Control method and system of high-speed double-frequency blower

Abstract

The invention discloses a control method and system of a high-speed double-frequency hair dryer. The control method of the high-speed double-frequency hair dryer comprises the following steps: setting and constructing a double-branch heterogeneous encoder, inputting the characteristic information of hair color, curling degree and operation data into a branch I, matching a response control interval, inputting the characteristic information of fracture, forking and operation data into a branch II, constructing an evaluation calculation model, and calculating the hair color, the curling degree and the operation data according to the evaluation calculation model. Generating a hair damage index to divide a plurality of damage levels, and triggering different branch fusion strategies; through mutual response of the first branch and the second branch, the double-frequency response value of the blower is accurately output, intelligent self-adaptive control is achieved, through the double-frequency control model, the nonlinear mapping relation between the video data and the operation data and the double-frequency response value can be established in real time, and therefore it is guaranteed that on the premise of high-speed control of the blower, the high-speed control of the blower is achieved. And the working state of the blower is accurately adjusted.

Description

Technical Field

[0001] The present invention relates to the technical field of hair dryers, and specifically to a control method and system for a high-speed dual-frequency hair dryer. Background Art

[0002] Hair dryers are essential household appliances in people's daily lives. With the development of technology, high-speed dual-frequency hair dryers have emerged, which combine the advantages of high-speed wind and wind temperature switching and are realized by using a high-speed electric motor (such as a high-speed brushless motor), thereby accelerating the evaporation of moisture on the hair surface, enabling the hair to dry faster, and not causing excessive heat damage to the hair while blowing at high speed;

[0003] On the one hand, traditional hair dryers mostly adopt fixed gear control and cannot dynamically adjust the power according to environmental light, user hair quality or real-time usage scenarios, resulting in high energy consumption or poor user experience. Moreover, the design of traditional hair dryers mostly relies on a single wind speed or temperature control. Although some hair dryers support multi-gear switching, they lack an intelligent adaptive control logic and cannot accurately regulate various operating parameters, resulting in the inability to optimize the adjustment according to the hair during use;

[0004] On the other hand, existing hair dryers lack a real-time monitoring and feedback system and cannot effectively control the coordination between temperature, wind speed, etc., making it difficult to ensure that the hair dryer maintains the best working state during operation. Especially in the high-speed operation state, the wind speed and rotation speed data of the hair dryer are prone to fluctuations, resulting in unstable heating efficiency, energy waste and even heat damage to the hair. Summary of the Invention

[0005] (I) Technical Problems to be Solved

[0006] In view of the deficiencies of the prior art, the present invention provides a control method and system for a high-speed dual-frequency hair dryer, which solves the problems raised in the background art.

[0007] (II) Technical Solutions

[0008] To achieve the above objectives, the present invention is realized through the following technical solutions:

[0009] The present invention provides a control method for a high-speed dual-frequency hair dryer, including the following steps:

[0010] Collect the head video data and operation data of the high-speed dual-frequency hair dryer, number the head video data in the scanning order, and use image processing technology to extract the feature information of the hair;

[0011] Among them, the feature information at least includes breakage, splitting, hair color and curl;

[0012] After normalizing the characteristic information of the hair and synchronizing the time series, it is respectively cascaded with the classified characteristic information of the extracted operating data, and fusion and dimensionality reduction are completed. A fusion framework is constructed using BLSTM to identify the fusion features, obtaining a dual-frequency control model and outputting the dual-frequency response values of the corresponding sub-regions;

[0013] Among them, the dual-frequency response values at least include rotational speed, temperature, and wind speed;

[0014] Based on the dual-frequency response values, calculate the efficiency and power ratio of the corresponding high-speed dual-frequency hair dryer at time t, fit to obtain the efficiency function and power ratio function, calculate the optimization function, import the optimization function into the dual-frequency control model for training and optimization, and use the gradient descent method to optimize the dual-frequency control model.

[0015] Furthermore, the steps of extracting the characteristic information of the hair using image processing technology include:

[0016] Extract key frames: Perform frame difference processing on the sequentially numbered head video data frames, divide the head region into several sub-regions, and use the SSIM technology to analyze the processed results in each sub-region to obtain several key frame images;

[0017] Filtering process: Filter the key frame images using wavelet basis functions;

[0018] Hair color recognition: Extract the color components of the filtered image on each RGB channel, weighted average the color components to calculate the initial brightness, introduce the indoor light intensity to correct the initial brightness, and construct a color space in each sub-region to identify the hair color;

[0019] Fork recognition: Use the Zhang-Suen thinning algorithm to determine whether there are Y-shaped intersections in each sub-region. If so, mark them as forks and obtain the fork density and fork angle variance;

[0020] Curvature recognition: Based on each sub-region, obtain the local curvature, statistically calculate the maximum value, minimum value, average value, fluctuation value, skewness, and kurtosis of the local curvature to calculate the curl evaluation coefficient, and combine the standard curl threshold to judge the curl type; among them, the curl types include straight hair type, wavy hair type, and spiral hair type;

[0021] Breakage recognition: Use Canny edge detection and Hough line transformation to identify the breakage lines in each sub-region. When the length of the breakage line is less than the preset breakage threshold, mark it as a breakage and obtain the breakage density.

