnormal smoke area and the location information of the gas sensor in the abnormal gas area, and map the location information of the smoke sensor and the gas sensor to the corresponding pixel position of the flame image , and the pixel position in the flame detection result 's confidence level is increased.

[0030] As a further improvement of the technical solution, the formula for increasing the confidence is as follows:

[0031] ,in Indicates the smoke concentration reading of the smoke sensor; Indicates the gas concentration reading of the gas sensor; Indicates the smoke concentration threshold; Indicates the gas concentration threshold; and They represent the smoke concentration adjustment coefficient and gas concentration adjustment coefficient respectively.

[0032] As a further improvement of the present technical solution, the smoke and gas impact analysis fusion module is used to confirm the final flame area, and re-determine the confidence threshold based on the flame detection result after adjusting the confidence. If the flame detection confidence of a pixel point in the flame detection result exceeds the preset confidence threshold, the area is marked as the final flame area and an alarm is triggered immediately.

[0033] Compared with the prior art, the present invention has the following beneficial effects:

[0034] The intelligent system based on multi-source sensor data analysis uses a multi-scale convolutional neural network to process flame weight image data, and gradually extracts low-level and high-level features through multiple convolutional layers, pooling layers, and upsampling layers to generate feature maps. In this process, temperature data affects the generation of feature maps through a weight adjustment mechanism, making the confidence of flame detection results more reliable. This multi-source data fusion method not only improves the confidence of flame detection, but also enhances the system's adaptability to complex environments.

[0035] By fusing smoke data and gas data with flame detection results, the confidence of each area in the flame detection results is adjusted; if an abnormality is detected in both smoke and gas data in a certain area, the module will significantly increase the flame detection confidence of that area, thereby improving the accuracy of the system's judgment of fire. This multi-source data fusion analysis method can effectively overcome the limitations of a single sensor in a complex environment and provide more reliable and comprehensive fire detection results. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a schematic diagram of the overall module of the present invention;

[0037] Figure 2 It is a schematic diagram of the temperature impact analysis and flame detection module unit of the present invention.

[0038] In the figure: 100, data collection module; 200, temperature impact analysis and flame detection module; 201, temperature impact analysis unit; 202, flame detection unit; 300, smoke gas impact analysis fusion module. DETAILED DESCRIPTION

[0039] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0040] Next, see Figure 1-Figure 2 The present invention provides a technical solution: an intelligent system based on multi-source sensor data analysis, including a data collection module 100, a temperature impact analysis and flame detection module 200 and a smoke gas impact analysis fusion module 300.

[0041] The data collection module 100 is the basis of the entire system and is responsible for collecting data from multiple sensors as multi-source sensor data. The collected multi-source sensor data includes flame image data, temperature data, smoke data and gas data, among which:

[0042] The flame image data is collected by the camera installed in the monitoring area. The flame image data collected by the camera is stored in the form of RGB or grayscale images. Multiple frames of images can be collected per second to form a video stream.

[0043] Temperature data is collected by temperature sensors installed in the monitoring area and stored in numerical form (e.g., Celsius, Fahrenheit). Each sensor collects temperature readings once per second;

[0044] Smoke data is collected by smoke sensors installed in the monitoring area and stored in concentration values ​​(such as mg / m³). Each sensor collects smoke concentration readings once a second.

[0045] Gas data is collected by gas sensors installed in the monitoring area and stored in concentration values ​​(such as ppm). Each sensor collects gas concentration readings once per second.

[0046] All data collected by sensors are time-stamped. The data of each sensor is aligned by time-stamp to ensure that the data at the same time point can be fused and processed.

[0047] The temperature impact analysis unit 201 in the temperature impact analysis and flame detection module 200 is used to modify the weight of the flame image data according to the temperature data, by detecting abnormally high temperature of the temperature data, and according to the detection result, increasing the weight of the corresponding area in the flame image data as the flame weight image data, specifically including:

[0048] Extract timestamp information from flame image data and temperature data, and calibrate data by timestamp to ensure that data at the same time point can be processed together. For example, if the timestamp difference exceeds 1 second, the data is considered to be out of sync and needs to be adjusted.

