Ultrasonic human sensing method and device for air conditioner, air conditioner and storage medium

Abstract

The invention relates to the technical field of intelligent household appliances, and discloses an ultrasonic human sensing method and device for an air conditioner, the air conditioner and a storage medium. The method comprises the following steps: receiving an echo signal of a signal transmitted by an ultrasonic sensor so as to extract spatial characteristics and dynamic characteristics in the echo signal; fusing the extracted spatial features and dynamic features; the fused features are input into a lightweight hybrid neural network model to obtain a classification result, and operation parameters of the air conditioner are adjusted according to the classification result; wherein the classification result comprises the perceived user position and the perceived user behavior state. According to the method, the lightweight hybrid neural network model can respond quickly under the condition of ensuring the accuracy of user perception. Therefore, the real-time control requirement of the user is met and the user experience is improved.

Description

Technical Field

[0001] This application relates to the technical field of smart home appliances, for example, to an ultrasonic sensing method and device for an air conditioner, an air conditioner, and a computer-readable storage medium. Background Art

[0002] With the development of air conditioner technology, consumers' demand for air conditioners has gradually increased, thus giving rise to the development of various air conditioner sensing and blowing systems. However, the existing air conditioner sensing and blowing systems have problems such as insufficient accuracy in sensing the user state and poor adaptability to complex scenarios.

[0003] The related art discloses a wireless sensing human behavior detection system integrating software and hardware, which relates to the technical field of human behavior detection. The system includes a data collection unit for collecting motion data of a human body when generating a motion behavior by using an ultrasonic device and converting the motion data into an ultrasonic signal for storage in a preset database; a data preprocessing unit for performing preprocessing operations on the ultrasonic signal by using a sliding window algorithm and wavelet transform technology and extracting the amplitude change of the human behavior action as a data set based on the operation result; a feature extraction unit for building a training neural network based on the data set and extracting action features based on the neural network to obtain the human behavior action category; and an action recognition extraction unit for obtaining real-time human behavior data, preprocessing the behavior data, and inputting it into the neural network to identify the human behavior action.

[0004] In the process of implementing the embodiments of the present disclosure, it is found that at least the following problems exist in the related art: The related art has serious delays in complex scenarios and cannot meet the real-time control requirements.

[0005] It should be noted that the information disclosed in the above background art section is only used to enhance the understanding of the background of this application, and thus may include information that does not constitute the prior art known to those of ordinary skill in the art. Summary of the Invention

[0006] To have a basic understanding of some aspects of the disclosed embodiments, a simple summary is given below. This summary is not a general review, nor is it intended to identify key / important elements or delineate the protection scope of these embodiments, but rather serves as a preamble to the subsequent detailed description.

[0007] The embodiments of the present disclosure provide an ultrasonic sensing method and device for an air conditioner, an air conditioner, and a computer-readable storage medium to reduce response delay and meet the real-time control requirements.

[0008] In some embodiments, the method includes: receiving an echo signal of an ultrasonic sensor transmitting signal to extract spatial features and dynamic features in the echo signal; fusing the extracted spatial features and dynamic features; inputting the fused features into a lightweight hybrid neural network model to obtain a classification result, and adjusting the operating parameters of the air conditioner according to the classification result; wherein the classification result includes the perceived user position and user behavior state.

[0009] In some embodiments, the device includes: a processor and a memory storing program instructions, and the processor is configured to execute the ultrasonic sensing method for an air conditioner as described above when running the program instructions.

[0010] In some embodiments, the air conditioner includes: an air conditioner body provided with a plurality of ultrasonic sensor transceiver arrays; and the ultrasonic sensing device for an air conditioner as described above, installed on the air conditioner body.

[0011] In some embodiments, the computer-readable storage medium stores program instructions, and when the program instructions are running, they are used to cause a computer to execute the ultrasonic sensing method for an air conditioner as described above.

[0012] The ultrasonic sensing method and device for an air conditioner, the air conditioner, and the computer-readable storage medium provided by the embodiments of the present disclosure can achieve the following technical effects: The user is sensed through an ultrasonic sensor array to obtain feature information at multiple angles and positions. The lightweight hybrid neural network model is used to fuse and classify the features to obtain information on the user position and behavior state. Then, the operating parameters of the air conditioner are controlled based on the obtained user position and behavior state. In this way, while ensuring the accuracy of user perception, the lightweight hybrid neural network model can respond quickly. Thus, it meets the real-time control requirements of users and improves the user experience.

