A method and system for adaptive adjustment of optical power of a laser methane detector

Abstract

The application provides a light power adaptive adjustment method and system of a laser methane detector, and relates to the technical field of gas detection. The method comprises the following steps: collecting ambient light data, target distance information and reflected laser signals, and calculating signal-to-noise ratio data; then inputting the ambient light data, the target distance information and the signal-to-noise ratio data into a convolutional neural network for feature extraction to obtain multi-dimensional environmental features; then inputting the multi-dimensional environmental features into a fuzzy logic processor for scene recognition and outputting a scene recognition result; finally, querying a preset mapping table according to the scene recognition result to obtain a target value of a driving current, combining an actual current value to determine a current adjustment amount, and adjusting the driving current of an emission module according to the current adjustment amount, so as to realize the light power adaptive adjustment of the laser methane detector. The application improves the detection stability and adaptability of the laser methane detector in different complex environments.

Description

Technical Field

[0001] This application relates to the field of gas detection technology, and in particular to a method and system for adaptive adjustment of optical power of a laser methane detector. Background Technology

[0002] Laser methane detectors play an important role in safety monitoring in industries such as chemical and gas processing. They identify leaks by detecting the absorption of specific wavelength laser light by methane gas, and have the advantages of high sensitivity and fast response. As the application scenarios expand from open areas to complex environments such as tunnels, pipelines, and indoor corridors, maintaining the stability of the detection signal has become crucial.

[0003] In existing technologies, solutions collect ambient light intensity information and adjust the laser's output power based on a pre-set threshold. Alternatively, methods combine an ambient light sensor with a distance measurement unit to control the light power through a fixed mapping relationship. These methods can meet basic requirements when environmental conditions change gradually.

[0004] However, the existing methods described above suffer from insufficient precision and timeliness in power adjustment when faced with complex and changing field conditions, such as strong ambient reflected light, dynamic changes in the distance to the target object, and complex background noise. This leads to fluctuations in the detection signal-to-noise ratio, affecting the reliability and stability of the final detection results. Therefore, the existing technology suffers from insufficient adaptability to complex working conditions. Summary of the Invention

[0005] This application provides a method and system for adaptive adjustment of optical power of a laser methane detector, which solves the problems of poor detection stability and poor adaptability of laser methane detectors in different complex environments in the prior art.

[0006] To address the aforementioned technical problems, in a first aspect, this application provides a method for adaptive adjustment of optical power in a laser methane detector, comprising:

[0007] Collect ambient light data, target distance information, and reflected laser signals;

[0008] Based on the ambient light data and the reflected laser signal, the signal-to-noise ratio data is calculated;

[0009] The ambient light data, target distance information, and signal-to-noise ratio data are input into a convolutional neural network for feature extraction to obtain multi-dimensional environmental features.

[0010] The multi-dimensional environmental features are input into a fuzzy logic processor for scene recognition, and the scene recognition result is output.

[0011] Based on the scene recognition results, a preset mapping table is queried to obtain the target value of the driving current of the emission module in the laser methane detector, and the current adjustment amount is determined based on the target value and the actual value of the driving current.

[0012] The driving current of the emission module is adjusted based on the current regulation amount to achieve adaptive adjustment of the optical power of the laser methane detector.

[0013] Optionally, the step of inputting the multi-dimensional environmental features into a fuzzy logic processor for scene recognition and outputting the scene recognition result includes:

[0014] The multi-dimensional environmental features are input to the fuzzification interface of the fuzzy logic processor. Through multiple preset membership functions in the fuzzification interface, the light intensity distribution features, distance change features, and signal quality features in the multi-dimensional environmental features are converted into corresponding fuzzy sets.

[0015] All fuzzy sets are input into the rule base of the fuzzy logic processor. The rule base contains multiple preset inference rules. Inference operations are performed on all fuzzy sets according to the inference rules to obtain fuzzy output results. Based on the fuzzy output results, scene recognition results are output.

[0016] Optionally, the step of performing inference operations on all fuzzy sets according to the inference rules to obtain fuzzy output results, and outputting scene recognition results based on the fuzzy output results, includes:

[0017] The confidence level of the corresponding conclusion part is calculated based on the degree to which the fuzzy set satisfies the corresponding condition part in the inference rule.

[0018] The confidence scores of the conclusion parts of all fuzzy sets are combined to generate fuzzy output results;

[0019] Based on the fuzzy output results, the improved centroid method is used to calculate multiple definite values ​​corresponding to multiple scene categories, forming a sequence of definite values.

[0020] The determined value sequence is input into a gated recurrent unit network for temporal feature enhancement to generate an enhanced determined value sequence.

[0021] Based on the enhanced sequence of determined values, a decision tree model is used for classification to generate scene recognition results.

[0022] Optionally, the step of classifying the enhanced sequence of determined values ​​using a decision tree model to generate scene recognition results includes:

[0023] The enhanced sequence of determined values ​​is input into a decision tree model, which contains a root node, multiple branch nodes and multiple leaf nodes. Each branch node corresponds to an attribute judgment condition, and each leaf node corresponds to a scene category.

[0024] The decision tree model starts from the root node, and sequentially determines whether the attribute judgment conditions of the branch node are met according to the enhanced sequence of determined values, and traverses along the path that meets the conditions to a leaf node.

[0025] The scene category corresponding to the leaf node that has been traversed is output as the scene recognition result.

[0026] Optionally, calculating the signal-to-noise ratio data based on the ambient light data and the reflected laser signal includes:

[0027] The reflected laser signal is analyzed to obtain a first component and a second component;

[0028] Extract the background light data corresponding to the time from the ambient light data;

[0029] The first component and the background light data are input into the adaptive filtering model, and the effective signal data is calculated by the adaptive filtering model. The second component is used as noise data.

