A Smart Noise Adjustment Method for Range Hoods

Abstract

This application relates to the field of kitchen appliances and discloses an intelligent noise adjustment method for a range hood. The method includes: acquiring an environmental image output by a camera module and gas concentration data output by a gas sensor; converting the environmental image into an environmental feature matrix and extracting the contrast attenuation coefficient corresponding to a reference area in the environmental feature matrix; calculating a first confidence weight based on the contrast attenuation coefficient; extracting the slope of the gas concentration data and calculating a second confidence weight based on the slope; performing feature concatenation between the time-varying edge gradient in the environmental feature matrix and the instantaneous concentration value of the gas concentration data based on the first confidence weight and the second confidence weight to generate a cooking scene feature vector; matching a target noise reduction speed threshold corresponding to the cooking scene feature vector; and generating a fan drive duty cycle corresponding to the target noise reduction speed threshold. This technical solution avoids the lag in speed adjustment caused by unreliable image data.

Description

Technical Field

[0001] This application relates to the field of kitchen appliances, and in particular to an intelligent method for adjusting the noise of a range hood. Background Technology

[0002] As a home appliance that improves air quality in the kitchen environment, the main function of a range hood is to use a fan to generate negative pressure to absorb and expel the fumes produced during cooking; traditional range hoods are mostly manually controlled by setting different speeds.

[0003] In related technologies, some range hoods incorporate camera modules that capture images of the cookware area, run neural network models to identify cooking behaviors, and then match the corresponding fan speed. After acquiring the images, the camera module inputs the image stream to a cloud or local advanced processor, extracts the image features of the oil fume concentration, and then feeds back a speed command to the fan.

[0004] However, the relevant technology has the following shortcomings: In the high-temperature and high-oil environment of a kitchen, volatile grease will inevitably adhere quickly to the lens surface of the camera module and form an oil film. This oil film will cause severe refraction and optical scattering of external light, resulting in a sharp decrease in the contrast of the image captured by the camera module and the loss of edge features. In order to process this contaminated image data to ensure recognition accuracy, the system must rely on a huge image restoration model or a complex anti-interference algorithm, which far exceeds the computing power of the microcontroller built into the range hood. If processing is carried out in the cloud, it will also lead to high data processing latency, causing the fan speed adjustment to lag far behind the real-time cooking status. Summary of the Invention

[0005] This application provides an intelligent noise adjustment method for a range hood, which at least partially solves the above-mentioned technical problems.

[0006] To achieve the above objectives, this application provides an intelligent noise control method for a range hood, comprising: Acquire environmental images output by the camera module and gas concentration data output by the gas sensor; The environmental image is converted into an environmental feature matrix, and the contrast attenuation coefficient corresponding to the reference area is extracted from the environmental feature matrix. The first confidence weight is calculated based on the contrast attenuation coefficient; Extract the slope of the gas concentration data and calculate a second confidence weight based on the slope; Based on the first confidence weight and the second confidence weight, the time-varying edge gradient in the environmental feature matrix and the instantaneous concentration value of the gas concentration data are concatenated to generate a cooking scene feature vector; Match the target noise reduction speed threshold corresponding to the feature vector of the cooking scene; Generate the fan drive duty cycle corresponding to the target noise reduction speed threshold and send the fan drive duty cycle to the fan actuator.

[0007] In this embodiment of the application, through the above technical solution, the image data of the camera module and the concentration data of the gas sensor are weighted and fused by confidence weight. When the image data is unreliable due to lens oil film, the visual weight is reduced and the gas weight is increased. This allows the cooking scene recognition to maintain effective operation based on gas data even when the camera module is contaminated. This avoids the lag in speed adjustment caused by unreliable image data and also reduces the amount of data calculation.

[0008] Other features and advantages of this application will be described in detail in the following detailed description section. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This application provides an exemplary system framework for an intelligent noise control method for a range hood, as described in an exemplary embodiment.

[0011] Figure 2 This is a flowchart illustrating the steps of an intelligent noise control method for a range hood provided in an exemplary embodiment of this application. Detailed Implementation

[0012] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the protection scope of this application.

[0013] Reference Figure 1 This is the system framework for the implementation embodiment of the method in this application. The range hood has a built-in camera module and a gas sensor; the camera module is deployed at the bottom of the smoke collection chamber, and the gas sensor is deployed at the front of the fan inlet; the camera module is connected to the main control board microcontroller, and the gas sensor is connected to the main control board microcontroller through an analog-to-digital converter channel; the main control board microcontroller is electrically connected to the motor drive module of the fan actuator through a pulse width modulation output pin.

[0014] This application provides an intelligent noise control method for range hoods. Please refer to [link / reference]. Figure 2 The intelligent noise adjustment method for a range hood provided in this application includes the following steps: Step 101: Acquire the environmental image output by the camera module and the gas concentration data output by the gas sensor. Specifically, the environmental image of the cookware area is acquired by the camera module located at the bottom of the smoke collection chamber of the range hood, and the gas concentration data characterizing the content of volatile organic compounds is acquired by the gas sensor deployed at the front of the air inlet of the fan. The environmental image here refers to the optical video frames or image data acquired by the camera module; the gas concentration data refers to the digital signal queue output by the gas sensor on a continuous time axis.

[0015] Step 102: Convert the environmental image into an environmental feature matrix and extract the contrast attenuation coefficient corresponding to the reference area from the environmental feature matrix. Specifically, based on the grayscale algorithm, the environmental image is converted into a two-dimensional numerical array as the environmental feature matrix. A pixel sub-matrix corresponding to the reference area is extracted from this matrix. After applying a high-pass filter operator to filter out the slowly varying illumination components, the absolute values ​​of the grayscale differences between adjacent pixels are calculated and summed to obtain the current sharpness characterization value. This value is compared with the initial edge characterization benchmark value under uncontaminated conditions to obtain the contrast attenuation coefficient. The reference area referred to here is a region in the image that is in a relatively static state and has characteristic reference value. The contrast attenuation coefficient reflects the degree of influence of the oil film on the lens surface of the camera module on the image sharpness.