[0013] The above general description and the following description are only exemplary and explanatory, and are not used to limit the present application. Description of the Drawings

[0014] One or more embodiments are exemplarily illustrated by corresponding drawings. These exemplary illustrations and the drawings do not constitute limitations on the embodiments. Elements with the same reference numerals in the drawings are shown as similar elements. The drawings do not constitute a proportional limitation, and among them: Figure 1 is a schematic diagram of an ultrasonic sensing method for an air conditioner provided by an embodiment of the present disclosure; Figure 2 is a schematic diagram of another ultrasonic sensing method for an air conditioner provided by an embodiment of the present disclosure; Figure 3It is a schematic diagram for determining the current noise baseline in the method provided by an embodiment of the present disclosure; Figure 4 It is a schematic diagram of another ultrasonic sensing method for an air conditioner provided by an embodiment of the present disclosure; Figure 5 It is a schematic diagram of an ultrasonic sensing device for an air conditioner provided by an embodiment of the present disclosure; Figure 6 It is a schematic diagram of an air conditioner provided by an embodiment of the present disclosure.

[0015] Reference numerals: 100: Ultrasonic sensing device for an air conditioner; 101: Processor; 102: Memory; 103: Communication interface; 104: Bus; 200: Air conditioner. Detailed implementation manners

[0016] In order to be able to understand the features and technical content of the embodiments of the present disclosure in more detail, the implementation of the embodiments of the present disclosure will be described in detail below with reference to the accompanying drawings. The accompanying drawings are only for reference and explanation purposes and are not used to limit the embodiments of the present disclosure. In the following technical description, for the convenience of explanation, sufficient understanding of the disclosed embodiments is provided through multiple details. However, one or more embodiments can still be implemented without these details. In other cases, well-known structures and devices can be shown in a simplified manner.

[0017] In the description of the embodiments of the present disclosure, terms such as "first" and "second" in the specification, claims and the above-mentioned drawings are used to distinguish similar objects and do not necessarily need to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so as to implement the embodiments of the present disclosure described here. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion.

[0018] Unless otherwise specified, the term "plurality" means two or more.

[0019] In the embodiments of the present disclosure, the character " / " indicates that the objects before and after are in an "or" relationship. For example, A / B means: A or B.

[0020] The term "and / or" is a description of the association relationship of an object and indicates that three relationships can exist. For example, A and / or B means: A or B, or, the three relationships of A and B.

[0021] The term "corresponding" can refer to an association relationship or a binding relationship. A corresponding to B means that there is an association relationship or a binding relationship between A and B.

[0022] In the embodiments of the present disclosure, an intelligent household appliance device refers to a household appliance product formed by introducing microprocessors, sensor technology, and network communication technology into household appliance devices, and has the characteristics of intelligent control, intelligent perception, and intelligent application. The operation process of intelligent household appliance devices often relies on the application and processing of modern technologies such as the Internet of Things, the Internet, and electronic chips. For example, an intelligent household appliance device can be connected to an electronic device to achieve remote control and management of the intelligent household appliance device by the user.

[0023] In the embodiments of the present disclosure, multiple ultrasonic sensor transceiver arrays can be arranged only on the air conditioner, or can be arranged on the air conditioner and other intelligent household appliance devices to form a cross-device collaborative array. When multiple ultrasonic sensors are all arranged on the air conditioner, the multiple ultrasonic sensors are deployed as distributed transceiver units to form a multi-directional coverage of the detection range. Exemplarily, a set of transmitting arrays and two sets of receiving arrays are arranged on the top of the air conditioner; a set of transmitting arrays and a set of receiving arrays are arranged on both the left and right sides of the air conditioner. In this way, a detection range of 180° in the horizontal direction and 90° in the vertical direction is formed.

[0024] When multiple ultrasonic sensors are arranged in a cross-device collaborative manner, the air conditioner and other intelligent household appliance devices (such as intelligent speakers, floor cleaning robots, etc.) form a master-slave architecture. Among them, the air conditioner is the master device and periodically sends ultrasonic signals. Other intelligent household appliance devices are slave devices, which transmit / receive signals at preset intervals and send the data back to the master device for fusion processing.