[0030] The effective signal data and the noise data are input into the evaluation network for calculation to obtain the signal-to-noise ratio data.

[0031] Optionally, the step of inputting the ambient light data, target distance information, and signal-to-noise ratio data into a convolutional neural network for feature extraction to obtain multi-dimensional environmental features includes:

[0032] The ambient light data, the target distance information, and the signal-to-noise ratio data are respectively input into the first branch, the second branch, and the third branch of the convolutional neural network for processing;

[0033] In the first branch, a first convolution operation is performed on the ambient light data to extract light intensity distribution features; in the second branch, a second convolution operation is performed on the target distance information to extract distance change features; and in the third branch, a third convolution operation is performed on the signal-to-noise ratio data to extract signal quality features.

[0034] The light intensity distribution characteristics, the distance variation characteristics, and the signal quality characteristics are combined to generate multi-dimensional environmental characteristics.

[0035] Optionally, the step of querying a preset mapping table based on the scene recognition result to obtain the target value of the driving current of the emission module in the laser methane detector, and determining the current adjustment amount based on the target value and the actual value of the driving current, includes:

[0036] Based on the scene recognition result, the corresponding drive current setting value is found in the mapping table, and the drive current setting value is used as the target value.

[0037] The current value of the drive current of the transmitting module is collected as the actual value;

[0038] Calculate the difference between the target value and the actual value, and use the difference as the current adjustment amount.

[0039] Secondly, this application provides an adaptive optical power adjustment system for a laser methane detector, comprising:

[0040] The acquisition module is used to acquire ambient light data, target distance information, and reflected laser signals;

[0041] The calculation module is used to calculate the signal-to-noise ratio data based on the ambient light data and the reflected laser signal;

[0042] The extraction module is used to input the ambient light data, target distance information and signal-to-noise ratio data into a convolutional neural network for feature extraction to obtain multi-dimensional environmental features;

[0043] The recognition module is used to input the multi-dimensional environmental features into the fuzzy logic processor for scene recognition and output the scene recognition result;

[0044] The determination module is used to query a preset mapping table based on the scene recognition result to obtain the target value of the driving current of the emission module in the laser methane detector, and to determine the current adjustment amount based on the target value and the actual value of the driving current.

[0045] An adjustment module is used to adjust the drive current of the emission module based on the current adjustment amount, so as to realize adaptive adjustment of the optical power of the laser methane detector.

[0046] Thirdly, this application provides an electronic device, comprising:

[0047] Memory, used to store computer programs;

[0048] A processor is configured to execute the computer program to implement the steps of the adaptive adjustment method for the optical power of the laser methane detector as described in the first aspect above.

[0049] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, can implement the steps of the adaptive adjustment method for the optical power of a laser methane detector as described in the first aspect above.

[0050] This application provides a method for adaptive adjustment of optical power in a laser methane detector, which has the following advantages:

[0051] First, by collecting ambient light data, target distance information, and reflected laser signals, a comprehensive environmental parameter foundation is provided for subsequent adjustments. Then, based on this data, the signal-to-noise ratio (SNR) is calculated, directly assessing the quality of the current detection signal. Next, the aforementioned multi-source data is input into a convolutional neural network for feature extraction, uncovering deep correlations between ambient light, distance, and SNR. Subsequently, these features are input into a fuzzy logic processor for scene recognition, effectively addressing the uncertainty and ambiguity of environmental parameters, thus enabling accurate judgment of complex operating conditions. Then, based on the recognized scene results, a preset mapping table is consulted to quickly match the optimal target value of the driving current suitable for the current environment. Finally, by comparing the target value with the actual value, the adjustment amount is determined and executed, ultimately achieving precise and adaptive control of the laser's optical power, ensuring the stability of the methane detection signal under different environments.

[0052] Furthermore, this method effectively accommodates the uncertainty of input features through fuzzing, and then achieves robust recognition of complex scenes jointly determined by light intensity, distance and signal quality through rule-based reasoning, providing a reliable decision basis for subsequent precise power adjustment.

[0053] These or other aspects of this application will become more apparent in the following description of the embodiments. Attached Figure Description

[0054] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0055] Figure 1 A flowchart of an adaptive optical power adjustment method for a laser methane detector provided in this application embodiment;

[0056] Figure 2 A schematic diagram illustrating a specific implementation of an adaptive optical power adjustment method for a laser methane detector provided in this application embodiment;

[0057] Figure 3 This is a schematic diagram of the structure of an adaptive optical power adjustment system for a laser methane detector provided in an embodiment of this application. Detailed Implementation

[0058] Laser methane detectors face a core challenge when used in complex environments such as tunnels and pipeline corridors: fixed laser power or simple threshold adjustment methods struggle to simultaneously adapt to the coupled effects of sudden changes in ambient light intensity, significant variations in target distance, and complex background noise. Specifically, in scenarios with close proximity of personnel or strong reflected light, excessively high laser power introduces interference, reducing the signal-to-noise ratio; while at long distances or under low light conditions, insufficient power results in a weak effective signal, ultimately affecting the accuracy and reliability of methane detection. Therefore, existing adjustment methods have limitations in terms of precision and dynamic response capability when dealing with such multidimensional and ambiguous complex conditions.

[0059] In view of this, this application proposes an adaptive optical power adjustment method for a laser methane detector. This method first simultaneously acquires ambient light, target distance, and reflected laser signals, and calculates the real-time signal-to-noise ratio to construct a multi-dimensional data foundation characterizing the current operating conditions. Next, a convolutional neural network is used to extract deep features from these data, uncovering the intrinsic correlation between light, distance, and noise to form a high-quality environmental feature description. Then, these features are input into a fuzzy logic processor, leveraging its advantage in handling uncertain information to accurately identify specific scenarios such as "person approaching strong light" or "low light at a distance." Finally, based on the identified scenario, a preset current mapping table is queried, and closed-loop adjustment is executed to dynamically match the optimal laser driving current. Therefore, this solution, by constructing a complete technical chain of "data acquisition - feature fusion - intelligent recognition - precise execution," transforms the multi-dimensional and complex changes in the environment into precise power control commands, effectively solving the problem of insufficient adaptability of existing technologies to dynamic environmental changes, and achieving the goal of maintaining high stability of the detection signal under various complex operating conditions.