[0022] Furthermore, the steps of calculating the curl evaluation coefficient include:

[0023] Obtain the local curvature of a sub-region, construct a six-dimensional monitoring vector [max, min, mean, volatility, skewness, kurtosis] with the maximum value, minimum value, average value, volatility value, skewness, and kurtosis of the local curvature, and construct judgment vectors vec1, vec2, and vec3 according to different combinations;

[0024] Combine vec1, vec2, and vec3, and set a formula to calculate the crimp evaluation coefficient:

[0025]

[0026] In the formula, crimp represents the crimp evaluation coefficient, vec1_th represents the baseline threshold, σ is the scaling factor, ε1, ε2, and ε3 all represent adjustment factors, and σ, ε1, ε2, and ε3 are all greater than 0.

[0027] Furthermore, the steps of the classification cascade include:

[0028] Construct a dual-branch heterogeneous encoder, including Branch One and Branch Two;

[0029] Input the characteristic information of hair color, crimp, and running data into Branch One, match the response control interval, and generate several modes based on the response control interval, including the gentle wind mode, quick-drying mode, and hair care mode; among them, the response control interval includes the temperature interval and wind speed interval of the hair dryer;

[0030] At the same time, input the characteristic information of breakage, splitting, and running data into Branch Two, build an evaluation calculation model, generate a hair damage index, divide several damage levels, and trigger different branch fusion strategies;

[0031] If branch fusion strategy one is triggered, perform gated residual fusion;

[0032] If branch fusion strategy two is triggered, perform bidirectional attention fusion;

[0033] If branch fusion strategy three is triggered, perform multi-scale gated fusion.

[0034] Furthermore, the calculation formula of the evaluation calculation model is:

[0035]

[0036] In the formula, H damage represents the hair damage index, F1 represents the breakage density, F2 represents the splitting density, ψ represents the constant term, α1, α2, and α3 all represent adjustment coefficients, and α1, α2, and α3 are all greater than 0.

[0037] Furthermore, the formula on which the efficiency function is based:

[0038]

[0039] In the formula, η(t) represents the efficiency of the hair dryer obtained at time t, and b in , b out both represent constants related to the characteristics of the hair dryer and the motor. wind(t) represents the wind speed obtained at time t, temp(t) represents the temperature obtained at time t, and speed(t) represents the rotational speed obtained at time t;

[0040] The obtained efficiency function by fitting is marked as: η(wind, speed, temp) = gn1;

[0041] The calculation formula of the power ratio function is:

[0042]

[0043] In the formula, PowerRate(t) represents the power ratio obtained at time t;

[0044] The obtained power ratio function by fitting is marked as: PowerRate(wind, speed, temp) = gn2.

[0045] Furthermore, the calculation formula of the optimization function is:

[0046]

[0047] In the formula, O(t) represents the optimization function at time t, κ1, κ2, and κ3 all represent weight coefficients, and κ1, κ2, and κ3 are all greater than 0, and (κ3 * H damage ) represents the penalty term.

[0048] The present invention provides a control system for a high-speed dual-frequency hair dryer, including:

[0049] A data acquisition module that acquires the head video data and operation data of the high-speed dual-frequency hair dryer, and numbers the head video data in the scanning order;

[0050] A feature extraction module that extracts the feature information of the hair by using image processing technology; wherein, the feature information includes breakage, splitting, hair color, and curl;

[0051] A classification and fusion module that normalizes the feature information of the hair and synchronizes the time series, and then classifies and cascades the feature information of the extracted operation data respectively, and completes fusion and dimensionality reduction. A fusion framework is constructed by using BLSTM to identify the fusion features, obtains a dual-frequency control model, and outputs the dual-frequency response values of the corresponding sub-regions; wherein, the dual-frequency response values include rotational speed, temperature, and wind speed;

[0052] The model optimization module calculates the efficiency and power ratio of the high-speed dual-frequency hair dryer at time t based on the dual-frequency response value, fits to obtain the efficiency function and power ratio function, calculates the optimization function, and uses the gradient descent method to optimize the dual-frequency control model.

[0053] (III) Beneficial Effects

[0054] The present invention provides a control method and system for a high-speed dual-frequency hair dryer, having the following beneficial effects:

[0055] 1. By building a dual-frequency control model, the present invention can establish a non-linear mapping relationship between video data and operation data and the dual-frequency response value in real time, and can capture the dynamic changes in the monitoring of hair properties during the operation of the hair dryer through the dual-frequency control model, so as to realize automatic adjustment of the control work of the hair dryer;

[0056] 2. By setting up a dual-branch heterogeneous encoder, the present invention inputs the characteristic information of hair color, curl degree, and operation data into Branch 1 to match the response control interval, inputs the characteristic information of breakage, split ends, and operation data into Branch 2, builds an evaluation calculation model, generates a hair damage index, divides several damage levels, and triggers different branch fusion strategies; through the mutual response of Branch 1 and Branch 2, the dual-frequency response value of the hair dryer is accurately output to realize intelligent adaptive control;

[0057] 3. By using the efficiency function and power ratio function, the present invention calculates the optimization function and uses the gradient descent method to optimize the dual-frequency control model to ensure that the hair dryer maintains the best working state under the premise of high-speed control. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 is a schematic flowchart of a control method for a high-speed dual-frequency hair dryer shown according to an exemplary embodiment;

[0059] Figure 2 is an operation block diagram of a dual-frequency control model shown according to an exemplary embodiment;

[0060] Figure 3 is a schematic diagram of the modules of a control system for a high-speed dual-frequency hair dryer shown according to an exemplary embodiment. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0061] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0062] Example 1