[0016] Step 103: Calculate the first confidence weight based on the contrast attenuation coefficient. Specifically, a preset weight multiplier is matched according to the numerical range of the contrast attenuation coefficient as the first confidence weight; when the contrast attenuation coefficient is less than the first attenuation threshold, the first confidence weight is set to the full-load weight benchmark value; when the coefficient is between the first attenuation threshold and the second attenuation threshold, the coefficient is input into a non-linear smooth attenuation function to calculate the decreasing first confidence weight; when the coefficient is greater than or equal to the second attenuation threshold, the first confidence weight is forcibly set to zero; the aforementioned first confidence weight refers to a numerical multiplier used to characterize the degree of influence of image data in the fusion calculation.

[0017] Step 104: Extract the slope of the gas concentration data and calculate the second confidence weight based on the slope. Specifically, a sliding time window is applied to the gas concentration data to extract the concentration difference within the time window. The concentration difference is divided by the time span of the sliding time window to obtain the instantaneous derivative. The instantaneous derivatives output from multiple consecutive sliding time windows are smoothed and filtered to obtain the slope. The second confidence weight is calculated based on this slope. The aforementioned second confidence weight refers to a numerical multiplier used to characterize the degree of influence of gas sensing data in the fusion calculation.

[0018] Step 105: Based on the first and second confidence weights, feature concatenation is performed on the time-varying edge gradient in the environmental feature matrix and the instantaneous concentration value of the gas concentration data to generate a cooking scene feature vector. Specifically, spatial dimensionality reduction is performed on the time-varying edge gradient in the environmental feature matrix to obtain scalar edge feature values. The edge feature values ​​are multiplied by the first confidence weight to obtain visual weight features, and the instantaneous concentration value is multiplied by the second confidence weight to obtain gas weight features. After normalization processing, the visual weight features and gas weight features are concatenated to generate a one-dimensional cooking scene feature vector. The time-varying edge gradient mentioned here refers to the feature component extracted from the environmental feature matrix that reflects the change of the oil fume contour over time during the cooking process.

[0019] Step 106: Match the target noise reduction speed threshold corresponding to the cooking scene feature vector. Specifically, calculate the distance between the cooking scene feature vector and multiple pre-stored standard scene vectors, and select the speed parameter associated with the standard scene vector with the smallest distance as the target noise reduction speed threshold.

[0020] Step 107: Generate the fan drive duty cycle corresponding to the target noise reduction speed threshold and send the fan drive duty cycle to the fan actuator. Specifically, calculate the speed difference between the target noise reduction speed threshold and the current reference operating speed of the fan actuator, input the speed difference into a preset smooth transition function to calculate the transition speed node, convert the transition speed node into the corresponding fan drive duty cycle step by step, and send the duty cycle to the motor drive module of the fan actuator through the pulse width modulation output pin.

[0021] In one implementation, the generation of the cooking scene feature vector can replace the splicing operation by weighted summation of visual weight features and gas weight features, directly outputting the scene discrimination value in scalar form.

[0022] In another implementation, the target noise reduction speed threshold can be matched by using decision tree classification instead of distance calculation, and the speed gear to be entered can be determined step by step based on the value of each dimension of the feature vector.

[0023] The above technical solution integrates image data from the camera module with concentration data from the gas sensor using confidence weights. When the image data becomes unreliable due to lens oil film, the visual weight is reduced and the gas weight is increased. This allows the cooking scene recognition to maintain effective operation based on gas data even when the camera module is contaminated. It avoids the lag in speed adjustment caused by unreliable image data and also reduces the amount of data computation.

[0024] In some embodiments, converting the environmental image into an environmental feature matrix and extracting the contrast attenuation coefficient corresponding to the reference area from the environmental feature matrix includes: Step 201: Based on the preset coordinate mapping relationship, obtain the pixel sub-matrix corresponding to the reference area from the environmental feature matrix. Specifically, the coordinate mapping relationship is pre-calibrated and stored in the parameter area of ​​the microcontroller. This mapping relationship records the row and column start and end indices of the reference area in the entire environmental image. The corresponding pixel sub-matrix is ​​cropped from the environmental feature matrix according to this index. The reference area mentioned here refers to the area in the image that is in a relatively static state and has feature reference value. It is usually selected as the stove surface area outside the edge of the pot.

[0025] Step 202: Apply a preset high-pass filter operator to the pixel submatrix to filter out slowly varying illumination components and generate an edge matrix. Specifically, use the Sobel operator or Laplacian operator to perform convolution filtering on the pixel submatrix to extract the rapidly changing components of pixel grayscale as the edge matrix, while filtering out the slowly varying DC components generated by uniform ambient light illumination.

[0026] Step 203: Calculate the absolute value of the grayscale difference between adjacent pixels in the edge matrix. Specifically, traverse each pixel position in the edge matrix, calculate the grayscale difference between that pixel and its right and lower neighbor pixels, and take the absolute value. Organize all the absolute values ​​into a difference matrix.

[0027] Step 204: Sum the absolute values ​​to obtain the absolute error sum, which is used as the current sharpness representation value. Specifically, perform a summation operation on all elements in the difference matrix, and the resulting scalar value is the current sharpness representation value; the larger the value, the richer the edge details of the reference area, that is, the higher the image sharpness; the smaller the value, the more severely the edge information is obscured by the oil film.