[0060] To enable those skilled in the art to better understand the present application, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely some embodiments of the present application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0061] The core of this application is to provide a method for adaptive adjustment of optical power in a laser methane detector, and a flowchart of one specific implementation is shown below. Figure 1 As shown, the method includes:

[0062] Step 101: Collect ambient light data, target distance information, and reflected laser signals.

[0063] In step 101, ambient light data refers to quantitative information obtained by photoelectric sensors that reflects the overall illumination level of the detection site; target distance information refers to the spatial interval data between the current target object and the laser methane detector measured by a ranging device; and reflected laser signal refers to the light wave signal output by the receiving module in the laser methane detector.

[0064] In this embodiment, ambient light data reflecting the background brightness is first collected by an ambient light sensor. At the same time, target distance information indicating the distance to the target is obtained by an ultrasonic sensor or a laser ranging module. Simultaneously, the reflected laser signal emitted by the laser and reflected back by the target is received by a photodetector, thereby providing the necessary raw data input for subsequent analysis and adjustment.

[0065] Step 102: Calculate the signal-to-noise ratio data based on the ambient light data and the reflected laser signal.

[0066] In this embodiment, step 102 includes the following process:

[0067] Step 1021: Analyze the reflected laser signal to obtain the first component and the second component.

[0068] In step 1021, the first component refers to the DC component of the reflected laser signal that represents stable light intensity, and the second component refers to the AC component of the reflected laser signal that represents random disturbance.

[0069] In this embodiment of the application, the reflected laser signal is input into a signal separation module. The module extracts the second component and the first component through a high-pass filter and a low-pass filter, respectively. The high-pass filter is used to separate the high-frequency random disturbance part to obtain the second component, and the low-pass filter is used to extract the low-frequency stable light intensity part to obtain the first component.

[0070] In practical applications, the signal separation module receives a reflected laser signal with a voltage amplitude of 5.0 volts. After passing through a high-pass filter, it obtains a voltage sequence containing random fluctuations as the second component. At the same time, after passing through a low-pass filter, it obtains a DC signal with a stable voltage value of 4.8 volts as the first component.

[0071] Step 1022: Extract the background light data corresponding to the time from the ambient light data.

[0072] In step 1022, the background light data refers to the sampled value of the ambient light data at the same time point when the reflected laser signal is acquired.

[0073] In this embodiment of the application, based on the acquisition timestamp of the reflected laser signal, a data point that perfectly matches the timestamp is indexed from the continuous sampling sequence of the ambient light data, and then the value recorded by the data point is used as the background light data.

[0074] In practical applications, assuming the reflected laser signal is recorded at 10:30:05 AM, and the value recorded at 10:30:05 AM is retrieved from the ambient light data storage queue, for example, if the value is 0.5 lux, then this 0.5 lux is used as the background light data.

[0075] Step 1023: Input the first component and the background light data into the adaptive filtering model, calculate the effective signal data through the adaptive filtering model, and use the second component as noise data.

[0076] In step 1023, the adaptive filtering model is a signal processing model that automatically adjusts its internal parameters according to the characteristics of the input signal. Its function is to filter out specific interference mixed in the useful signal. The effective signal data refers to the laser intensity information generated only by the reflection of the target object after removing the background light intensity influence represented by the background light data from the first component. The noise data refers to the signal disturbance information represented by the second component.

[0077] It should be noted that the embodiments of this application do not specifically limit the structure, function, implementation process, etc. of the adaptive filtering model.

[0078] In this embodiment, the adaptive filtering model takes the first component as the main input and the background light data as the reference input, and continuously adjusts a filtering weight coefficient to subtract the components related to the reference input from the main input to the maximum extent. The final output of the model is the effective signal data. At the same time, the second component is marked as the noise data.

[0079] In practical applications, the main input of the adaptive filtering model is 4.8 volts, and the reference input is 0.5 lux corresponding to the background light intensity. After model iteration and adjustment, the model outputs a voltage value of 4.3 volts as valid signal data. At the same time, the second component obtained in step 1021 is marked as noise data.

[0080] Step 1024: Input the effective signal data and the noise data into the evaluation network for calculation to obtain the signal-to-noise ratio data.

[0081] In step 1024, the evaluation network is a computational module with mathematical operation capabilities, used to calculate the power ratio of signal to noise; the signal-to-noise ratio data is a quantitative result describing the ratio between the strength of the effective signal data and the strength of the noise data.

[0082] It should be noted that the embodiments of this application do not specifically limit the specific structure, function, implementation process, etc. of the evaluation network.

[0083] In this embodiment, the evaluation network calculates the power of the effective signal data and the power of the noise data, where the power of the effective signal data is the square of its voltage value and the power of the noise data is the mean square value of its voltage sequence. Then, the evaluation network uses the ratio of the power value of the effective signal data to the power value of the noise data as the signal-to-noise ratio data.

[0084] In practical applications, the power of the effective signal data calculated by the evaluation network is 4.3 × 4.3 = 18.49V. 2 Assuming the power of the noise data is 0.25V 2 Therefore, the signal-to-noise ratio is 73.96.

[0085] This application separates the effective portion characterizing methane reflection from a mixed reflected laser signal and filters out ambient background light interference. At the same time, it quantifies and evaluates the noise level in the signal and finally calculates the accurate signal-to-noise ratio, providing a precise basis for subsequent judgment of signal quality and power adjustment.