[0063] An embodiment of the present invention provides a control method for a high-speed dual-frequency hair dryer; Figure 1 It is a schematic flowchart of a control method for a high-speed dual-frequency hair dryer shown according to an exemplary embodiment; Figure 2 It is an operation block diagram of a dual-frequency control model shown according to an exemplary embodiment; Please refer to Figure 1 - Figure 2 , the method includes the following steps:

[0064] S1. Collect the head video data and operation data of the high-speed dual-frequency hair dryer, number the head video data according to the scanning order, and use image processing technology to extract the characteristic information of the hair; wherein, the characteristic information includes breakage, splitting, hair color, and curl;

[0065] Among them, the step of using image recognition technology to extract the characteristic information of the hair includes:

[0066] Extract key frames: Perform frame difference processing on the sequentially numbered head video data frames, divide the head area into several sub-areas, and use the SSIM technology to analyze the processed results in each sub-area to obtain several key frame images;

[0067] Specifically, for any sub-area, perform differential processing on the corresponding video data at the same time interval, calculate the absolute difference between the current frame and the previous frame to obtain the forward difference image, calculate the absolute difference between the next frame and the current frame to obtain the backward difference image; for each current frame, calculate the structural similarity index of the forward difference image and the backward difference image, and the SSIM value ranges from [0, 1], the closer to 1, the more similar; when the SSIM value corresponding to the current frame is less than the preset similarity threshold, it indicates that the movement or scene change before and after this frame is significant, then use the frame image of the current frame as the key frame image; wherein, the time interval is a user-defined parameter, usually set to 0.1 second;

[0068] In addition, the difference image usually needs to be converted to a grayscale image to reduce the calculation amount;

[0069] It should be noted that in the process of collecting the head video data and operation data of the high-speed dual-frequency hair dryer, multiple sensors are used for collection. The head video data is collected through a micro camera. The operation data includes the wind speed, flow rate, vibration, temperature, rotation speed of the hair dryer. The temperature is monitored through a temperature sensor, the vibration is monitored through a mechanical sensor, the wind speed is monitored through a wind speed sensor, the flow rate is monitored through a gas flow sensor, and the rotation speed of the hair dryer is monitored through a rotation speed sensor. At the same time, multi-axis synchronization is set in the collection process. For example, the MEMS mechanical sensor can be embedded in the fan or the air outlet area of the hair dryer to monitor the force of the air flow on the hair, or placed in the handle area of the hair dryer to detect the holding force and vibration. The micro camera is installed at the front end of the hair dryer, close to the air outlet position, to monitor the state of the hair during blowing. To avoid interference from excessive temperature or other external environmental factors, the sensor may need to be equipped with certain cooling or isolation devices, which will not be elaborated here. Usually, the transmitting and receiving parts of the sensor need to be closely arranged to ensure the accuracy of the signal. These sensors are integrated into a miniaturized system, and synchronous signals or clocks are used to ensure that the data collection of each sensor can be consistent. The clock synchronization of the sensor can be achieved through hardware or software (such as through an external clock signal or a synchronization protocol), and the above sensors or devices are not shown in the figure.

[0070] In addition, by obtaining key frame images in each sub-region, it is also possible to first determine whether the obtained images belong to the hair roots, the middle part of the hair, or the hair tips, and set initial temperature and wind speed ranges based on the region, providing a premise for subsequent algorithm design, which will not be elaborated here.

[0071] Filtering process: Use wavelet basis functions to filter the key frame images to obtain high-frequency sub-images and low-frequency sub-images. Among them, the filtering process includes using high-pass filtering and low-pass filtering.

[0072] High-pass filtering: It is used to suppress low frequencies (the main body of healthy hair), retain high frequencies, and quickly and accurately distinguish split / fracture features. For example, when the high-frequency energy ratio (the total energy of the high-frequency region of the filtered image to the energy of the original spectrum) is greater than a preset ratio threshold, assumed to be 50%, it is marked as a fracture / split candidate area to prepare for subsequent analysis of fracture and split features.

[0073] Low-pass filtering: It is used to retain low-frequency components to observe the overall density and uniformity of the hair, and accurately analyze the hair color.

[0074] In addition, homomorphic filtering is used to enhance high-frequency details and retain low-frequency features. The formula is designed as:

[0075]

[0076] Wherein, H(u, v) represents a binarization transfer function for low-frequency retention and high-frequency enhancement, ω Low and ω high respectively control the amplitudes of low-frequency and high-frequency enhancement, (u, v) represents the frequency domain coordinates, D(u, v) represents the distance from the frequency domain point to the center of the frequency domain, and D0 represents the cut-off frequency for dividing the frequency range of retention or suppression;

[0077] Feature extraction:

[0078] Hair color recognition: Extract the color components of the filtered image on each RGB channel, and perform weighted averaging on the color components to calculate the initial brightness. Introduce the indoor light intensity to correct the initial brightness. By taking the red / green component as the X-axis, the blue component as the Y-axis, and the corrected brightness as the Z-axis, establish a color space in each sub-region to recognize the hair color;

[0079] Among them, the formula for calculating the initial brightness is: L = 0.30R + 0.59G + 0.11B;

[0080] In the formula, L is the initial brightness, R is the red component, G is the green component, and B is the blue component;

[0081] Then the formula for correcting the initial brightness L is:

[0082] In the formula, L' represents the corrected brightness, E env represents the collected ambient light intensity, E std represents the standard ambient light intensity, γ represents the correction coefficient, usually taking a value of 2.2 for simulating the non-linearity of human eye perception, and β represents the brightness offset for compensating the absolute brightness loss in low light;

[0083] Formula explanation: The lower the value of L’ (the darker the color), the higher the heat absorption efficiency, and the drying time is negatively correlated with the value of L’; the higher the value of the red / green component (R / G) (more reddish), the higher the drying efficiency (because the red light has a longer wavelength and stronger penetration), and the blue component (high B value) belongs to short-wavelength visible light (450 - 490nm) and has a higher reflectivity to near-infrared light (700 - 2500nm). Since the heat of the hair dryer is mainly transmitted in the infrared band, it takes longer time or higher wind speed to dry (it is preferred to increase the wind speed rather than the temperature to accelerate the evaporation of moisture by forced convection);

[0084] Effect: Through the combination of the R / G ratio and the B value, fine color deviations can be distinguished (for example: red-brown and wine red). After brightness correction, the color space is more in line with the actual visual perception, providing accurate color modeling support for the subsequent hair drying scenario;

[0085] In addition, by constructing a color space, the pixels of each key-frame image are in the coordinates of (X(R / G), Y(B), Z(L)), that is, the hair color can be represented. The hair color is compared with the standard color library, and the Euclidean distance is used to classify the hair color into light color level, medium color level, and dark color level;

[0086] Fork recognition: Use the Zhang-Suen thinning algorithm to determine whether there are Y-shaped intersections in each sub-region. If there are, mark them as forks, and obtain the fork point density and the variance of the fork angle;

[0087] Among them, the steps to determine whether there is a Y-shaped intersection include:

[0088] Obtain the single hair skeleton through the Zhang-Suen thinning algorithm, and select any skeleton point P(x, y) that satisfies the condition: Then it is determined that there is a Y-shaped intersection; in the formula, Cp represents the connection number, θ represents the branch angle, and there are 3 θs;

[0089] Fork point density: the number of fork nodes / area of the region, indicating the density of forks;

[0090] Variance of fork angle: var(θ), indicating the irregularity of the fork direction;

[0091] Curvature recognition: Based on each sub-region, convert the skeleton line into a continuous parametric curve, obtain the local curvature, and count the maximum value, minimum value, average value, fluctuation value, skewness, and kurtosis of the local curvature in any sub-region, and combine the standard curvature threshold interval to judge the curl type; among them, it includes straight hair type, wavy hair type, and spiral hair type;

[0092] Specifically, in the process of obtaining the local curvature, the B-spline method can be used to smooth the set of skeleton points, generate a continuous curve, and calculate the first and second derivatives of the parametric curve;

[0093] Assume that the B-spline method is selected, then the code for calculating the local curvature is as follows:

[0094] # Calculate the first and second derivatives of the B-spline curve

[0095] x_prime = splev(xnew, tck, der = 1) # First derivative

[0096] x_double_prime = splev(xnew, tck, der = 2) # Second derivative

[0097] # Calculate the local curvature (local curvature of a two-dimensional curve)

[0098] curvature = np.abs(x_prime[1] * x_double_prime[0] - x_prime[0] * x_double_prime[1]) / np.linalg.norm(x_prime) ** 3

[0099] # Plot local curvature

[0100] plt.plot(xnew, curvature, 'g-', label='Curvature')

[0101] plt.legend()

[0102] plt.show()

[0103] Furthermore, in any sub-region, a six-dimensional monitoring vector [max, min, mean, volatility, skewness, kurtosis] is constructed using the maximum value, minimum value, average value, fluctuation value, skewness, and kurtosis of the local curvature, and judgment vectors are constructed according to different combinations;

[0104] The constructed judgment vectors are as follows:

[0105] vec1 = [|max * min * mean * volatility * skewness * kurtosis|];

[0106] Feature: Design the product of multi-dimensional features to capture the non-linear synergistic effect between statistics, and amplify significant abnormal combinations. For example, capture the joint features of high curvature, high volatility, and distribution pattern to distinguish spiral curls (high Vec1) from straight hair (low Vec1);

[0107] vec2 = [max - min, |mean - volatility|, skewness + kurtosis];

[0108] Feature: Design the difference characteristics to identify the distribution asymmetry by comparing the range, mean - volatility balance, and skewness - kurtosis superposition;

[0109] vec3 = [max 2 + min 2 , mean 2 + volatility 2 , skewness 2 + kurtosis 2 ;

[0110] Feature: Design the sum of squares form to quantify the feature intensity and suppress noise interference;

[0111] Combine vec1, vec2, and vec3 to set up a formula to calculate the crimp evaluation coefficient:

[0112]

[0113] In the formula, crimp represents the crimp evaluation coefficient, vec1_th represents the baseline threshold for controlling the exponential decay amplitude, σ is the scaling factor for suppressing the overflow of excessive vec1 data, and ε1, ε2, and ε3 all represent adjustment factors for controlling the influence degree of different factors on the crimp evaluation coefficient, and σ, ε1, ε2, and ε3 are all greater than 0;

[0114] Formula explanation: The formula combines an exponential function and a logarithmic function, making the influence of the judgment vector vec1 on the crimp evaluation coefficient a non-linear relationship. When vec1 is small, the value of this part is close to 1. On the contrary, when the value of vec1 is larger, the value of this part gradually increases, and the increasing speed becomes faster and faster;