[0028] Step 205: Obtain the initial edge characterization reference value of the reference area under a pollution-free state. Specifically, the initial edge characterization reference value is pre-collected and stored during the factory calibration stage of the range hood, that is, the absolute error and value obtained by performing the same calculation process as steps 201 to 204 on the reference area under a clean lens state.

[0029] Step 206: Calculate the difference between the initial edge representation reference value and the current sharpness representation value, and obtain the contrast attenuation coefficient based on the difference and the initial edge representation reference value. Specifically, divide the difference by the initial edge representation reference value to obtain the normalized contrast attenuation coefficient; to avoid excessive computational power consumption in the division operation, a bitwise right shift instruction can be used to divide the difference by the power of two corresponding to the reference value to achieve approximate normalization, thereby improving the real-time response speed of the control.

[0030] In one implementation, the high-pass filter operator can use the Canny operator instead of the Sobel operator to suppress isolated noise points while maintaining edge extraction performance.

[0031] In another implementation, the sharpness representation value can be calculated using a variance metric instead of the absolute error sum, reflecting the richness of image detail by statistically analyzing the gray-level variance of the pixel submatrix.

[0032] The above technical solution allows the system to obtain the effect of lens oil film on image quality through the contrast attenuation coefficient of the reference area, enabling the system to adjust the reliability of image data in real time according to the degree of lens contamination.

[0033] In some embodiments, calculating a first confidence weight based on a contrast attenuation coefficient includes: Step 301: Obtain the preset first attenuation threshold and second attenuation threshold, wherein the second attenuation threshold is greater than the first attenuation threshold. Specifically, the first attenuation threshold and the second attenuation threshold are determined based on the lens oil film accumulation experimental data during the factory calibration stage and preset in the parameter area of ​​the microcontroller; the first attenuation threshold corresponds to the critical value of slight lens contamination, and the second attenuation threshold corresponds to the critical value of severe lens contamination that renders the image data unusable.

[0034] Step 302: When the contrast attenuation coefficient is less than the first attenuation threshold, the first confidence weight is set to the full-load weight benchmark value. Specifically, when the contrast attenuation coefficient is lower than the first attenuation threshold, it indicates that the lens contamination is slight and the image data is still reliable. At this time, the first confidence weight is taken as the full-load weight benchmark value, that is, the image data occupies the largest influence proportion in the fusion calculation.

[0035] Step 303: When the contrast attenuation coefficient is between the first attenuation threshold and the second attenuation threshold, the contrast attenuation coefficient is input into a preset nonlinear smooth attenuation function to calculate the decreasing first confidence weight. Specifically, the nonlinear smooth attenuation function can be a Sigmoid function or a piecewise linear interpolation function to ensure that the weight decreases smoothly between the two thresholds and avoids control output jumps caused by sudden weight changes; as the contrast attenuation coefficient increases, the first confidence weight gradually decreases from the full-load weight benchmark value to close to zero.

[0036] Step 304: When the contrast attenuation coefficient is greater than or equal to the second attenuation threshold, the first confidence weight is forcibly set to zero. Specifically, when the contrast attenuation coefficient reaches or exceeds the second attenuation threshold, it indicates that the lens oil film has severely blocked the image data, making it completely unusable. At this time, the first confidence weight is forcibly set to zero to completely shield the noise interference of the image channel.

[0037] In one implementation, the nonlinear smooth decay function can be an exponential decay function, and the decay rate is controlled by a preset decay constant.

[0038] In another implementation, the first attenuation threshold and the second attenuation threshold can be updated based on the calibration results after lens surface cleaning and maintenance to adapt to the differences in oil film accumulation rate under different usage environments.

[0039] The above technical solution uses a segmented confidence weight calculation strategy to maintain high weight of image data when the lens is slightly contaminated, completely block the image channel when it is severely contaminated, and smoothly attenuate in the intermediate transition range. This avoids the interference of visual noise on scene recognition and prevents the fan speed jump caused by sudden weight changes.

[0040] In some embodiments, extracting the slope of the gas concentration data and calculating a second confidence weight based on the slope includes: Step 401: Apply a sliding time window to the gas concentration data to extract the concentration difference within the time window. Specifically, a fixed-length sliding array structure is constructed in the microcontroller's memory. In each data push cycle, the difference between the head element and the tail element of the array is calculated as the concentration difference. The sliding time window mentioned here refers to a fixed-length time interval that slides continuously over time to capture local variation features of the gas concentration data.

[0041] Step 402: Divide the concentration difference by the time span of the sliding time window to obtain the instantaneous derivative. Specifically, call the displacement command to divide the concentration difference by the time interval constant to obtain the instantaneous derivative value characterizing the rate of change of gas concentration.

[0042] Step 403: Smooth the instantaneous derivatives output from multiple consecutive sliding time windows to obtain the slope. Specifically, perform moving average or first-order low-pass filtering on the instantaneous derivatives output from multiple consecutive sliding time windows to eliminate random fluctuations in single sampling and obtain the smoothed slope; the slope mentioned above refers to the trend rate of change of gas concentration over time.

[0043] Step 404: Calculate the second confidence weight based on the slope. Specifically, when the absolute value of the slope is greater than the preset concentration change threshold, it indicates that the oil fume concentration is changing drastically, and the second confidence weight is set to a high weight value. When the absolute value of the slope is small, the second confidence weight is reduced to avoid interference from the zero-point drift of the gas sensor.

[0044] In one implementation, the length of the sliding time window can be adjusted according to the response delay characteristics of the gas sensor; the greater the response delay, the longer the window.

[0045] By employing the above technical solution, the trend slope of gas concentration is extracted through sliding time window and smoothing filter, eliminating random fluctuations in single sampling. This ensures that the calculation of the second confidence weight is based on the stable trend of concentration change rather than transient noise, thereby improving the reliability of gas channel data in fusion calculation.