[0086] Step 103: Input the ambient light data, target distance information and signal-to-noise ratio data into a convolutional neural network for feature extraction to obtain multi-dimensional environmental features.

[0087] In this embodiment, step 103 includes the following process:

[0088] Step 1031: Input the ambient light data, the target distance information, and the signal-to-noise ratio data into the first branch, the second branch, and the third branch of the convolutional neural network for processing.

[0089] In step 1031, the first branch, the second branch, and the third branch are three independent parallel data processing channels in the convolutional neural network. Each branch has its own independent convolutional layer and parameter settings, which are used to process different types of data.

[0090] It should be noted that the number of convolutional layers and parameter settings for each branch can be the same or different. Therefore, the embodiments of this application do not impose specific limitations on the number of convolutional layers and parameter settings for each branch.

[0091] In this embodiment of the application, the collected ambient light data sequence, target distance information sequence, and signal-to-noise ratio data sequence are formatted respectively. Then, the formatted ambient light data is directed to the input interface of the first branch, the target distance information is directed to the input interface of the second branch, and the signal-to-noise ratio data is directed to the input interface of the third branch through the data routing function.

[0092] In practical applications, assuming the ambient light data sequence is [0.5, 0.6, 0.55] lux, the target distance information sequence is [10, 12, 11] meters, and the signal-to-noise ratio data sequence is [70, 73, 69], these sequences are then sent to the corresponding branches for processing.

[0093] Step 1032: In the first branch, a first convolution operation is performed on the ambient light data to extract light intensity distribution features; in the second branch, a second convolution operation is performed on the target distance information to extract distance change features; and in the third branch, a third convolution operation is performed on the signal-to-noise ratio data to extract signal quality features.

[0094] In step 1032, the first convolution operation, the second convolution operation, and the third convolution operation are respectively the mathematical calculation processes of extracting features from the input data using convolution kernels within the three branches; the light intensity distribution feature is the feature data characterizing the change pattern of ambient light intensity in time or space, the distance change feature is the feature data characterizing the fluctuation pattern of target distance, and the signal quality feature is the feature data characterizing the change pattern of signal-to-noise ratio.

[0095] In this embodiment, the process of extracting each feature can be parallel. Specifically: in the first branch, the first convolution operation uses a set of pre-trained convolution kernels to perform a convolution operation on the ambient light data sequence. The specific steps include sliding the convolution kernels on the data sequence and performing dot product summation to extract the light intensity distribution features characterizing the light intensity fluctuation pattern; in the second branch, the second convolution operation uses another set of pre-trained convolution kernels to perform a similar convolution operation on the target distance information sequence to extract the distance change features characterizing the distance change trend; simultaneously, in the third branch, the third convolution operation uses a third set of pre-trained convolution kernels to perform a convolution operation on the signal-to-noise ratio data sequence to extract the signal quality features characterizing the signal quality stability.

[0096] In practical applications, for the first branch, a 3-dimensional convolution kernel [0.25, 0.5, 0.25] is used to convolve the ambient light data sequence [0.5, 0.6, 0.55]. The specific calculation is 0.5 × 0.25 + 0.6 × 0.5 + 0.55 × 0.25 = 0.5625, so one value of the light intensity distribution feature is 0.5625. For the second branch, the same convolution kernel [0.25, 0.5, 0.25] can be used to convolve the target distance information sequence. The operation on the sequence [10, 12, 11] yields 10 × 0.25 + 12 × 0.5 + 11 × 0.25 = 11.25, so one value of the distance change feature is 11.25. Similarly, for the third branch, the convolution kernel [0.25, 0.5, 0.25] can be used to operate on the signal-to-noise ratio data sequence [70, 73, 69], yielding 70 × 0.25 + 73 × 0.5 + 69 × 0.25 = 71.25, so one value of the signal quality feature is 71.25. It should be noted that the above calculation is only an example of a single convolution operation; in practice, multiple convolution kernels can be used to extract multiple feature values ​​to form a feature vector.

[0097] Step 1033: Combine the light intensity distribution features, the distance variation features, and the signal quality features to generate multi-dimensional environmental features.

[0098] In step 1033, the multi-dimensional environmental features are the combination of information from three different dimensions—light intensity distribution features, distance variation features, and signal quality features—into a unified data expression form, which is used to comprehensively describe the overall state of the detection environment.

[0099] In this embodiment, the multiple feature values ​​contained in the light intensity distribution feature, the distance change feature, and the signal quality feature are first flattened and converted into feature vectors. Then, these three feature vectors are concatenated in the feature dimension direction to form a longer and more comprehensive feature vector, which serves as the multi-dimensional environmental feature.

[0100] In practical applications, assuming the light intensity distribution feature corresponds to the vector [0.5625, 0.58], the distance change feature corresponds to the vector [11.25, 11.1], and the signal quality feature corresponds to the vector [71.25, 72.0], then concatenating these three vectors in sequence will generate a multi-dimensional environmental feature vector of [0.5625, 0.58, 11.25, 11.1, 71.25, 72.0].

[0101] This application utilizes convolutional neural networks to perform parallel and deep feature mining on different types of environmental parameters, and integrates all the mined features into a unified comprehensive environmental description, providing a high-quality, multi-dimensional information foundation for accurate scene recognition in the future.

[0102] Step 104: Input the multi-dimensional environmental features into the fuzzy logic processor for scene recognition and output the scene recognition result.

[0103] In this embodiment, step 104 includes the following process, such as... Figure 2 As shown:

[0104] Step 1041: Input the multi-dimensional environmental features into the fuzzification interface of the fuzzy logic processor. Through multiple preset membership functions in the fuzzification interface, convert the light intensity distribution features, distance change features, and signal quality features in the multi-dimensional environmental features into corresponding fuzzy sets.