[0115] (ε2*‖vec2||2): The formula combines a norm term, which reflects the comprehensive strength of the judgment vector vec2 on the crimp evaluation coefficient. The larger the norm, the more curly it indicates;

[0116] [ε3*log(1 + vec3)]: The formula combines a logarithmic function to suppress noise and outliers and improve robustness. When vec3 is small, the response is close to linearity and details are retained; when vec3 is large, the hair is more curly, so that high-crimp data is compressed into a reasonable range;

[0117] In summary, by introducing mathematical functions to increase the accuracy of the formula calculation results, a balance between sensitivity and robustness is achieved in crimp evaluation, capturing critical damage while avoiding overfitting noise. In addition, during the calculation process, the adjustment parameters and scaling factors are updated with the change of training;

[0118] Example:

[0119] Data 1:

[0120] Parameter settings: max = 10.0, min = 5.0, mean = 7.0, volatility = 2.0, skewness = 1.5, kurtosis = 2.5; ε1 = 0.1, ε2 = 0.2, ε3 = 0.3, σ = 0.5, vec1_th = 100; we have:

[0121] vec1 = 10 * 5 * 7 * 2 * 1.5 * 2.5 = 2625;

[0122] vec2 = [10 - 5, |7 - 2|, 1.5 + 2.5];

[0123] vec3 = [10 2 + 5 2 , 7 2 + 2 2 , 1.5 2 + 2.5 2 ;

[0124] Then crimp ≈ 1.49 + 1.62 + 1.48 = 4.59;

[0125] Data 2:

[0126] Parameter settings: max = 20.0, min = 5.0, mean = 7.0, volatility = 2.0, skewness = 1.5, kurtosis = 2.5; ε1 = 0.1, ε2 = 0.2, ε3 = 0.3, σ = 0.5, vec1_th = 100; There is:

[0127] vec1 = 20 * 5 * 7 * 2 * 1.5 * 2.5 = 5250;

[0128] vec2 = [20 - 5, |7 - 2|, 1.5 + 2.5];

[0129] vec3 = [20 2 + 5 2 , 7 2 + 2 2 , 1.5 2 + 2.5 2 ;

[0130] Then crimp ≈ 1.594 + 3.262 + 3.688 = 8.544;

[0131] Comparing Data 1 and Data 2, 4.59 < 8.544, the hair corresponding to Data 2 is curlier than the hair corresponding to Data 1;

[0132] Compare and analyze the curl evaluation coefficient with the standard curl threshold range:

[0133] When the curl evaluation coefficient is less than or equal to the minimum value of the standard curl threshold range, it is determined to be a straight hair type;

[0134] When the curl evaluation coefficient is within the standard curl threshold range, it is determined to be a wavy hair type;

[0135] When the curl evaluation coefficient is greater than or equal to the maximum value of the standard curl threshold range, it is determined to be a spiral hair type;

[0136] Fracture identification: Use Canny edge detection and Hough line transformation to identify fracture lines. When the length of the fracture line is less than the preset fracture threshold, it is marked as a fracture, and the fracture density is obtained;

[0137] Fracture density: The number of fracture pixels / area of the region, indicating the severity of fractures per unit area;

[0138] S2. After normalizing the characteristic information of the hair and synchronizing the time series, classify and cascade the characteristic information of the extracted operating data respectively, and complete fusion and dimensionality reduction. Use BLSTM to construct a fusion framework to identify the fusion features and obtain the dual-frequency response values of the corresponding sub-regions; among them, the dual-frequency response values include wind speed, rotation speed, and temperature;

[0139] Among them, the steps of classification and cascading include:

[0140] Construct a dual-branch heterogeneous encoder, including branch one and branch two;

[0141] Input the hair color, curl degree, and the characteristic information of the operating data into branch one, match the response control interval, and generate several modes based on the response control interval, including gentle wind mode, quick-drying mode, and hair care mode; among them, the response control interval includes the temperature interval and wind speed interval of the hair dryer;

[0142] Specifically, different hair colors have corresponding temperature or wind speed intervals, and different curl degrees also have corresponding temperature or wind speed intervals. Use machine learning algorithms (such as support vector machine, decision tree, K-nearest neighbor, etc.) to train the data, and map the input features (such as hair color, curl degree) and operating data to these preset response control intervals; during the process, the control strategy can also be optimized through a reward mechanism to enable the system to adjust the strategy according to feedback during operation to achieve goals such as gentle wind mode and quick-drying mode; among them, the gentle wind mode means gentle drying, medium temperature and medium wind speed; the quick-drying mode means efficient drying, high temperature and high wind speed; the hair care mode means low temperature protection, precise temperature and wind control; for example, high curl degree may correspond to the quick-drying mode with higher wind speed and temperature; the specific steps are not elaborated here;

[0143] In addition, during the actual use process, when adjusting the temperature and wind speed of the hair dryer, the adjustment span of the control interval corresponding to the quick-drying mode is greater than that of the control interval corresponding to the gentle wind mode, which is greater than that of the control interval corresponding to the hair care mode; the specific adjustment strategy is not elaborated here;

[0144] Input the fracture, fork, and the characteristic information of the operating data into branch two, build an evaluation calculation model, generate a hair damage index, divide several damage levels, and trigger different branch fusion strategies;

[0145] The calculation formula of the evaluation calculation model is:

[0146]