[0046] In some embodiments, based on a first confidence weight and a second confidence weight, the time-varying edge gradient in the environmental feature matrix and the instantaneous concentration value of the gas concentration data are concatenated to generate a cooking scene feature vector, including: Step 501: Perform spatial dimensionality reduction on the time-varying edge gradients in the environmental feature matrix to obtain scalar edge feature values. Specifically, the two-dimensional time-varying edge gradients are compressed into scalar edge feature values ​​through regional dimensionality reduction. The dimensionality reduction method can be regional mean, regional maximum, or weighted summation. The spatial dimensionality reduction mentioned here refers to the operation process of compressing multi-dimensional feature data in matrix form into a single scalar value.

[0047] Step 502: Multiply the edge feature value by the first confidence weight to obtain the visual weight feature. Specifically, the multiplier accumulator performs the multiplication operation between the edge feature value and the first confidence weight. When the first confidence weight is zero, the visual weight feature is also zero, and at this time the image channel data is completely masked.

[0048] Step 503: Multiply the instantaneous concentration value by the second confidence weight to obtain the gas weight feature. Specifically, perform the multiplication operation between the instantaneous concentration value and the second confidence weight. The higher the second confidence weight, the greater the influence of gas data in the fusion result.

[0049] Step 504: Perform normalization processing on the visual weight features and gas weight features. Specifically, divide the visual weight features and gas weight features by the sum of their absolute values ​​to eliminate the numerical magnitude differences between different modes, so that the two normalized feature values ​​are within the same numerical range.

[0050] Step 505: Concatenate the normalized visual weight features and gas weight features to generate a one-dimensional cooking scene feature vector. Specifically, the normalized visual weight features and gas weight features are stored in the same contiguous memory address space in a preset order and concatenated to generate a one-dimensional array-like cooking scene feature vector; this vector simultaneously encodes the weighted information of the image channel and the gas channel.

[0051] In one implementation, feature concatenation can use weighted summation instead of vector concatenation, directly adding the two normalized feature values ​​to output a scalar-form scene discrimination value.

[0052] In another implementation, normalization can be achieved by using maximum value normalization instead of absolute value sum normalization, scaling based on the maximum of the two eigenvalues.

[0053] The above technical solution uses confidence weights to weight and then stitch together the visual and gas modal data, enabling the cooking scene feature vector to adjust the proportion of the two data sources in real time according to the degree of lens contamination. When the image data is reliable, it makes full use of image information, and when the image data fails, it smoothly transitions to gas data as the main data source, ensuring the continuity of scene recognition.

[0054] In some embodiments, matching the target noise reduction speed threshold corresponding to the feature vector of the cooking scene includes: Step 601: Obtain multiple standard scene vectors and the anti-jitter compensation values ​​associated with each standard scene vector. Specifically, read the lookup table preset in the read-only memory. This lookup table contains multiple standard scene vectors and the rotational speed parameters and anti-jitter compensation values ​​associated with each vector. The anti-jitter compensation value mentioned here refers to the penalty term superimposed on the distance metric to prevent frequent jumps in scene classification during adjacent control cycles.

[0055] Step 602: Calculate the sum of the absolute values ​​of the differences between the feature vector of the cooking scene and the feature values ​​of the corresponding dimensions of each standard scene vector to obtain the first distance value. Specifically, call the arithmetic logic unit to calculate the Manhattan distance or Euclidean distance between the current cooking scene feature vector and each standard scene vector in the lookup table, as the first distance value; the smaller the first distance value, the more similar the current scene is to the standard scene.

[0056] Step 603: Compare each standard scene vector with the historical output scene of the previous control cycle to see if they are consistent. Specifically, compare each standard scene vector with the target scene output in the previous control cycle to determine whether the current candidate scene is the same as the scene selected in the previous cycle.

[0057] Step 604: For inconsistent standard scene vectors, add the anti-shake compensation value to the corresponding first distance value to obtain the corrected distance; for consistent standard scene vectors, directly use the corresponding first distance value as the corrected distance. Specifically, when the candidate scene is inconsistent with the previous output scene, the anti-shake compensation value is superimposed on the distance value of the candidate scene as an additional penalty, so that scene switching needs to overcome a certain distance margin before it can be triggered, thereby suppressing the back-and-forth jumps of the scene under boundary conditions.

[0058] Step 605: Select the rotational speed parameter associated with the standard scene vector with the smallest corresponding correction distance as the target noise reduction rotational speed threshold. Specifically, iterate through all correction distances and extract the rotational speed parameter bound to the standard scene vector corresponding to the smallest correction distance as the target noise reduction rotational speed threshold.

[0059] In one implementation, distance calculation can use Euclidean distance instead of Manhattan distance, which is more sensitive to differences between feature dimensions.

[0060] In another implementation, the anti-shake compensation value can be increased according to the number of consecutive inconsistent cycles. The longer the inconsistency continues, the lower the switching threshold becomes, thus avoiding being stuck in the wrong scenario for a long time.

[0061] By introducing anti-shake compensation values ​​into the distance metric, the scene switching needs to overcome additional distance margins, effectively suppressing the frequent jumps in scene classification caused by sensor noise between adjacent cycles, and ensuring the smoothness of the fan speed adjustment.

[0062] In some embodiments, extracting the slope of the gas concentration data and calculating a second confidence weight based on the slope includes: Step 701: Extract the slope of the gas concentration data. Specifically, following the methods in steps 401 to 403, apply a sliding time window to the gas concentration data and obtain the slope after smoothing and filtering.

[0063] Step 702: Obtain the temperature sequence output by the temperature sensing component. Specifically, temperature data is collected by a temperature sensor deployed near the air inlet of the range hood, and the rate of change of temperature over time is calculated to obtain the temperature sequence; the temperature sequence referred to here is the derivative sequence of temperature change over time, reflecting the changing trend of heat radiation intensity in the cooking area.