[0049] Set a temperature threshold in the system configuration, such as 60°C; compare the acquired temperature data with the temperature threshold, and if the temperature data exceeds the temperature threshold, mark it as an abnormally high temperature area. For example, if the temperature reading in a certain area is 75°C, mark the area as an abnormally high temperature area;

[0050] Obtain the location information of the temperature sensor in the abnormally high temperature area, and map the temperature reading of the temperature sensor to the corresponding pixel position of the flame image according to the location information , and calculate the corresponding position weight to generate flame weight image data, where the calculation formula is as follows:

[0051] ,in Indicates the pixel position The flame detection weight; represents the weight adjustment coefficient; Indicates the temperature reading of the temperature sensor; Indicates the temperature threshold.

[0052] The flame detection unit 202 in the temperature impact analysis and flame detection module 200 calculates the flame detection result of each area according to the flame weight image data, wherein the flame detection result includes the confidence level. The specific process includes:

[0053] In the process of calculating the flame detection results, the flame weight image data is input into the pre-trained flame detection model for processing; the flame detection model adopts a multi-scale convolutional neural network, which can effectively capture the characteristics of flames at different scales through multi-level feature extraction and fusion, thereby improving the detection accuracy;

[0054] The flame weight image data passes through multiple convolutional layers, pooling layers, and upsampling layers in the flame detection model to gradually extract low-level features (such as edges and textures) and high-level features (such as the unique shape and color distribution of the flame) in the image as feature maps;

[0055] In each convolutional layer, the flame detection model uses activation functions (such as ReLU) to perform nonlinear transformations on the feature map to enhance the expressiveness of the features. Subsequently, the feature map is gradually restored to the resolution of the original image through the upsampling layer to generate the flame detection results.

[0056] Each pixel in the flame detection result corresponds to a confidence level , indicating the probability that the pixel belongs to the flame area; for example, the flame detection confidence of a pixel is 0.8, which means that there is an 80% probability that there is a flame at the pixel; the confidence calculation is completed by the last layer of the flame detection model (usually the sigmoid or softmax layer), which converts the feature map output by the previous layer into a probability map, and in the case of the sigmoid activation function, the output value ranges from 0 to 1, indicating the probability that the pixel is a flame, where the probability value is the confidence ;

[0057] For each confidence level in the flame detection result A confidence threshold is determined. If the flame detection confidence of a pixel in the flame detection result exceeds a preset confidence threshold (such as 0.5), the area is marked as a flame area and a corresponding flame detection result is generated; these results will be transmitted to the subsequent smoke and gas impact analysis fusion module 300 for further verification.

[0058] The temperature impact analysis and flame detection module 200 solves the problem that flame image data in traditional systems is easily affected by changes in ambient light and visual interference, leading to false alarms, by combining temperature data and flame image data; the temperature impact analysis unit 201 in this module detects abnormally high temperature areas by setting temperature thresholds and marks these areas; the flame detection unit 202 uses a multi-scale convolutional neural network to process flame weight image data, gradually extracts low-level and high-level features, and generates high-confidence flame detection results; in this way, the system can not only identify flames more accurately, but also reduce the false alarm rate and improve the sensitivity and reliability of fire detection through the assistance of temperature data.