[0147] In the formula, H damage represents the hair damage index, F1 represents the fracture density, F2 represents the splitting density, and max(F1, F2) is used for normalization to avoid being dominated by a single factor (fracture density, splitting density). When the splitting area and fractures are in a common area, the hair damage index increases. ψ represents a constant term to prevent the denominator from being zero. α1, α2, and α3 all represent adjustment coefficients, and α1, α2, and α3 are all greater than 0;

[0148] Formula explanation: Quantify the linear superposition effect of the splitting density and the fracture density. The setting of the square root suppresses the overestimation in the high-density region and reflects the non-linear growth characteristic of the damage; Represents the synergistic effect between splitting and fracture and the amplification effect of the angle on the damage;

[0149] Example:

[0150] Parameter setting: F1 = 5, F2 = 3, var(θ) = 0.2, α1 = 0.3, α2 = 0.4, α3 = 0.3, ψ = 10 -6 ;

[0151] Then H damage ≈1.643+(3 / 5.000001)=1.832;

[0152] Compare and analyze the hair damage index H damage with the preset evaluation threshold interval [qj1, qj2]:

[0153] When H damage < qj1, it is determined that the hair damage is small, and a low-risk signal is generated;

[0154] When qj1 ≤ H damage ≤ qj2, it is determined that the hair damage is moderate, and a medium-risk signal is generated;

[0155] When H damage > qj2, it is determined that the hair damage is severe, and a high-risk signal is generated;

[0156] Trigger different branch fusion strategies based on different risk signals:

[0157] In response to the low-risk signal, trigger branch fusion strategy one and perform gated residual fusion, and its representation form is:

[0158]

[0159] In the formula, Z fuseddenotes the fused feature vector, g denotes the gating vector, ch1 denotes branch one, ch2 denotes branch two, ⊙ denotes the matrix dot product operation, denotes the element-wise addition operation, W r denotes the weight matrix;

[0160] Formula features: Residual connection ensures low computational overhead, the gating vector g suppresses noise, and ReLU(W r ·) enhances non-linearity;

[0161] In response to the risk signal, trigger branch fusion strategy two, perform bidirectional attention fusion, and its representation form is:

[0162]

[0163] In the formula, A 1→2 denotes the attention of branch one to branch two, A 2→1 denotes the attention of branch two to branch one, sigmoid(·) denotes the activation function;

[0164] Specifically, dark hair colors can better hide the forking features, light hair colors can clearly reveal the forking features, high curl degrees will amplify the fracture features, and low curl degrees will reduce the fracture features. Therefore, there is a correlation between the two, and it is necessary to construct the attention between the two;

[0165]

[0166] In the formula, W inquire denotes the query matrix, W key denotes the key matrix, (·) T denotes the transpose matrix, d denotes the encoding length of the feature vector, and when the dimensions are inconsistent, the common dimension is selected;

[0167] In response to the high-risk signal, trigger branch fusion strategy three, perform multi-scale gating fusion, and its representation form is: ; In the formula, s denotes the scale. When the value is 1, it represents the original scale, Scale s (ch1,ch2) denotes the multi-scale transformation calculation. For example: perform a 3*3 convolution on branch one and a dilated convolution on branch two. H(·) denotes the entropy statistic of the multi-scale feature pyramid, and Sparsemax(·) denotes forced sparsity;

[0168] By establishing a BLSTM fusion framework, the "fusion framework" here generally refers to the fusion of multiple features or data sources, and uses the context information of the input sequence (i.e., bidirectional propagation) for recognition and prediction; BLSTM is a special type of long short-term memory network, and its key advantage is the ability to capture long-term dependencies, understand the mutual influence between the front and back parts of the input sequence, effectively fuse the feature information of the head with the feature model of the running data, thereby improving the recognition ability of the model and more accurately judging the blowing mode and dual-frequency response value of the hair dryer;

[0169] In addition, a classifier is built into the dual-frequency control model. The fused features are input into the classifier. The classification space of the classifier is represented as the dual-frequency response value of the hair dryer, and the dual-frequency response value is used as the sample label of the training samples of the dual-frequency control model;

[0170] To sum up, by taking fracture and bifurcation as the calculation premise, an evaluation calculation model is designed to divide the damage level of the hair, so as to design a corresponding fusion strategy, which is convenient for feature fusion on Branch 1 and Branch 2. By using hair color and curl as auxiliary calculations, the training accuracy and calculation speed of the evaluation calculation model are improved, and the relationship of mutual influence between Branch 1 and Branch 2 is established;

[0171] S3. Calculate the efficiency and power ratio of the corresponding hair dryer at time t based on the dual-frequency response value, fit to obtain the efficiency function and power ratio function, calculate the optimization function, optimize the dual-frequency control model, and input the collected head video data into the optimized dual-frequency control model again to output the dual-frequency response value;

[0172] Among them, the calculation formula of the efficiency function is:

[0173] In the formula, η(t) represents the hair dryer efficiency obtained at time t, b in 、b out both represent power constants, which are related to the characteristics of the hair dryer and the motor. wind(t) represents the wind speed obtained at time t, temp(t) represents the temperature obtained at time t, and speed(t) represents the rotation speed obtained at time t;

[0174] Mark the obtained efficiency function as: η(wind, speed, temp) = gn1;

[0175] It should be noted that the fitting process uses piecewise polynomial regression to adapt to different blowing modes. For example, when it is the soft wind mode, the least squares method is used to fit the polynomial regression model. Its advantage is that it can capture the non-linear relationship between variables; when it is the quick-dry mode, the exponential function is used for fitting, and when it is the hair care mode, the logarithmic function is used for fitting. The specific fitting process will not be elaborated here;