[0064] Step 703: Compare the slope with the rising slope of the temperature sequence. Specifically, compare the directionality of the gas concentration slope with the rising slope of the temperature gradient to determine whether they change in a consistent manner; when the temperature rises sharply while the gas concentration changes slowly, it indicates that the current release is mainly water vapor rather than oil fume.

[0065] Step 704: When the rising slope is greater than the first threshold and less than the second threshold, reduce the second confidence weight. Specifically, when the temperature rising slope exceeds the first threshold, it indicates a significant increase in thermal radiation, but the gas concentration slope is lower than the second threshold, it indicates that the oil fume concentration has not increased synchronously. At this time, it is determined to be a water vapor interference scenario, and the second confidence weight is actively reduced to prevent the gas sensor from falsely triggering the high windshield due to water vapor.

[0066] In one implementation, the temperature sensing component can be a thermocouple or a thermistor, selected based on the response speed and measurement range.

[0067] In another implementation, the first and second thresholds can be calibrated through cooking experiments, and differentiated settings can be made for the steam release characteristics of different cooking appliances.

[0068] By introducing the thermodynamic gradient of the temperature sensor as an auxiliary criterion, the system identifies water vapor interference scenarios and reduces the gas confidence weight when the gas concentration slope and temperature rise slope do not match. This avoids the gas sensor being affected by water vapor and falsely triggering high speed in scenarios such as boiling water or steaming, and further improves the accuracy of scene recognition.

[0069] In some embodiments, after forcibly setting the first confidence weight to zero, the following is included: Step 801: Trigger a shutdown command for the camera module. Specifically, the main control program cuts off the power supply to the camera module or reduces its main control clock frequency to a minimum through register configuration, stopping the entire pipeline of image acquisition and transmission, and releasing the computing power and memory resources of the microcontroller.

[0070] Step 802: Stop extracting time-varying edge gradients from the environmental feature matrix. Specifically, after the camera module is turned off, visual feature calculation steps such as grayscale conversion, high-pass filtering, and edge extraction are no longer executed, and the corresponding memory buffer is marked as invalid.

[0071] Step 803: Switch to generating the target noise reduction speed threshold solely based on gas concentration data. Specifically, the main control program switches the scene recognition logic to a gas channel single-modal mode, matching the target noise reduction speed threshold solely based on a table lookup of the slope and instantaneous concentration value of the gas concentration data, no longer relying on visual features.

[0072] Step 804: Generate a lens surface cleaning reminder signal. Specifically, the main control program sends a lens cleaning reminder to the user interface module of the range hood through the display interface or communication interface, reminding the user to wipe and maintain the lens surface of the camera module.

[0073] In one implementation, the shutdown command can be executed in stages, first stopping the allocation of image processing computing power, and then delaying the power supply to the camera module to avoid interference from power supply transients to other modules.

[0074] In another implementation, the output frequency of the lens surface cleaning prompt signal can be set to an upper limit to avoid repeatedly pushing the same prompt within multiple consecutive control cycles.

[0075] The above technical solution actively shuts down the camera module pipeline and switches to gas single-mode control when the image data is completely lost. This not only frees up the microcontroller's computing resources for rapid sampling and calculation of the gas channel, but also guides users to maintain the lens in a timely manner through cleaning prompts, thus shortening the recovery time of the image channel.

[0076] In some embodiments, obtaining the time-varying edge gradient in the environmental feature matrix includes: Step 901: Obtain the historical environment feature matrix for multiple consecutive historical periods. Specifically, the microcontroller allocates a circular buffer in memory to cache the environment feature matrix for the past N control periods; the circular buffer here refers to a circular storage structure where new data overwrites the oldest data to maintain a fixed length of historical records.

[0077] Step 902: The historical environmental feature matrices are weighted and summed using preset time decay weights to generate a background reference matrix. Specifically, the arithmetic logic unit is invoked to assign larger decay weights to matrices with more recent times, and weighted fusion is performed on the N historical environmental feature matrices to construct a slowly updating static kitchen background model. The aforementioned background reference matrix refers to the reference matrix obtained after weighted fusion of the historical feature matrices, reflecting the static background features of the kitchen, such as stationary walls and slowly moving cookware.

[0078] Step 903: Calculate the absolute difference matrix between the current environmental feature matrix and the background reference matrix. Specifically, perform matrix subtraction between the current frame's environmental feature matrix and the background reference matrix and take the absolute value to obtain the absolute difference matrix; the regions with larger values ​​in this difference matrix correspond to the regions in the current frame that have undergone significant changes relative to the static background, i.e., the areas of oil fume activity.

[0079] Step 904: Perform edge extraction on the absolute difference matrix to obtain the time-varying edge gradient after filtering out static background interference. Specifically, perform edge detection operator operations on the absolute difference matrix to extract the edge contours of the difference region as the time-varying edge gradient; this gradient only contains the time-varying edge information that changes over time in the current frame, and static background interference has been effectively filtered out.

[0080] In one implementation, the time decay weight can adopt an exponential decay model, where the weight of frames further away from the current time is smaller.

[0081] In another implementation, the background reference matrix can be updated using a selective update strategy, updating the background value only at pixel locations determined to be non-smoke areas, thus avoiding pollution of the background model by smoke areas.

[0082] The above technical solution constructs a background reference matrix by weighted fusion of historical frames and then performs a difference with the current frame. This removes the static background of the kitchen from the current frame, retaining only the time-varying oil fume contour information as the time-varying edge gradient. This effectively eliminates the interference of static backgrounds such as fixed stoves and cookware on oil fume edge detection.