[0059] The smoke and gas impact analysis fusion module 300 combines the smoke data and the gas data to verify the flame weight image data, performs anomaly detection on the smoke data and the gas data, and increases the confidence in the flame detection results of the corresponding area in the flame detection results according to the detection results, specifically including:

[0060] Use timestamps to synchronize flame image data, smoke data, and gas data to ensure that the three types of data are processed at the same time point; for example, by extracting and comparing timestamps, ensure that the time difference between flame image data and smoke data and gas data is within the allowable range (e.g., within 1 second); set smoke concentration thresholds (e.g., 0.5 mg / m³) and gas concentration thresholds (e.g., 100 ppm) to identify abnormal smoke and gas areas;

[0061] Compare the smoke data with the smoke concentration threshold, detect areas where the smoke concentration exceeds the threshold, and mark them as abnormal smoke areas; for example, if the smoke concentration reading in a certain area is 0.8mg / m³, which exceeds the preset 0.5mg / m³ threshold, the area is marked as an abnormal smoke area;

[0062] Compare the gas data with the gas concentration threshold, and detect areas where the gas concentration exceeds the threshold as abnormal gas areas; for example, if the gas concentration reading in a certain area is 150ppm, which exceeds the preset threshold of 100ppm, then the area is marked as an abnormal gas area;

[0063] Obtain the location information of the smoke sensor in the abnormal smoke area and the location information of the gas sensor in the abnormal gas area, and map the location information of the smoke sensor and the gas sensor to the corresponding pixel position of the flame image , and the pixel position in the flame detection result The confidence level is increased by the following formula:

[0064] ,in Indicates the smoke concentration reading of the smoke sensor; Indicates the gas concentration reading of the gas sensor; Indicates the smoke concentration threshold; Indicates the gas concentration threshold; and They represent the smoke concentration adjustment coefficient and gas concentration adjustment coefficient respectively.

[0065] The smoke and gas impact analysis fusion module 300 generates a flame detection result including an adjusted confidence level, and re-determines the confidence level in the adjusted flame detection result based on the confidence level threshold. If the flame detection confidence level of a pixel in the flame detection result exceeds the preset confidence level threshold, the area is marked as the final flame area and an alarm is immediately triggered to help firefighters respond quickly.

[0066] The smoke and gas impact analysis fusion module 300 further verifies the flame detection results by combining smoke data and gas data, solving the problem of false alarms and missed alarms when smoke sensors and gas sensors work alone in traditional systems. The module uses anomaly detection methods to identify abnormal smoke and gas areas by setting smoke concentration thresholds and gas concentration thresholds, and maps the location information of these areas to the corresponding pixel positions of the flame image, thereby increasing the confidence in the flame detection results. In this way, the system can not only reduce the false alarms of a single sensor through the fusion of multi-source data, but also improve the accuracy and early detection capabilities of flame detection in the early stages of a fire with the assistance of smoke and gas data.

[0067] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and descriptions are only preferred examples of the present invention and are not intended to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. An intelligent system based on multi-source sensor data analysis, characterized in that: It comprises a data collection module (100), a temperature impact analysis and flame detection module (200) and a smoke gas impact analysis fusion module (300), wherein: The data collection module (100) acquires multi-source sensor data, wherein the multi-source sensor data includes flame image data, temperature data, smoke data and gas data; The temperature impact analysis and flame detection module (200) is used to modify the weight of flame image data according to temperature data, detect abnormally high temperature of temperature data, and increase the weight of corresponding areas in flame image data according to the detection result to obtain flame weight image data; and calculate flame detection results of each area according to the flame weight image data, wherein the flame detection results include confidence levels; The smoke and gas impact analysis fusion module (300) combines smoke data and gas data to verify flame weight image data, performs abnormality detection on the smoke data and gas data, and increases the confidence in the flame detection results of the corresponding area in the flame detection results based on the detection results.

2. The intelligent system based on multi-source sensor data analysis according to claim 1 is characterized in that: The multi-source sensor data in the data collection module (100) includes a timestamp, which is used to synchronously process different data in the multi-source sensor data.