[0176] Among them, the calculation formula of the power ratio function is:

[0177]

[0178] In the formula, PowerRate(t) represents the power ratio obtained at time t;

[0179] Mark the power ratio function obtained by fitting as: PowerRate(wind, speed, temp) = gn2;

[0180] The support vector regression (SVR) is used in the fitting process. Its advantage lies in being applicable to high-dimensional data and being able to handle non-linear relationships. It maps the input space to a high-dimensional feature space to find the optimal regression model; the specific fitting process will not be elaborated here;

[0181] Then the calculation formula of the optimization function is:

[0182]

[0183] In the formula, O(t) represents the optimization function at time t, which is used to balance the maximization of efficiency and the minimization of power. κ1, κ2, and κ3 all represent weight coefficients, and κ1, κ2, and κ3 are all greater than 0. (κ3 * H damage ) represents the penalty term. The larger the hair damage index, the larger the penalty term;

[0184] Logical explanation: Minimize O(t), that is, pursue high efficiency, low power, and low damage at the same time; adjust the working parameters of the hair dryer device, such as (heating temperature, wind speed), through gradient descent to make g1 increase, g2 decrease, and H damage decrease, complete the control work of the hair dryer on the premise of realizing high-speed control, and realize the optimization and update of the dual-frequency control model;

[0185] Example:

[0186] Parameter setting: gn1 = 10, gn2 = 2, H damage = 2, κ1 = 0.5, κ2 = 0.3, κ3 = 0.4;

[0187] Then O(t) = 0.05 + 0.6 + 0.8 = 1.45;

[0188] Assume that the optimization function is to minimize O(t), and take the partial derivative of each parameter:

[0189] Use the gradient descent formula:

[0190] In the formula, represents a certain updated weight parameter, κm denotes the initial weight parameter, where m takes values of 1, 2, 3, and ▽ denotes the learning rate;

[0191] Parameter setting: Set the initial learning rate to 0.01, then we have:

[0192]

[0193] In summary, by balancing efficiency, power, and damage through itemized weights, gradient descent can automatically find the optimal parameter combination, regularly calibrate κ1, κ2, κ3 to ensure weight matching, and precisely adjust the working state of the hair dryer while ensuring high-speed operation.

[0194] Embodiment 2

[0195] The embodiment of the present invention provides a control system for a high-speed dual-frequency hair dryer; Figure 3 is a schematic module diagram of the control system of a high-speed dual-frequency hair dryer shown according to an exemplary embodiment; please refer to Figure 3 , the system includes: a data acquisition module, a feature extraction module, a classification and fusion module, and a model optimization module, and the data acquisition module, the feature extraction module, the classification and fusion module, and the model optimization module are communicatively connected;

[0196] The data acquisition module acquires the head video data and operation data of the high-speed dual-frequency hair dryer, and numbers the head video data according to the scanning order;

[0197] The feature extraction module uses image processing technology to extract the feature information of the hair; among them, the feature information includes breakage, split ends, hair color, and curl;

[0198] The classification and fusion module performs normalization processing and time series synchronization on the feature information of the hair, and then classifies and cascades the feature information of the extracted operation data respectively, completes fusion and dimensionality reduction, uses BLSTM to construct a fusion framework to identify the fusion features, obtains a dual-frequency control model, and outputs the dual-frequency response values of the corresponding sub-regions; among them, the dual-frequency response values include rotational speed, temperature, and wind speed;

[0199] The model optimization module calculates the efficiency and power ratio of the corresponding high-speed dual-frequency hair dryer at time t based on the dual-frequency response values, fits to obtain an efficiency function and a power ratio function, calculates an optimization function, and uses the gradient descent method to optimize the dual-frequency control model.

[0200] In the application, several formulas involved are calculated by taking their numerical values after dimensionless, and the formula is a formula obtained by software simulation of collecting a large amount of data to approximate the real situation, and the in the formula is set by those skilled in the art according to the actual situation.

[0201] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. Those of ordinary skill in the art will appreciate that the units and algorithm steps of the examples described in connection with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in hardware or software depends on the specific application and design constraints of the technical solution.

[0202] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units. They may be located in one place or distributed over multiple network units. Some or all of the units can be selected according to actual needs to achieve the objectives of the solution of this embodiment.

[0203] As described above, the above is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art within the technical scope disclosed in this application can easily think of changes or substitutions, which should all be covered within the protection scope of this application.

Claims

1. A control method for a high-speed dual-frequency hair dryer, characterized in that: The following steps are involved: Collect the head video data and operation data of the high-speed dual-frequency hair dryer, number the head video data according to the scanning sequence, and use image processing technology to extract the characteristic information of the hair; The characteristic information at least includes breakage, split ends, hair color and curl; After the hair feature information is normalized and time-series synchronized, it is classified and concatenated with the feature information of the extracted running data, and fusion and dimension reduction are completed. The fusion framework is constructed using BLSTM to identify the fusion features, obtain the dual-frequency control model, and output the dual-frequency response values ​​of the corresponding sub-areas. Wherein, the dual-frequency response value includes at least rotation speed, temperature and wind speed; Based on the dual-frequency response value, the efficiency and power ratio of the high-speed dual-frequency hair dryer at time t are calculated, and the efficiency function and power ratio function are fitted to obtain the optimization function. The optimization function is imported into the dual-frequency control model for training and optimization, and the dual-frequency control model is optimized using the gradient descent method.