[0025] Optionally, the ultrasonic sensor array includes a transmitting unit and a receiving unit. Among them, the transmitting unit is an ultrasonic speaker array, and the receiving unit is a microphone array. Multiple speakers form a beamforming structure to adjust the transmission angle to cover the target area. And the working frequency range is in the 40 kHz to 48 kHz frequency band, and the transmission power meets the effective detection distance of more than 10 meters and supports pulse modulation. The microphone array can adopt a microphone array of 1×2 or more (such as a 2-element linear array), and the element spacing ≤ λ / 2, where λ is the ultrasonic wavelength. Exemplarily, the wavelength corresponding to 40 kHz is 8.5 mm, and the spacing is taken as 5 to 8 mm. It supports direction-of-arrival estimation and signal phase difference calculation.

[0026] Optionally, the microphone array can reuse an intelligent household appliance device with a voice function. The microphone array distinguishes voice signals (20 Hz to 20 kHz) from ultrasonic signals (>20 kHz) through frequency division multiplexing technology, and separates ultrasonic echoes at the receiving end through a band-pass filter (passband 30 kHz to 50 kHz).

[0027] Optionally, when multiple ultrasonic sensors are all arranged in the air conditioner, the multiple transmitting units use time-division pulse transmission, and the receiving unit collects the echo signals corresponding to each time slot, and distinguishes the signal sources through timestamps. Exemplarily, the transmission period T = 10 ms, the transmission duration t of each array is 1 ms, and an interval of 1 ms is used to avoid overlap. When multiple ultrasonic sensors are cooperatively arranged across devices, the master device broadcasts a time slot allocation table during initialization. For example, device A uses time slots 1 to 3, device B uses time slots 4 to 6, and each time slot is 500 μs. The transmission frequency intervals of different devices are ≥2 kHz (for example, device A uses 40 kHz, device B uses 42 kHz, and device C uses 44 kHz), and the receiving unit extracts the echo signals of the corresponding frequencies through filters.

[0028] Combined with Figure 1 As shown, an ultrasonic sensing method for an air conditioner provided by an embodiment of the present disclosure includes: S101, the processor receives the echo signal of the ultrasonic sensor transmitting signal to extract the spatial feature and dynamic feature in the echo signal.

[0029] S102, the processor fuses the extracted spatial feature and dynamic feature.

[0030] S103, the processor inputs the fused features into a lightweight hybrid neural network model to obtain a classification result, and adjusts the operating parameters of the air conditioner according to the classification result; wherein, the classification result includes the perceived user position and user behavior state.

[0031] Here, during ultrasonic sensor sensing detection, an ultrasonic signal is transmitted and an echo signal is received (wherein, the echo signal may include continuous echo signals of multiple detection periods). After processing the echo signal (such as filtering and noise reduction, etc.), the spatial feature and dynamic feature therein are extracted. Because the multi-array ultrasonic sensors can obtain echo signals from multiple angles and distances, the position and behavior state information of the user can be obtained more accurately. Among them, the spatial feature refers to the feature in the spatial dimension, including but not limited to features such as signal strength, direction of arrival, and obstacle contour. The dynamic feature refers to the feature in the time dimension, including but not limited to features such as periodic change of the signal, human body movement speed, and acceleration.

[0032] Subsequently, the extracted spatial features and dynamic features are fused. Since the dynamic features are time-series vectors without spatial dimensions, they need to be converted into a format compatible with the spatial features for dimension alignment. After dimension alignment, concatenation is performed based on the channel dimension. Specifically, the dimension of the spatial features is C×H×W, where C is the number of channels, and H and W are the height and width of the spatial feature map (the spatial feature map is a collection of spatial features), respectively. The dimension of the dynamic features is T×D, where T is the number of time steps and D is the feature vector dimension. When fusing the dynamic features of a certain time series, the feature vector dimension D corresponding to the target time step is copied to the H×W grid to generate a D×H×W tensor. At this time, the spatial features and dynamic features are dimensionally aligned, having the same spatial dimension H×W, only with different numbers of channels. Further, the two feature maps are concatenated along the channel dimension to generate a fused feature map of (C + D)×H×W. Among them, the first C channels retain the spatial features, and the last D channels retain the dynamic features. The length of the feature vector at each spatial position (h i , w j ) is C + D. h i constitutes H, w j constitutes W, and i represents the i-th spatial position.