[0105] In step 1041, the fuzzification interface is a component in the fuzzy logic processor responsible for converting precise numerical inputs into fuzzy language descriptions. The membership function is a function used to calculate the degree to which a precise value belongs to a fuzzy language concept. The fuzzy set is a way of expressing the degree to which a precise value belongs to a certain fuzzy concept using membership values.

[0106] It should be noted that the embodiments of this application do not specifically limit the actual structure of the fuzzy interface, the expression of the membership function, etc.

[0107] In this embodiment, the fuzzification interface receives the precise values ​​of the light intensity distribution feature, distance variation feature, and signal quality feature, respectively, and then inputs these values ​​into their respective predefined membership functions for calculation. Each feature corresponds to multiple fuzzy language concepts. For example, the light intensity distribution feature corresponds to the three concepts of "weak light intensity", "medium light intensity", and "strong light intensity". Each concept has a membership function. The membership degree of the feature value relative to each concept is calculated, and the set of these membership values ​​is the fuzzy set corresponding to the feature.

[0108] In practical applications, assuming the light intensity distribution characteristic value is 0.5625, the membership function of the corresponding concept "weak light intensity" outputs a membership degree of 0.1 when the input is 0.5625, the membership function of the concept "medium light intensity" outputs 0.7, and the membership function of the concept "strong light intensity" outputs 0.2. Therefore, the fuzzy set corresponding to the light intensity distribution characteristic can be represented as follows: the membership degree for "weak light intensity" is 0.1, for "medium light intensity" it is 0.7, and for "strong light intensity" it is 0.2. Similarly, the distance change characteristic value is 11.25 meters, and the membership functions of the corresponding concepts "near distance," "medium distance," and "far distance" may output 0.8, 0.2, and 0.0, respectively. The signal quality characteristic value is 71.25, and the membership functions of the corresponding concepts "poor quality," "medium quality," and "good quality" may output 0.1, 0.4, and 0.5, respectively.

[0109] Step 1042: Input all fuzzy sets into the rule base of the fuzzy logic processor. The rule base contains multiple preset inference rules. Perform inference operations on all fuzzy sets according to the inference rules to obtain fuzzy output results. Output scene recognition results based on the fuzzy output results.

[0110] The rule base stores multiple "if-then" reasoning rules based on expert experience, used to describe the fuzzy logical relationship between input features and output scenarios; the fuzzy output result refers to the membership description of different output scenario categories obtained after reasoning operations.

[0111] Step 1042 may specifically include the following steps:

[0112] A1: Calculate the confidence level of the corresponding conclusion part based on the degree to which the fuzzy set satisfies the corresponding condition part in the inference rule.

[0113] In step A1, confidence level refers to the degree to which a certain inference rule is activated under the current input, that is, the credibility of the conclusion derived by the rule.

[0114] In this embodiment of the application, for each reasoning rule in the rule base, the condition part of the rule is usually composed of multiple fuzzy propositions connected by logical operators such as "AND" and "OR". First, the membership value of the corresponding fuzzy proposition is found from the input fuzzy set. Then, the degree of satisfaction of the overall condition part of the rule is calculated according to the logical operators. The "AND" operation usually takes the minimum value, and the "OR" operation usually takes the maximum value. This degree of satisfaction is the confidence level of the conclusion part of the rule.

[0115] In practical applications, suppose there is an inference rule: "If the light intensity is strong and the distance is close, then it is the first scenario." The membership degree of "strong light intensity" is 0.2, and the membership degree of "close distance" is 0.8. By taking the minimum value through the "AND" operation, we get the condition satisfaction level of the rule as 0.2. This 0.2 is the confidence level of the rule's conclusion "first scenario".

[0116] A2: Combine the confidence scores of the conclusion parts of all fuzzy sets to generate fuzzy output results.

[0117] In step A2, combination refers to aggregating the confidence scores of all rules pointing to the same output scene category to obtain the overall membership score of that scene category.

[0118] In this embodiment of the application, all inference rules are grouped according to the output scene category pointed to by their conclusions. For the same output scene category, such as "first scene", the confidence scores of all rules that conclude that scene are ORed (usually the maximum value is taken). The maximum value is the membership degree of the fuzzy concept "first scene" in the output fuzzy set. Similarly, the membership degrees of other scene categories are calculated. These membership degrees together constitute the fuzzy output result.

[0119] In practical applications, assuming there are three rules pointing to "the first scenario" with confidence levels of 0.2, 0.1, and 0.3 respectively, the maximum value obtained by using an "OR" operation is 0.3 for "the first scenario". There are two rules pointing to "the second scenario" with confidence levels of 0.6 and 0.4 respectively, so the membership level of "the second scenario" is 0.6. Therefore, the fuzzy output is that the membership level of "the first scenario" is 0.3 and the membership level of "the second scenario" is 0.6.

[0120] A3: Based on the fuzzy output results, the improved centroid method is used to calculate multiple definite values ​​corresponding to multiple scene categories, forming a sequence of definite values.

[0121] In step A3, the determined value is the quantized output after clarifying a certain scene category, and the determined value sequence is a list composed of these determined values ​​arranged in order.

[0122] In this embodiment of the application, each scene category in the fuzzy output result is regarded as a fuzzy set, and the membership degree of the scene category is used as the weight of its fuzzy set. The calculation of the improved centroid method involves the typical representative value of the scene category. For example, a typical value of 1 can be preset for "first scene" and a typical value of 2 can be preset for "second scene". Then, the calculation is performed on each scene category respectively, and the calculation results are arranged in the scene order to form a definite value sequence.