2. The control method of a high-speed dual-frequency hair dryer according to claim 1, characterized in that: The steps of extracting characteristic information of hair using image processing technology include: Extract key frames: perform frame difference processing on the sequentially numbered head video data, divide the head area into several sub-areas, and use the SSIM technology to analyze the processed results in each sub-area to obtain several key frame images; Filtering: Use wavelet basis function to filter the key frame image; Hair color recognition: extract the color components of the filtered image in each RGB channel, perform weighted average of the color components to calculate the initial brightness, introduce the indoor light intensity to correct the initial brightness, and construct a color space in each sub-area to identify the hair color; Bifurcation identification: Use the Zhang-Suen thinning algorithm to determine whether there is a Y-shaped intersection in each sub-area. If so, mark it as a bifurcation, and obtain the bifurcation point density and bifurcation angle variance; Curl recognition: Based on the local curvature of each sub-area, the maximum, minimum, average, fluctuation, skewness and kurtosis of the local curvature are counted to calculate the curl evaluation coefficient, and the curl type is determined in combination with the standard curl threshold; the curl type includes straight hair type, wavy type and spiral type; Fracture identification: Use Canny edge detection and Hough line transform to identify the fracture lines in each sub-area. When the length of the fracture line is less than the preset fracture threshold, it is marked as a fracture and the fracture density is obtained.

3. The control method of a high-speed dual-frequency hair dryer according to claim 2, characterized in that: The steps for calculating the curl assessment coefficient include: Get the local curvature of each sub-area, and construct a six-dimensional monitoring vector [max, min, mean, volatility, skewness, kurtosis] with the maximum, minimum, average, volatility, skewness, and kurtosis of the local curvature, and construct judgment vectors vec1, vec2, and vec3 according to different combinations; Combine vec1, vec2, and vec3 to set the formula to calculate the curl evaluation coefficient: Where crimp represents the curl assessment coefficient, vec1_th represents the baseline threshold, σ is the scaling factor, ε1, ε2, and ε3 all represent adjustment factors, and σ, ε1, ε2, and ε3 are all greater than 0.

4. The control method of a high-speed dual-frequency hair dryer according to claim 1, characterized in that: The steps of the classification cascade include: Construct a dual-branch heterogeneous encoder, including branch one and branch two; Input the characteristic information of hair color, curl and operation data into branch one, match the response control interval, and generate several modes based on the response control interval, including soft wind mode, quick drying mode and hair care mode; wherein the response control interval includes the temperature interval and wind speed interval of the hair dryer; At the same time, the characteristic information of breakage, bifurcation and operation data is input into branch 2 to build an evaluation calculation model to generate a hair damage index to divide it into several damage levels and trigger different branch fusion strategies; If branch fusion strategy 1 is triggered, gated residual fusion is performed; If branch fusion strategy 2 is triggered, bidirectional attention fusion is performed; If branch fusion strategy three is triggered, multi-scale gated fusion is performed.

5. The control method of a high-speed dual-frequency hair dryer according to claim 4, characterized in that: The calculation formula for the evaluation calculation model is: In the formula, H damage represents the hair damage index, F1 represents the breakage density, F2 represents the bifurcation density, ψ represents the constant term, α1, α2, α3 represent the adjustment coefficients, and α1, α2, α3 are all greater than 0.

6. The control method of a high-speed dual-frequency hair dryer according to claim 1, characterized in that: The efficiency function is based on the formula: Where η(t) represents the hair dryer efficiency at time t, b in , b out All represent constants related to the characteristics of the hair dryer and the motor, wind(t) represents the wind speed obtained at time t, temp(t) represents the temperature obtained at time t, speed(t) represents the speed obtained at time t; The efficiency function obtained by fitting is marked as: η(wind, speed, temp)=gn1; The power proportional function is calculated as: Where PowerRate(t) represents the power ratio obtained at time t; The power ratio function obtained by fitting is marked as: PowerRate(wind, speed, temp)=gn2.

7. A control method for a high-speed dual-frequency hair dryer according to claim 6, characterized in that: The calculation formula of the optimization function is: Where O(t) represents the optimization function at time t, κ1, κ2, and κ3 represent weight coefficients, and κ1, κ2, and κ3 are all greater than 0, (κ3*H damage ) represents a penalty term.

8. A control system for a high-speed dual-frequency hair dryer, characterized in that: include: The data acquisition module collects the head video data and operation data of the high-speed dual-frequency hair dryer, and numbers the head video data according to the scanning sequence; The feature extraction module uses image processing technology to extract the characteristic information of hair, including breakage, bifurcation, hair color and curl; The classification fusion module normalizes and synchronizes the characteristic information of the hair, and then classifies and concatenates it with the characteristic information of the extracted running data, completes fusion and dimensionality reduction, uses BLSTM to build a fusion framework to identify the fusion features, obtains the dual-frequency control model, and outputs the dual-frequency response values ​​of the corresponding sub-areas; the dual-frequency response values ​​include rotation speed, temperature, and wind speed; The model optimization module calculates the efficiency and power ratio of the high-speed dual-frequency hair dryer at time t based on the dual-frequency response value, and fits the efficiency function and power ratio function to calculate the optimization function, and uses the gradient descent method to optimize the dual-frequency control model.

Images