[0033] The fused features are input into a lightweight hybrid neural network model to obtain classification results. Among them, the classification results include the perceived user location and the user behavior state. The lightweight hybrid neural network model is obtained by training based on a dataset, and the dataset includes echo data in different scenarios and labels of user locations and behavior states matching the data. In this way, the hybrid neural network model is trained using the dataset with labels, and the loss function is optimized to obtain the trained lightweight hybrid neural network model. In this way, the user location and the user behavior state are obtained using the trained lightweight hybrid neural network model. The air conditioner can adjust the operating parameters based on the user location and the user behavior state so that the air outlet requirements of the air conditioner meet the user's needs. Exemplarily, parameters such as the air outlet wind speed and the air outlet direction of the air conditioner are adjusted.

[0034] In addition, the lightweight hybrid neural network model includes a convolutional layer, a gated recurrent unit layer, and a feature fusion layer. Among them, the feature fusion layer fuses the spatial features output by the convolutional layer and the dynamic features output by the gated recurrent unit layer, and outputs the classification results through a fully connected layer. To improve the response speed and reduce the computational complexity, the hybrid neural network model is optimized through pruning and quantization techniques to obtain a lightweight model.

[0035] Exemplarily, the pruning technique can be to prune the unimportant convolutional kernel channels in the convolutional layer according to the channel importance (such as the absolute value of the weight based on the L1 norm), and the pruning ratio is 30% - 50%. It can also be to sparsify the recurrent weight matrix of the gated recurrent unit layer, retain the top K weights with the largest absolute values (such as K = 20%), and set the remaining weights to zero. The quantization technique can be to quantize the model parameters (such as weights and activation values) from 32-bit floating-point type to 8-bit fixed-point type. In this way, while ensuring the model performance, the model is made as lightweight as possible, thereby improving the response speed and real-time performance.

[0036] By using the ultrasonic sensing method for air conditioners provided in the embodiments of the present disclosure, the ultrasonic sensor array is used to sense the user to obtain feature information from multiple angles and positions. The lightweight hybrid neural network model is used to fuse and classify the features to obtain information about the user's position and behavior state. Then, based on the obtained user position and behavior state, the operating parameters of the air conditioner are controlled. In this way, while ensuring the accuracy of user perception, the lightweight hybrid neural network model can respond quickly. Thus, it meets the user's real-time control requirements and improves the user experience.

[0037] Optionally, in step S101, the processor extracts the spatial features and dynamic features from the echo signal, including: The processor uses the convolutional layer of the lightweight hybrid neural network model to extract the spatial features from the echo signal received by each ultrasonic sensor; the spatial features include the signal intensity distribution and the direction of arrival.

[0038] The processor uses the gated recurrent unit layer of the lightweight hybrid neural network model to extract the dynamic features from the echo signals continuously received by each ultrasonic sensor; the dynamic features include the user's movement speed and acceleration.

[0039] Here, the convolutional layer can better process data with spatial structures to extract local features. Therefore, the convolutional layer is used to extract the spatial features from the echo signal. Among them, the spatial features mainly include the signal intensity distribution and the direction of arrival. The signal intensity distribution can be used to determine whether there is someone and the user's behavior state; for example, when there is someone in the target space, the signal becomes stronger. The signal also increases when the user is in a moving state. The direction of arrival can be used for user positioning and trajectory tracking.

[0040] The gated recurrent unit layer is suitable for processing sequential data and can capture the signal change rules of time-dependent relationships. Therefore, the gated recurrent unit is used here to perform temporal modeling on the continuous echo signals to extract dynamic features. Among them, the dynamic features include the user's movement speed and acceleration. Specifically, the user's movement speed and acceleration are calculated using the Doppler frequency shift in the echo signal of the gated recurrent unit. In this way, the structure of the hybrid neural network model is relatively simple, the robustness of the model is better, and the calculation efficiency is also relatively high.

[0041] Optionally, in step S102, the processor fuses the extracted spatial features and dynamic features, including: The processor splices and fuses the spatial feature map representing the spatial features and the dynamic features to obtain a spliced feature map.

[0042] Based on the attention mechanism, the processor calculates the spatial attention weight and the temporal attention weight for the spliced feature map.

[0043] Based on the spatial attention weight and the temporal attention weight, the processor obtains the fused features through weighted summation.

[0044] Here, the fusion splicing of the spatial feature map output by the convolutional layer and the dynamic features output by the gated recurrent unit layer is as described above and will not be elaborated here. To improve the perception ability of the lightweight hybrid neural network model for key regions (such as the signal features of the user's location) and key time points (such as the time sequence segments of the user's actions), the attention mechanism is introduced. The spatial attention weight and the temporal attention weight are calculated respectively, and the fused features are obtained by weighted summation of the spliced feature map based on the weights.