[0083] In some embodiments, acquiring the environmental image output by the camera module and the gas concentration data output by the gas sensor includes: Step 1001: Continuously extract the slope of the gas concentration data at the first sampling frequency. Specifically, the gas sensor continuously operates at the first sampling frequency and outputs concentration data, and the microcontroller continuously performs slope calculation on the concentration data; the first sampling frequency mentioned here refers to the sampling frequency under normal operating conditions of the gas sensor, and this frequency must meet the requirement of capturing rapid changes in oil fume concentration.

[0084] Step 1002: Determine whether the slope is greater than the preset visual wake-up threshold. Specifically, compare the calculated slope with the preset visual wake-up threshold; the visual wake-up threshold is a critical value for determining whether there is a cooking activity that requires the intervention of the image channel, and it is determined based on cooking experiment data during the factory calibration stage.

[0085] Step 1003: When the slope is less than or equal to the visual wake-up threshold, the camera module is kept in a low-power sleep state, and the first confidence weight is directly set to zero, skipping the extraction and calculation steps for the environmental feature matrix. Specifically, when the slope does not exceed the wake-up threshold, it indicates that the current cooking activity intensity is low and no image channel intervention is required. The main control program keeps the camera module in a low-power sleep state and directly overwrites the visually related variables to zero, skipping all visual feature calculation steps to save computing power and power consumption.

[0086] Step 1004: When the slope is greater than the visual wake-up threshold, a wake-up signal is generated. Specifically, when the slope exceeds the wake-up threshold, a wake-up signal is generated by the interrupt controller; the wake-up signal referred to here is the control level signal that triggers the camera module to switch from sleep mode to working mode.

[0087] Step 1005: Trigger the camera module to output an environmental image at a second sampling frequency using a wake-up signal, wherein the second sampling frequency is lower than the first sampling frequency. Specifically, in response to the wake-up signal, the main control program restores the power supply to the camera module and configures the second sampling frequency through a frequency divider; the second sampling frequency is lower than the first sampling frequency because the computing power required for processing a single frame of image in the image channel is much greater than that required for a single sampling in the gas channel, and reducing the frame rate can ensure the real-time performance of image processing under limited computing power.

[0088] In another implementation, the second sampling frequency can be adjusted in stages according to the magnitude of the current slope; the larger the slope, the higher the frame rate to capture denser visual information.

[0089] By using the above technical solution, the gas concentration slope is used as the wake-up condition for the image channel. The camera module is kept in sleep mode when the cooking activity intensity is low, and visual acquisition is only started when a significant increase in oil fume concentration is detected. This saves the power consumption of the camera module and the computing power of the microcontroller, while ensuring that it can be woken up in time when visual information is needed.

[0090] In some embodiments, after forcibly setting the first confidence weight to zero, the method further includes: Step 1101: Perform a difference operation on the slope of the gas concentration data to obtain the second-order rate of change of concentration. Specifically, the arithmetic logic unit in the main control chip performs a subtraction operation on the gas concentration slope of adjacent control cycles again to obtain the rate of change of the slope, i.e., the second-order rate of change of concentration; the second-order rate of change of concentration mentioned here refers to the rate of change of the gas concentration slope, reflecting the acceleration characteristics of the change in oil fume concentration.

[0091] Step 1102: Determine whether the second-order rate of change of concentration is greater than the preset mutation threshold. Specifically, compare the second-order rate of change of concentration with the preset mutation threshold; the mutation threshold is the critical value for determining whether the concentration of cooking fumes has changed drastically, and it is determined during the factory calibration stage based on experimental data from vigorous cooking such as stir-frying.

[0092] Step 1103: When the second-order rate of change of concentration exceeds the mutation threshold, a preset compensation mapping table is obtained. The compensation mapping table records the correspondence between the second-order rate of change of concentration and the edge compensation value. Specifically, when the rate of change exceeds the mutation threshold, the main control chip reads the compensation mapping table preset in the memory. The aforementioned compensation mapping table refers to a data grid that records the quantitative correspondence between the second-order rate of change of concentration and the edge compensation value, which is established through a control experiment during the calibration phase.

[0093] Step 1104: Match the corresponding edge compensation value from the compensation mapping table based on the current second-order rate of change of concentration. Specifically, use the calculated second-order rate of change of concentration as the addressing index to find the corresponding edge compensation value in the compensation mapping table; the aforementioned edge compensation value refers to the alternative edge feature estimate calculated based on the drastic change characteristics of gas concentration when the image channel fails.

[0094] Step 1105: Replace the time-varying edge gradient with the edge compensation value and concatenate it with the instantaneous concentration value to generate a cooking scene feature vector. Specifically, push the edge compensation value into the memory stack of the feature concatenation, replacing the originally missing time-varying edge gradient data, and perform normalization and concatenation with the instantaneous concentration value according to the methods in steps 504 and 505 to generate a cooking scene feature vector.

[0095] Step 1106: Within a preset number of control cycles, if the second-order rate of change of concentration is not greater than the mutation threshold, then stop matching edge compensation values. Specifically, within a preset number of control cycles, if the second-order rate of change of concentration falls back to a safe range, then exit the compensation mode and stop matching edge compensation values ​​from the compensation mapping table; the state mentioned here, less than the mutation threshold, indicates that the change in oil fume concentration has tended to be stable and no further edge compensation is needed.

[0096] In one implementation, the compensation mapping table can be implemented using linear interpolation, where the edge compensation value corresponding to the intermediate value is calculated between discrete calibration points using linear interpolation.

[0097] In another implementation, the decay of the edge compensation value can adopt a step-by-step decreasing strategy, gradually reducing the compensation value instead of immediately returning to zero when exiting the compensation method, so as to avoid the feature vector from jumping.