[0123] In practical applications, for fuzzy output results, the membership degree of "first scene" is 0.3, and the typical value is preset to 1; the membership degree of "second scene" is 0.6, and the typical value is preset to 2; then the first definite value is calculated as 0.3×1=0.3, and the second definite value is calculated as 0.6×2=1.2; the final definite value sequence is [0.3, 1.2].

[0124] A4: Input the determined value sequence into a gated recurrent unit network for temporal feature enhancement to generate an enhanced determined value sequence.

[0125] In step A4, the gated recurrent unit network is a neural network model that can process sequential data and capture its temporal dependencies; the enhanced deterministic value sequence is a new deterministic value sequence that contains historical or contextual temporal information after being processed by this network.

[0126] In this embodiment, the current time-determined value sequence is first combined with the stored past time-determined value sequences in chronological order to form a time-series sequence. Then, this time-series sequence is input into a pre-trained gated recurrent unit network. The network selectively memorizes and forgets the sequence information through its internal gating mechanism and outputs a new sequence corresponding to the length of the input sequence. The last output of the new sequence is taken as the enhanced determined value sequence for the current time.

[0127] In practical applications, assuming the current time-determined value sequence is [0.3, 1.2], and the sequences of the past two time points are [0.25, 1.0] and [0.28, 1.1], they can be combined into a time series sequence {[0.25, 1.0], [0.28, 1.1], [0.3, 1.2]}. This sequence is then input into a gated recurrent unit network. After processing, the network may output a new sequence, which is the enhanced determined value sequence [0.31, 1.25].

[0128] A5: Based on the enhanced sequence of determined values, a decision tree model is used for classification to generate scene recognition results.

[0129] Step A5 may specifically include the following steps:

[0130] B1: Input the enhanced sequence of determined values ​​into the decision tree model. The decision tree model contains a root node, multiple branch nodes and multiple leaf nodes. Each branch node corresponds to an attribute judgment condition, and each leaf node corresponds to a scene category.

[0131] In this embodiment of the application, after receiving the enhanced sequence of determined values, the decision tree model starts making judgments from the root node.

[0132] In practical applications, the enhanced sequence of determined values ​​is [0.31, 1.25], and the root node judgment condition of the decision tree model is "whether the first determined value is less than 0.5".

[0133] B2: The decision tree model starts from the root node, and sequentially determines whether the attribute judgment conditions of the branch node are met according to the enhanced sequence of determined values, and traverses along the path that meets the conditions to a leaf node.

[0134] In this embodiment, the model starts from the root node based on the values ​​in the enhanced sequence of determined values. At each branch node, it determines whether the corresponding attribute condition is true. If the condition is true, it moves to a child node; if it is false, it moves to another child node. This process continues until a leaf node with no more branches is reached.

[0135] In practical applications, the root node determines whether the "first definite value is less than 0.5". Since 0.31 is less than 0.5, the condition is met, so it moves to the left child node. The condition for the left child node may be whether the "second definite value is greater than 1.2". Since 1.25 is greater than 1.2, the condition is met, so it continues to move to its corresponding child node, eventually reaching a leaf node marked as "first scenario".

[0136] B3: Output the scene category corresponding to the leaf node that has been traversed as the scene recognition result.

[0137] In this embodiment, the model outputs the preset scene category label stored in the last leaf node as the final judgment result.

[0138] In practical applications, the leaf node that is finally reached is pre-labeled as "first scene", so the decision tree model outputs "first scene" as the scene recognition result.

[0139] This application first uses fuzzy logic to handle the uncertainty and fuzziness of environmental features, then enhances the continuity of decision-making through temporal networks, and finally uses decision trees to make explicit classification decisions, thereby improving the robustness and accuracy of scene recognition under complex and ever-changing environmental conditions.

[0140] Step 105: Based on the scene recognition results, query the preset mapping table to obtain the target value of the driving current of the emission module in the laser methane detector, and determine the current adjustment amount based on the target value and the actual value of the driving current.

[0141] The mapping table is a pre-generated data table stored in memory, which records the fixed correspondence between different scene recognition results and suggested drive current values.

[0142] In this embodiment, step 105 includes the following process:

[0143] Step 1051: Based on the scene recognition result, find the corresponding drive current setting value in the mapping table and use the drive current setting value as the target value.

[0144] In step 1051, the drive current setting value is a current value pre-configured in the mapping table for the specific scene recognition result; the target value is determined based on the current scene, and it is expected that the drive current will eventually be adjusted to an ideal value.

[0145] In this embodiment of the application, the scene recognition result output by step 104 is first read, for example, the result is "first scene". Then, the data row with the key column "first scene" is searched in the mapping table, and the corresponding stored driving current value is read from the data row. Finally, the read value is designated as the target value to be achieved in this adjustment process.

[0146] In practical applications, assuming the scene recognition result is "Scene 1", and the preset driving current setting value for "Scene 1" in the mapping table is 300 mA, then 300 mA will be used as the target value for this adjustment.

[0147] Step 1052: Collect the current value of the drive current of the transmitting module as the actual value.

[0148] In step 1052, the actual value refers to the current value that is flowing through the laser drive circuit in the transmitting module, which is obtained in real time through the current measurement circuit.

[0149] In this embodiment, a voltage signal is obtained by using a sampling resistor connected in series in the laser driving circuit. Then, the voltage signal is converted into a digital quantity by an analog-to-digital converter. Finally, the magnitude of the current flowing through the sampling resistor is calculated based on the resistance value and the voltage-current ratio, and this is taken as the actual value.

[0150] In practical applications, the resistance of the sampling resistor is 0.1 ohms. The analog-to-digital converter measures a voltage of 35 millivolts across the sampling resistor. According to Ohm's law, the actual current value is equal to the voltage divided by the resistance. Therefore, the current actual value is 350 milliamperes.

[0151] Step 1053: Calculate the difference between the target value and the actual value, and use the difference as the current adjustment amount.