[0045] Specifically, global average pooling is performed on the spatial feature part in the spliced feature map to generate a channel description vector. Based on the channel description vector, a spatial attention map is generated, and the spatial attention map is a set of spatial attention weights. Similarly, the mean value of the dynamic features in the spliced feature map is calculated to generate a temporal description vector. Based on the temporal description vector, a temporal attention vector is generated. The temporal attention vector is a set of temporal attention weights.

[0046] Exemplarily, the spatial attention weight a s =σ(W2×ReLU(W1×z c ))), where W1∈R C / r×C , W2∈R H×W×C / r (r is the compression ratio, such as 16), z c is the channel description vector, z c ∈R C . Then a s ∈R H×W , and each position weight value is between [0,1]. ReLU is the activation function. The temporal attention vector a t =Softmax(W4×ReLU(W3×z t ))), W3∈R D / r×D , W4∈R T×D / r ; z t is the temporal description vector, z t ∈R D . Finally, a t ∈RT , the sum of weights is 1. R is a vector space. Thus, it helps to improve the accuracy of the classification result.

[0047] Optionally, the lightweight hybrid neural network model includes a depthwise separable convolutional layer and a simplified gated recurrent unit layer. Among them, the depthwise separable convolutional layer reduces the computational amount from DK 2 ×M×N×DF 2 to DK 2 ×M×DF + M×N×DF 2 (DK is the convolutional kernel size, M / N is the number of input / output channels, and DF is the feature map size), and changes the fully connected layer of the depthwise separable convolutional layer to global average pooling. The simplified gated recurrent unit layer means sharing the parameters of the update gate and the reset gate and reducing the number of weights of the recurrent connections. Thus, the structure of the hybrid neural network model is optimized and the computational amount of the model is reduced. At the same time, the model introduces an attention mechanism when fusing the extracted spatial features and dynamic features to ensure the model's ability to capture features.

[0048] In addition, after the structure of the above hybrid neural network model is optimized, quantization technology and / or pruning technology can be used to compress the model to obtain a lightweight hybrid neural network model. The compression of the model is detailed above.

[0049] Combined Figure 2 as shown, another ultrasonic sensing method for an air conditioner provided by an embodiment of the present disclosure includes: S101, the processor receives the echo signal of the ultrasonic sensor transmitting signal to extract the spatial features and dynamic features in the echo signal.

[0050] S121, the processor determines the current noise baseline to obtain an initial threshold.

[0051] S122, when the signal intensity in the extracted spatial features is greater than or equal to the initial threshold, the processor preliminarily determines the user situation in the target space.

[0052] S123, when there is a user in the target space, the processor fuses the extracted spatial features and dynamic features.

[0053] S103, the processor inputs the fused features into the lightweight hybrid neural network model to obtain a classification result, and adjusts the operating parameters of the air conditioner according to the classification result; wherein, the classification result includes the perceived user position and user behavior state.

[0054] Here, after extracting the spatial features and dynamic features in the echo signal, it is possible to initially determine whether there is a user in the target space based on the spatial features. If there is a user, the extracted features are further fused and classified to obtain a classification result. Specifically, the current noise baseline is determined (that is, the reference value of the current environmental background noise signal, which can be obtained by continuously collecting echo signals when there is no one in the target space and calculating the statistical features of the continuously collected echo signals to calculate the noise baseline). According to the current noise baseline, an initial threshold is calculated. Among them, the initial threshold is positively correlated with the current noise baseline.

[0055] Compare the signal strength in the spatial features extracted based on the echo signal with the initial threshold. If the signal strength in the extracted spatial features is greater than the initial threshold, it indicates that there is a user in the target space. If the signal strength in the extracted spatial features is less than or equal to the initial threshold, it indicates that there is no user in the target space. Optionally, compare the signal strength and the initial threshold in multiple detection cycles. When the signal strength in multiple detection cycles is greater than the initial threshold, it is determined that there is a user in the target space. In this way, sudden noise-induced momentary high noise is avoided, preventing misjudgment. In this way, after detecting a user, the analysis of the user's position and user behavior state is carried out. This helps to reduce computing power and save the memory occupancy of edge devices.