[0098] By using the above technical solution, when the image channel fails due to severe lens contamination, the second-order rate of change of gas concentration is used to match the edge compensation value from the compensation mapping table to replace the time-varying edge gradient. This allows for the generation of an approximate cooking scene feature vector based on the abrupt change characteristics of the gas channel even when image data is missing, thus avoiding the problem of complete interruption of scene recognition during visual failure.

[0099] In some embodiments, after triggering the camera module to output an environmental image at a second sampling frequency using a wake-up signal, the method further includes: Step 1201: Determine whether the absolute value of the slope within a preset number of consecutive control cycles is within a preset lower limit fluctuation range. Specifically, the microcontroller counts whether the absolute value of the gas concentration slope within multiple consecutive control cycles remains within a preset lower limit fluctuation range; the lower limit fluctuation range refers to a range where the absolute value of the slope is at an extremely low level, indicating that the intensity of cooking activity has significantly decreased.

[0100] Step 1202: When the absolute value is within the lower limit fluctuation range, a frame adjustment command is sent to the camera module to limit the output frame of the camera module to the preset center observation area. Specifically, when the microcontroller determines that the slope is continuously within the lower limit fluctuation range, the main control program writes preset frame format parameters to the control register of the camera module through the control interface, switching it from full-frame output mode to hardware cropping mode; in hardware cropping mode, the clock readout of the pixels on the outer periphery of the sensor target surface is blocked, and only the pixel stream aligned with the core area of ​​the stove is output.

[0101] Step 1203: Obtain the pixel array corresponding to the central observation area and use the pixel array as the environment image. Specifically, the microcontroller uses the received local pixel array of the central observation area as a new environment image for subsequent feature extraction and scene recognition calculations; due to the reduced image size, the amount of data processed is reduced accordingly, further reducing the computational burden on the microcontroller.

[0102] In one implementation, the position of the central observation area can be adjusted by a preset offset parameter to accommodate the differences in the relative installation positions of the camera module and the cooktop in different models of range hoods.

[0103] In another implementation, the frame adjustment command can be automatically canceled and the full-frame output mode restored when the absolute value of the slope exceeds the lower limit fluctuation range again.

[0104] By using the above technical solution, when the intensity of cooking activities is reduced, the output frame of the camera module is limited to the central observation area, which reduces the amount of data processing for irrelevant background areas and reduces the computing power consumption of the microcontroller while ensuring the integrity of the visual information of the core area of ​​the stove.

[0105] In some embodiments, generating a fan drive duty cycle corresponding to a target noise reduction speed threshold and sending the fan drive duty cycle to the fan actuator includes: Step 1301: Obtain the current reference operating speed of the fan actuator. Specifically, the current actual operating speed is read through the speed feedback channel of the fan actuator as the reference operating speed; this reference operating speed reflects the working state of the fan before receiving a new duty cycle command.

[0106] Step 1302: Calculate the speed difference between the target noise reduction speed threshold and the reference operating speed. Specifically, subtract the reference operating speed from the target noise reduction speed threshold to obtain the speed difference; a positive speed difference indicates that acceleration is needed, and a negative speed difference indicates that deceleration is needed.

[0107] Step 1303: Input the speed difference into the preset smooth transition function equation to calculate multiple time-increasing or decreasing transition speed nodes. Specifically, the speed difference is input into the smooth transition function equation, which decomposes the speed change into multiple time-increasing or decreasing transition speed nodes according to preset acceleration constraints. The smooth transition function equation mentioned here refers to a mathematical function that transforms the speed step command into a continuously changing speed sequence, which can be achieved using trapezoidal speed planning or S-shaped speed curves. The transition speed node refers to the intermediate speed value during the process of speed transitioning from the reference value to the target value.

[0108] Step 1304: Convert the transition speed nodes step by step into the corresponding fan drive duty cycles, and send the fan drive duty cycles to the fan actuator. Specifically, based on the clock frequency inside the microcontroller, convert each transition speed node into the set value of the duty cycle register, change the high-level duration of the pulse width modulation signal, and control the stator coil current of the fan motor step by step through the drive circuit to achieve a smooth speed transition.

[0109] In one implementation, the smooth transition function equation can adopt an S-shaped velocity curve, and jerk constraints can be introduced during acceleration and deceleration to make the speed change smoother.

[0110] In another implementation, the number of transition speed nodes can be adjusted according to the size of the speed difference; the larger the difference, the more nodes are needed to keep the speed change rate constant per unit time.

[0111] The above technical solution decomposes the speed step command into gradual transition speed nodes through a smooth transition function, so that the fan speed smoothly transitions from the current value to the target value, avoiding the mechanical shock and additional noise of the fan caused by sudden speed changes, and ensuring that the noise of the noise reduction adjustment process itself will not have a negative impact on the user experience.

[0112] Those skilled in the art will 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, or a combination of computer software and electronic hardware. 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.

[0113] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0114] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0115] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0116] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0117] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program verification codes, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0118] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more features. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0119] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0120] The embodiments, implementation methods, and related technical features of this application can be combined and substituted for each other without conflict.

[0121] The above are merely preferred embodiments of this application and are not intended to limit this application in any way. Any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of this application without departing from the technical solution of this application shall still fall within the scope of the technical solution of this application.

Claims

1. A method for intelligently adjusting the noise of a range hood, characterized in that, include: Acquire environmental images output by the camera module and gas concentration data output by the gas sensor; The environmental image is converted into an environmental feature matrix, and the contrast attenuation coefficient corresponding to the reference area is extracted from the environmental feature matrix. The first confidence weight is calculated based on the contrast attenuation coefficient; Extract the slope of the gas concentration data and calculate a second confidence weight based on the slope; Based on the first confidence weight and the second confidence weight, the time-varying edge gradient in the environmental feature matrix and the instantaneous concentration value of the gas concentration data are concatenated to generate a cooking scene feature vector; Match the target noise reduction speed threshold corresponding to the feature vector of the cooking scene; Generate the fan drive duty cycle corresponding to the target noise reduction speed threshold and send the fan drive duty cycle to the fan actuator.