3. The intelligent system based on multi-source sensor data analysis according to claim 2 is characterized in that: The temperature impact analysis and flame detection module (200) comprises a temperature impact analysis unit (201), and the process of the temperature impact analysis unit (201) generating flame weight image data specifically comprises: Set a temperature threshold; compare the acquired temperature data with the temperature threshold. If the temperature data exceeds the temperature threshold, it is marked as an abnormally high temperature area; Obtain the location information of the temperature sensor in the abnormally high temperature area, and map the temperature reading of the temperature sensor to the corresponding pixel position of the flame image according to the location information , and calculate the corresponding position weight to generate flame weight image data.

4. The intelligent system based on multi-source sensor data analysis according to claim 2 is characterized in that: The calculation formula of the flame detection weight is as follows: ,in Indicates the pixel position The flame detection weight; represents the weight adjustment coefficient; Indicates the temperature reading of the temperature sensor; Indicates the temperature threshold.

5. The intelligent system based on multi-source sensor data analysis according to claim 4 is characterized in that: The temperature impact analysis and flame detection module (200) comprises a flame detection unit (202), and the process of the flame detection unit (202) calculating the flame detection result specifically comprises: In the process of calculating the flame detection results, the flame weight image data is input into the pre-trained flame detection model for processing. The flame detection model uses a multi-scale convolutional neural network; The flame weight image data passes through multiple convolutional layers, pooling layers, and upsampling layers in the flame detection model to gradually extract low-level and high-level features in the image as feature maps; In each convolutional layer, the flame detection model uses activation functions to perform nonlinear transformations on feature maps to enhance the expressiveness of features. The feature maps are gradually restored to the resolution of the original image through upsampling layers to generate flame detection results. Each pixel in the flame detection result corresponds to a confidence level , which indicates the probability that the pixel belongs to the flame area. For each confidence level in the flame detection result Determine the confidence threshold.

6. The intelligent system based on multi-source sensor data analysis according to claim 5 is characterized in that: The confidence threshold determination process is as follows: If the flame detection confidence of a certain pixel in the flame detection result exceeds a preset confidence threshold, the area is marked as a flame area and a corresponding flame detection result is generated.

7. The intelligent system based on multi-source sensor data analysis according to claim 6, characterized in that: The confidence is calculated by the last layer of the flame detection model, which converts the feature map output by the previous layer into a probability map, and in the case of the sigmoid activation function, the output value ranges from 0 to 1, indicating the probability that the pixel is a flame, where the probability value is the confidence .

8. The intelligent system based on multi-source sensor data analysis according to claim 7 is characterized in that: The process of increasing the confidence level by the smoke and gas impact analysis fusion module (300) specifically includes: Use timestamps to synchronize flame image data, smoke data, and gas data to ensure that the three types of data are processed at the same time point; set smoke concentration thresholds and gas concentration thresholds to identify abnormal smoke and gas areas; Compare the smoke data with the smoke concentration threshold, detect the area where the smoke concentration exceeds the threshold, and use it as the abnormal smoke area; Compare the gas data with the gas concentration threshold, and detect the area where the gas concentration exceeds the threshold as the abnormal gas area; Obtain the location information of the smoke sensor in the abnormal smoke area and the location information of the gas sensor in the abnormal gas area, and map the location information of the smoke sensor and the gas sensor to the corresponding pixel position of the flame image , and the pixel position in the flame detection result 's confidence level is increased.

9. The intelligent system based on multi-source sensor data analysis according to claim 8, characterized in that: The formula for increasing the confidence is as follows: ,in Indicates the smoke concentration reading of the smoke sensor; Indicates the gas concentration reading of the gas sensor; Indicates the smoke concentration threshold; Indicates the gas concentration threshold; and They represent the smoke concentration adjustment coefficient and gas concentration adjustment coefficient respectively.

10. The intelligent system based on multi-source sensor data analysis according to claim 9, characterized in that: The smoke and gas impact analysis fusion module (300) is used to confirm the final flame area, and re-determine the confidence threshold according to the flame detection result after adjusting the confidence; if the flame detection confidence of a pixel point in the flame detection result exceeds a preset confidence threshold, the area is marked as the final flame area, and an alarm is immediately triggered.

Images