[0056] Combined with Figure 3 As shown, optionally, S121, the processor determines the current noise baseline including: S1211, the processor collects background noise signals when there is no user in the target space.

[0057] S1212, the processor calculates the statistical features of the background noise signals and establishes an initial noise baseline model.

[0058] S1213, the processor uses the statistical features of the initial noise baseline model as the current noise baseline.

[0059] Here, during the process of no user in the target space, background noise signals are collected through an ultrasonic sensor (that is, the echo signals of the ultrasonic emission signals are collected. At this time, since there is no user interference, the echo signals can be used as background noise signals). To ensure the accuracy of the initial noise baseline model, background noise signals for a certain period or different time periods can be taken. Calculate the statistical features of this part of the background noise signals, such as the mean and standard deviation, etc. Thus, an initial noise baseline model is established, and the initial noise baseline model includes statistical features. Furthermore, the statistical features of the initial noise baseline model are used as the current noise baseline. In this way, based on the current noise baseline, combined with the detection sensitivity requirements, an initial threshold is set. To initially determine whether there is a user.

[0060] Optionally, after step S1212 where the processor establishes the initial noise baseline model, the following steps are further included: The processor uses a sliding window mechanism to update the background noise signal in real time.

[0061] The processor calculates the statistical features within the updated sliding window to dynamically update the initial noise baseline model.

[0062] Here, considering the changes or interferences of the background noise, to ensure the robustness of the model in complex environments, the initial noise baseline model is dynamically updated. Specifically, the sliding window mechanism is used to update the background noise signal in real time. That is, whenever it is detected that there is no user in the target space, the sliding window can be updated so that the statistical features of the latest background noise signal are stored in the sliding window. In this way, the initial noise baseline model is dynamically updated using the sliding window mechanism.

[0063] In addition, the size of the sliding window can be set based on the corresponding loudness and stability requirements. Exemplarily, the sliding window size w = 100 is applicable to second-level changes. The sliding window size w = 500 is applicable to minute-level changes.

[0064] Optionally, after step S1213 where the processor uses the statistical features of the initial noise baseline model as the current noise baseline, the following steps are further included: The processor uses a sliding window mechanism to update the background noise signal in real time.

[0065] The processor calculates the statistical features within the updated sliding window to dynamically update the current noise baseline.

[0066] When the dynamically updated current noise baseline and the initial noise baseline (i.e., the statistical feature values in the initial noise baseline model) satisfy the trigger condition, the processor dynamically updates the initial noise baseline model.

[0067] Here, to avoid short-term environmental changes from causing changes in the background noise, only the sliding window mechanism is used to dynamically update the current noise baseline. Only when the dynamically updated current noise baseline satisfies the trigger condition, the initial noise baseline model is dynamically updated. Exemplarily, the trigger condition is within the first continuous duration . Among them, is the mean of the dynamically updated current noise baseline, is the mean and standard deviation of the initial noise baseline model. In this way, the current noise baseline can be adjusted in real time in the short term without affecting the initial noise baseline model. Only when the trigger condition is met, that is, the baseline shifts and the offset is greater than the allowable value, the initial noise baseline model is updated. In this way, it can adapt to environmental changes in real time and optimize the detection accuracy.

[0068] Optionally, step S204, where the processor determines the current noise baseline to obtain the initial threshold, includes: The processor calculates the initial threshold .

[0069] Wherein, is the mean of the current noise baseline, is the standard deviation of the current noise baseline; k is the safety factor. Optionally, k takes values from 2 to 3.

[0070] Here, it can be understood that after the current noise baseline or the initial noise baseline model is updated, the updated initial threshold is calculated based on the updated current noise baseline. To ensure that the initial threshold is calculated based on reliable noise statistics during each detection.

[0071] Optionally, after the processor obtains the initial threshold, it further includes: the processor adjusts the initial threshold according to the preliminary determination situation and the actual situation.

[0072] Here, when the preliminary determination of whether there is a person and the actual situation are inconsistent in multiple consecutive detection cycles, the initial threshold is dynamically adjusted. Exemplarily, the multiple detection cycles can be 5 detection cycles. Specifically, the initial threshold is adjusted through the following formula , where β is the learning rate (exemplarily, taking values from 0.1 to 0.3), is the signal feature value (such as the mean signal intensity) during false alarm / missed alarm, is the adjusted initial threshold, is the initial threshold before adjustment. Thus, the dynamic adjustment of the initial threshold is ensured, avoiding false alarms or missed alarms caused by not learning the environmental noise.