2. The method according to claim 1, characterized in that, The environmental image is converted into an environmental feature matrix, and the contrast attenuation coefficient corresponding to the reference area in the environmental feature matrix is ​​extracted, including: Extract the luminance component of the environmental image; construct a two-dimensional numerical array based on the luminance component to obtain the environmental feature matrix; According to the preset coordinate mapping relationship, the pixel submatrix corresponding to the reference area is obtained from the environmental feature matrix; A preset high-pass filter operator is applied to the pixel submatrix to filter out slowly varying illumination components, thereby generating an edge matrix; Calculate the absolute value of the grayscale difference between adjacent pixels within the edge matrix; The absolute values ​​are summed to obtain the absolute error, which is used as the current sharpness representation value. Obtain the initial edge characterization reference value of the reference area in a contamination-free state; The difference between the initial edge representation benchmark value and the current sharpness representation value is calculated, and the contrast attenuation coefficient is obtained based on the difference and the initial edge representation benchmark value.

3. The method according to claim 2, characterized in that, The first confidence weight is calculated based on the contrast attenuation coefficient, including: Obtain a preset first attenuation threshold and a second attenuation threshold, wherein the second attenuation threshold is greater than the first attenuation threshold; When the contrast attenuation coefficient is less than the first attenuation threshold, the first confidence weight is set to the full load weight benchmark value; When the contrast attenuation coefficient is between the first attenuation threshold and the second attenuation threshold, the contrast attenuation coefficient is input into a preset nonlinear smooth attenuation function to calculate the decreasing first confidence weight. When the contrast attenuation coefficient is greater than or equal to the second attenuation threshold, the first confidence weight is forcibly set to zero.

4. The method according to claim 3, characterized in that, Extracting the slope of the gas concentration data and calculating a second confidence weight based on the slope includes: A sliding time window is applied to the gas concentration data to extract the concentration difference within the time window; The instantaneous derivative is obtained by dividing the concentration difference by the time span of the sliding time window. The slope is obtained by smoothing and filtering the instantaneous derivatives output by multiple consecutive sliding time windows; The second confidence weight is calculated based on the slope.

5. The method according to claim 4, characterized in that, Based on the first confidence weight and the second confidence weight, the time-varying edge gradient in the environmental feature matrix and the instantaneous concentration value of the gas concentration data are concatenated to generate a cooking scene feature vector, including: Spatial dimensionality reduction is performed on the time-varying edge gradient in the environmental feature matrix to obtain scalar edge feature values; Multiply the edge feature value by the first confidence weight to obtain the visual weight feature; Multiply the instantaneous concentration value by the second confidence weight to obtain the gas weighted feature; Normalization is performed on the visual weight features and the gas weight features; The normalized visual weight features are concatenated with the gas weight features to generate a one-dimensional feature vector of the cooking scene.

6. The method according to claim 5, characterized in that, Matching the target noise reduction speed threshold corresponding to the feature vector of the cooking scene includes: Obtain multiple standard scene vectors and the anti-shake compensation values ​​associated with each of the standard scene vectors; The first distance value is obtained by summing the absolute values ​​of the differences between the feature vector of the cooking scene and the feature values ​​of the corresponding dimensions of each standard scene vector; Compare each of the standard scenario vectors with the historical output scenario of the previous control cycle to see if they are consistent. For inconsistent standard scene vectors, the corresponding first distance value is added to the anti-shake compensation value to obtain the corrected distance; for consistent standard scene vectors, the corresponding first distance value is directly used as the corrected distance. The rotational speed parameter associated with the standard scene vector that has the smallest correction distance is selected as the target noise reduction rotational speed threshold.

7. The method according to claim 6, characterized in that, Obtaining the slope of the gas concentration data and calculating the second confidence weight based on the slope includes: Extract the slope of the gas concentration data; Acquire the temperature sequence output by the temperature sensing component; Compare the slope with the rising slope of the temperature sequence; When the upward slope is greater than a first threshold and the slope is less than a second threshold, the second confidence weight is reduced.

8. The method according to claim 7, characterized in that, After forcibly setting the first confidence weight to zero, the method further includes: Trigger a shutdown command for the camera module; Stop extracting the time-varying edge gradient from the environmental feature matrix; Switch to generating the target noise reduction speed threshold solely based on the gas concentration data; Generate a lens surface cleaning prompt signal.

9. The method according to claim 8, characterized in that, Obtaining the time-varying edge gradient in the environmental feature matrix includes: Obtain the historical environmental feature matrix for multiple consecutive historical periods; The background reference matrix is ​​generated by weighting and summing the historical environmental feature matrices using a preset time decay weight. Calculate the absolute difference matrix between the environmental feature matrix and the background reference matrix at the current moment; An edge extraction operation is performed on the absolute difference matrix to obtain the time-varying edge gradient after filtering out static background interference.

10. The method according to claim 9, characterized in that, Acquire environmental images output by the camera module and gas concentration data output by the gas sensor, including: The slope of the gas concentration data is continuously extracted at a first sampling frequency; Determine whether the slope is greater than a preset visual wake-up threshold; When the slope is less than or equal to the visual wake-up threshold, the camera module is kept in a low-power sleep state, and the first confidence weight is directly set to zero, skipping the extraction and calculation steps for the environmental feature matrix. When the slope is greater than the visual wake-up threshold, a wake-up signal is generated; The wake-up signal triggers the camera module to output the environmental image at a second sampling frequency, wherein the second sampling frequency is lower than the first sampling frequency.

Images