[0152] In step 1053, the current adjustment amount is a signed numerical value used to represent the magnitude and direction of the change required to adjust the actual current value to the target current value.

[0153] In this embodiment of the application, the target value obtained in step 1051 is first used as the minuend, and the actual value collected in step 1052 is used as the subtrahend, and a subtraction operation is performed; wherein the absolute value of the calculation result represents the magnitude of the current that needs to be adjusted, and its positive or negative sign indicates the direction of adjustment, with a positive sign indicating that the current needs to be increased and a negative sign indicating that the current needs to be decreased.

[0154] In practical applications, if the target value is 300 milliamperes and the actual value is 350 milliamperes, then the current adjustment is equal to 300 milliamperes minus 350 milliamperes, and the calculation result is -50 milliamperes. This -50 milliamperes means that in order to achieve the target, the drive current needs to be reduced by 50 milliamperes.

[0155] This application accurately maps the scene recognition result to a specific current target, and calculates the required adjustment amount by comparing the current target with the current actual current, thereby providing clear and quantitative control instructions for the final execution of precise optical power adjustment.

[0156] Step 106: Adjust the drive current of the emission module based on the current adjustment amount to achieve adaptive adjustment of the optical power of the laser methane detector.

[0157] In step 106, adjusting the driving current of the emission module refers to the process of adjusting the current flowing through the laser by changing the output of the driving circuit according to the current adjustment amount; adaptive optical power adjustment refers to the ability of the laser output optical power to be automatically adjusted according to changes in the detection environment, thereby maintaining the final effect of a stable methane detection signal.

[0158] In this embodiment of the application, based on the current adjustment amount determined in step 105, the control unit sends a corresponding pulse width modulation signal or voltage command to the current drive circuit. The command causes the output current of the drive circuit to change according to the magnitude and direction of the adjustment amount, thereby adjusting the injection current of the laser. The optical power emitted by the laser is proportional to its drive current, so changing the drive current achieves adaptive adjustment of the optical power of the laser methane detector.

[0159] Figure 3 A schematic diagram of an adaptive optical power adjustment system for a laser methane detector provided in this application embodiment is shown below. Figure 3 As shown, the system includes:

[0160] The acquisition module 31 is used to acquire ambient light data, target distance information, and reflected laser signals.

[0161] The calculation module 32 is used to calculate the signal-to-noise ratio data based on the ambient light data and the reflected laser signal.

[0162] The extraction module 33 is used to input the ambient light data, target distance information and signal-to-noise ratio data into a convolutional neural network for feature extraction to obtain multi-dimensional environmental features.

[0163] The recognition module 34 is used to input the multi-dimensional environmental features into the fuzzy logic processor for scene recognition and output the scene recognition result.

[0164] The determination module 35 is used to query a preset mapping table based on the scene recognition result to obtain the target value of the driving current of the emission module in the laser methane detector, and to determine the current adjustment amount based on the target value and the actual value of the driving current.

[0165] The adjustment module 36 is used to adjust the drive current of the emission module based on the current adjustment amount, so as to realize the adaptive adjustment of the optical power of the laser methane detector.

[0166] The adaptive optical power adjustment system for the laser methane detector in this application is used to implement the aforementioned adaptive optical power adjustment method for the laser methane detector. Therefore, the specific implementation of the adaptive optical power adjustment system for the laser methane detector can be found in the embodiment section of the adaptive optical power adjustment method for the laser methane detector mentioned above. The specific implementation can be referred to the description of the corresponding embodiments, and will not be repeated here.

[0167] This application also provides an electronic device, comprising: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of the adaptive optical power adjustment method for the laser methane detector described above.

[0168] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the adaptive optical power adjustment method for the laser methane detector described above.

[0169] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as USB flash drives, read-only memory, random access memory, portable hard drives, magnetic disks, or optical disks.

[0170] Embodiments of this application also provide a computer program product, which includes a computer program that, when executed by a processor, implements the steps in any of the embodiments of the adaptive adjustment method for optical power of a laser methane detector.

[0171] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0172] The above provides a detailed description of the adaptive optical power adjustment method and system for a laser methane detector provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and its core ideas. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. A method for adaptive adjustment of optical power in a laser methane detector, characterized in that, include: Collect ambient light data, target distance information, and reflected laser signals; Based on the ambient light data and the reflected laser signal, the signal-to-noise ratio data is calculated; The ambient light data, target distance information, and signal-to-noise ratio data are input into a convolutional neural network for feature extraction to obtain multi-dimensional environmental features. The multi-dimensional environmental features are input into a fuzzy logic processor for scene recognition, and the scene recognition result is output. Based on the scene recognition results, a preset mapping table is queried to obtain the target value of the driving current of the emission module in the laser methane detector, and the current adjustment amount is determined based on the target value and the actual value of the driving current. The driving current of the emission module is adjusted based on the current adjustment amount to achieve adaptive adjustment of the optical power of the laser methane detector; The step of inputting the multi-dimensional environmental features into a fuzzy logic processor for scene recognition and outputting the scene recognition result includes: The multi-dimensional environmental features are input to the fuzzification interface of the fuzzy logic processor. Through multiple preset membership functions in the fuzzification interface, the light intensity distribution features, distance change features, and signal quality features in the multi-dimensional environmental features are converted into corresponding fuzzy sets. All fuzzy sets are input into the rule base of the fuzzy logic processor. The rule base contains multiple preset inference rules. Inference operations are performed on all fuzzy sets according to the inference rules to obtain fuzzy output results. Based on the fuzzy output results, scene recognition results are output. The step of performing inference operations on all fuzzy sets according to the inference rules to obtain fuzzy output results, and outputting scene recognition results based on the fuzzy output results, includes: The confidence level of the corresponding conclusion part is calculated based on the degree to which the fuzzy set satisfies the corresponding condition part in the inference rule. The confidence scores of the conclusion parts of all fuzzy sets are combined to generate fuzzy output results; Based on the fuzzy output results, the improved centroid method is used to calculate multiple definite values ​​corresponding to multiple scene categories, forming a sequence of definite values. The determined value sequence is input into a gated recurrent unit network for temporal feature enhancement to generate an enhanced determined value sequence. Based on the enhanced sequence of determined values, a decision tree model is used for classification to generate scene recognition results. The process of classifying the enhanced determined value sequence using a decision tree model to generate scene recognition results includes: The enhanced sequence of determined values ​​is input into a decision tree model, which contains a root node, multiple branch nodes and multiple leaf nodes. Each branch node corresponds to an attribute judgment condition, and each leaf node corresponds to a scene category. The decision tree model starts from the root node, and sequentially determines whether the attribute judgment conditions of the branch node are met according to the enhanced sequence of determined values, and traverses along the path that meets the conditions to a leaf node. The scene category corresponding to the leaf node that has been traversed is output as the scene recognition result.