[0073] Combined with Figure 4 shown, the embodiments of the present disclosure provide another ultrasonic sensing method for an air conditioner, including: S101, the processor receives the echo signal of the ultrasonic sensor transmitting signal to extract the spatial features and dynamic features in the echo signal.

[0074] S121, the processor determines the current noise baseline to obtain the initial threshold.

[0075] S122, when the signal intensity in the extracted spatial features is greater than or equal to the initial threshold, the processor preliminarily determines that there is a user in the target space.

[0076] S123, when there is a user in the target space, the processor fuses the extracted spatial features and dynamic features.

[0077] In S103, the processor inputs the fused features into a lightweight hybrid neural network model to obtain a classification result, and adjusts the operating parameters of the air conditioner according to the classification result; wherein, the classification result includes the perceived user position and user behavior state.

[0078] In S204, the processor corrects the initial threshold according to the user behavior state in the classification result.

[0079] Wherein, in the case where the user behavior state is a relatively static state, the initial threshold is corrected downward; in the case where the user behavior state is a dynamic state, the initial threshold is corrected upward.

[0080] Here, after updating the initial threshold based on the preliminary determination situation and environmental noise, the initial threshold is further dynamically corrected based on the classification result of the lightweight hybrid neural network model. Specifically, when the user behavior state in the classification result is a relatively static state, the initial threshold is corrected downward. That is, the sensitivity of the initial threshold is reduced to avoid missing the detection of static users. When the user behavior state in the classification result is a dynamic state, the initial threshold is corrected upward. That is, the sensitivity of the threshold is increased to improve the anti-interference ability to filter out high-frequency noise. Among them, relatively static means that the user makes small movements or local movements. The dynamic state means that the user is in a moving state, such as walking or moving in place. In this way, while improving the response speed, the accuracy of detection is ensured.

[0081] Combined Figure 5 As shown, an ultrasonic sensing device 100 for an air conditioner provided by an embodiment of the present disclosure includes a processor 101 and a memory 102. Optionally, the device may further include a communication interface 103 and a bus 104. Among them, the processor 101, the communication interface 103, and the memory 102 can complete mutual communication through the bus 104. The communication interface 103 can be used for information transmission. The processor 101 can call the logical instructions in the memory 102 to execute the ultrasonic sensing method for the air conditioner in the above embodiment.

[0082] In addition, when the logical instructions in the above-mentioned memory 102 are implemented in the form of a software functional unit and sold or used as an independent product, they can be stored in a computer-readable storage medium.

[0083] The memory 102, as a computer-readable storage medium, can be used to store software programs and computer-executable programs, such as the program instructions / modules corresponding to the method in the embodiment of the present disclosure. The processor 101 executes functional applications and data processing by running the program instructions / modules stored in the memory 102, that is, implements the ultrasonic sensing method for the air conditioner in the above embodiment.

[0084] The memory 102 may include a program storage area and a data storage area. Among them, the program storage area may store an operating system and application programs required for at least one function; the data storage area may store data created according to the use of the terminal device and the like. In addition, the memory 102 may include high-speed random access memory and may also include non-volatile memory.

[0085] Combined Figure 6 As shown, an embodiment of the present disclosure provides an air conditioner 200, including: an air conditioner body, and the above ultrasonic sensing device 100 for an air conditioner. The ultrasonic sensing device 100 for an air conditioner is installed on the air conditioner body. The installation relationship described here is not limited to being placed inside the air conditioner body, but also includes installation connections with other components of the air conditioner 200, including but not limited to physical connections, electrical connections, or signal transmission connections, etc. Those skilled in the art can understand that the ultrasonic sensing device 100 for an air conditioner can be adapted to a feasible air conditioner body, thereby implementing other feasible embodiments.

[0086] An embodiment of the present disclosure provides a computer-readable storage medium storing computer-executable instructions, and the computer-executable instructions are set to execute the above ultrasonic sensing method for an air conditioner.

[0087] The technical solution of the embodiment of the present disclosure can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes one or more instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in the embodiment of the present disclosure. The foregoing storage medium may be a non-transitory storage medium, for example: a USB flash drive, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk, or an optical disc, etc., which are various media that can store program codes.

[0088] The above description and the accompanying drawings fully illustrate the embodiments of the present disclosure, enabling those skilled in the art to practice them. Other embodiments may include structural, logical, electrical, process, and other changes. The embodiments merely represent possible variations. Unless explicitly required, individual components and functions are optional, and the order of operations may vary. Parts and features of some embodiments may be included in or replace parts and features of other embodiments. Moreover, the terms used in this application are only for describing the embodiments and are not used to limit the claims. As used in the description of the embodiments and the claims, unless the context clearly indicates otherwise, the singular forms "a", "an", and "the" are intended to also include the plural forms. Similarly, as used in this application, the term "and / or" refers to any and all possible combinations of one or more of the associated listed items. Additionally, when used in this application, the term "comprise" and its variants "comprises" and / or "comprising" etc. mean the presence of the stated features, wholes, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components, and / or groups thereof. Without further limitation, an element defined by the statement "comprising an..." does not exclude the presence of additional identical elements in the process, method, or apparatus comprising the element. Herein, each embodiment may focus on the differences from other embodiments, and the same or similar parts among the various embodiments may be referred to each other. For the methods, products, etc. disclosed in the embodiments, if they correspond to the method parts disclosed in the embodiments, the relevant parts may refer to the description of the method parts.

Claims

1. An ultrasonic sensing method for an air conditioner, characterized in that, A plurality of ultrasonic sensor transceiver arrays are arranged on the air conditioner; the method includes: Receiving the echo signal of the ultrasonic sensor transmitting signal to extract the spatial features and dynamic features in the echo signal; Fusing the extracted spatial features and dynamic features; Inputting the fused features into a lightweight hybrid neural network model to obtain a classification result, and adjusting the operating parameters of the air conditioner according to the classification result; wherein, the classification result includes the perceived user position and user behavior state.

2. The method according to claim 1, wherein Extracting the spatial features and dynamic features in the echo signal includes: Using the convolutional layer of the lightweight hybrid neural network model to extract the spatial features in the echo signal received by each ultrasonic sensor; the spatial features include signal intensity distribution and direction of arrival; Using the gated recurrent unit layer of the lightweight hybrid neural network model to extract the dynamic features in the echo signals continuously received by each ultrasonic sensor; the dynamic features include user movement speed and acceleration.

3. The method according to claim 1, characterized in that, Fusing the extracted spatial features and dynamic features includes: Stitching and fusing the spatial feature map representing the spatial features and the temporal feature map representing the dynamic features to obtain a stitched feature map; Calculating the spatial attention weight and the temporal attention weight for the stitched feature map based on the attention mechanism; Based on the spatial attention weight and the temporal attention weight, obtaining the fused features by weighted summation.

4. The method according to claim 1, characterized in that, Fusing the extracted spatial features and dynamic features includes: Determining the current noise baseline to obtain an initial threshold; When the signal intensity in the extracted spatial features is greater than or equal to the initial threshold, preliminarily determining the user situation in the target space; When there is a user in the target space, fusing the extracted spatial features and dynamic features.

5. The method according to claim 4, wherein Determining the current noise baseline includes: Collecting background noise signals when there is no user in the target space; Calculating the statistical features of the background noise signals and establishing an initial noise baseline model; Taking the statistical features of the initial noise baseline model as the current noise baseline.

6. The method according to claim 5, wherein After establishing the initial noise baseline model, it further includes: Using the sliding window mechanism to update the background noise signals in real time; Calculating the statistical features of the background noise signals in the updated sliding window to dynamically update the initial noise baseline model.

7. The method according to any one of claims 4 to 6, characterized in that After inputting the fused features into the hybrid neural network model to obtain a classification result, it further includes: Correcting the initial threshold according to the user behavior state in the classification result; Wherein, when the user behavior state is a relatively static state, the initial threshold is corrected downward; when the user behavior state is a dynamic state, the initial threshold is corrected upward.

8. An ultrasonic sensing device for an air conditioner, comprising a processor and a memory storing program instructions, characterized in that, The processor is configured to execute the ultrasonic sensing method for an air conditioner according to any one of claims 1 to 7 when running the program instructions.

9. An air conditioner, characterized in that, It includes: An air conditioner body provided with a plurality of ultrasonic sensor transceiver arrays; And the ultrasonic sensing device for an air conditioner according to claim 8, installed on the air conditioner body.

10. A computer-readable storage medium storing program instructions, characterized in that, When the program instructions are running, they are used to cause the computer to execute the ultrasonic sensing method for an air conditioner according to any one of claims 1 to 7.

Images