2. The method according to claim 1, characterized in that, The calculation of signal-to-noise ratio data based on the ambient light data and the reflected laser signal includes: The reflected laser signal is analyzed to obtain a first component and a second component; Extract the background light data corresponding to the time from the ambient light data; The first component and the background light data are input into the adaptive filtering model, and the effective signal data is calculated by the adaptive filtering model. The second component is used as noise data. The effective signal data and the noise data are input into the evaluation network for calculation to obtain the signal-to-noise ratio data.

3. The method according to claim 1, characterized in that, The ambient light data, target distance information, and signal-to-noise ratio data are input into a convolutional neural network for feature extraction to obtain multi-dimensional environmental features, including: The ambient light data, the target distance information, and the signal-to-noise ratio data are respectively input into the first branch, the second branch, and the third branch of the convolutional neural network for processing; In the first branch, a first convolution operation is performed on the ambient light data to extract light intensity distribution features; in the second branch, a second convolution operation is performed on the target distance information to extract distance change features; and in the third branch, a third convolution operation is performed on the signal-to-noise ratio data to extract signal quality features. The light intensity distribution characteristics, the distance variation characteristics, and the signal quality characteristics are combined to generate multi-dimensional environmental characteristics.

4. The method according to claim 1, characterized in that, The step of querying a preset mapping table based on the scene recognition result to obtain the target value of the driving current of the emission module in the laser methane detector, and determining the current adjustment amount based on the target value and the actual value of the driving current, includes: Based on the scene recognition result, the corresponding drive current setting value is found in the mapping table, and the drive current setting value is used as the target value. The current value of the drive current of the transmitting module is collected as the actual value; Calculate the difference between the target value and the actual value, and use the difference as the current adjustment amount.

5. An adaptive optical power adjustment system for a laser methane detector, characterized in that, include: The acquisition module is used to acquire ambient light data, target distance information, and reflected laser signals; The calculation module is used to calculate the signal-to-noise ratio data based on the ambient light data and the reflected laser signal; The extraction module is used to input the ambient light data, target distance information and signal-to-noise ratio data into a convolutional neural network for feature extraction to obtain multi-dimensional environmental features; The recognition module is used to input the multi-dimensional environmental features into the fuzzy logic processor for scene recognition and output the scene recognition result; The determination module is used to query a preset mapping table based on the scene recognition result to obtain the target value of the driving current of the emission module in the laser methane detector, and to determine the current adjustment amount based on the target value and the actual value of the driving current. An adjustment module is used to adjust the drive current of the emission module based on the current adjustment amount, so as to realize adaptive adjustment of the optical power of the laser methane detector; The step of inputting the multi-dimensional environmental features into a fuzzy logic processor for scene recognition and outputting the scene recognition result includes: The multi-dimensional environmental features are input to the fuzzification interface of the fuzzy logic processor. Through multiple preset membership functions in the fuzzification interface, the light intensity distribution features, distance change features, and signal quality features in the multi-dimensional environmental features are converted into corresponding fuzzy sets. All fuzzy sets are input into the rule base of the fuzzy logic processor. The rule base contains multiple preset inference rules. Inference operations are performed on all fuzzy sets according to the inference rules to obtain fuzzy output results. Based on the fuzzy output results, scene recognition results are output. The step of performing inference operations on all fuzzy sets according to the inference rules to obtain fuzzy output results, and outputting scene recognition results based on the fuzzy output results, includes: The confidence level of the corresponding conclusion part is calculated based on the degree to which the fuzzy set satisfies the corresponding condition part in the inference rule. The confidence scores of the conclusion parts of all fuzzy sets are combined to generate fuzzy output results; Based on the fuzzy output results, the improved centroid method is used to calculate multiple definite values ​​corresponding to multiple scene categories, forming a sequence of definite values. The determined value sequence is input into a gated recurrent unit network for temporal feature enhancement to generate an enhanced determined value sequence. Based on the enhanced sequence of determined values, a decision tree model is used for classification to generate scene recognition results. The process of classifying the enhanced determined value sequence using a decision tree model to generate scene recognition results includes: The enhanced sequence of determined values ​​is input into a decision tree model, which contains a root node, multiple branch nodes and multiple leaf nodes. Each branch node corresponds to an attribute judgment condition, and each leaf node corresponds to a scene category. The decision tree model starts from the root node, and sequentially determines whether the attribute judgment conditions of the branch node are met according to the enhanced sequence of determined values, and traverses along the path that meets the conditions to a leaf node. The scene category corresponding to the leaf node that has been traversed is output as the scene recognition result.

6. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the steps of the adaptive optical power adjustment method for a laser methane detector as described in any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, enables the adaptive adjustment method of optical power for a laser methane detector as described in any one of claims 1 to 4.

Images