Air conditioners and their control methods, storage media, and program products

Abstract

This application relates to an air conditioner and its control method, storage medium, and program product. The method includes: acquiring multi-source information of a target animal; the multi-source information includes radar data, infrared thermal imaging data, and image data; extracting features from the multi-source information to obtain a first feature vector; the first feature vector is composed of a radar feature vector, an infrared feature vector, and an image feature vector; the three feature vectors have the same number of dimensions; inputting the first feature vector into a pre-trained attention weight model to obtain dynamic weight coefficients corresponding to the three feature vectors; performing weighted calculation and concatenation of the three feature vectors according to the dynamic weight coefficients to obtain a second feature vector; inputting the second feature vector into a preset spatial channel attention model to obtain a third feature vector; and identifying the current abnormal state of the target animal based on preset recognition rules and the third feature vector. This method enables the air conditioner to more accurately identify the current abnormal state of the target animal.

Description

Technical Field

[0001] This application relates to the field of air conditioning technology, and in particular to an air conditioner and its control method, storage medium, and program product. Background Technology

[0002] With the increasing popularity of pet ownership, the comfort of pets' living environment has gradually become a core concern for owners. To provide a comfortable living environment for pets, some air conditioners have a pet mode. Specifically, in pet mode, air conditioners use various sensors to collect data such as the pet's body temperature, heart rate, and activity trajectory. This data is then weighted and fused to identify the pet's current state, and the air conditioner's operating parameters (such as set temperature and fan speed) are adjusted to improve the pet's comfort. However, current weighted fusion methods mostly use fixed weights. This method struggles to reflect the complex relationship between various data points such as body temperature and heart rate and the pet's behavioral state. Consequently, the weighted fused data may not accurately reflect the pet's actual condition, resulting in low accuracy in the air conditioner's identification of the pet's current state. Summary of the Invention

[0003] Based on this, this application provides an air conditioning control method that enables the air conditioning to more accurately identify the current abnormal state of the target animal.

[0004] On one hand, this application provides a method for controlling an air conditioner, the method comprising: acquiring multi-source information of a target animal; the multi-source information including radar data, infrared thermal imaging data, and image data; extracting features from the multi-source information to obtain a first feature vector; the first feature vector being composed of a radar feature vector, an infrared feature vector, and an image feature vector concatenated; the radar feature vector, the infrared feature vector, and the image feature vector having the same number of dimensions; inputting the first feature vector into a pre-trained attention weight model to obtain dynamic weight coefficients corresponding to the radar feature vector, the infrared feature vector, and the image feature vector respectively; performing weighted calculations and re-concatenating the radar feature vector, the infrared feature vector, and the image feature vector according to the dynamic weight coefficients to obtain a second feature vector; inputting the second feature vector into a preset spatial channel attention model to obtain a third feature vector; and identifying the current abnormal state of the target animal based on preset recognition rules and the third feature vector.

[0005] Optionally, the multi-source information also includes the acquisition time of the image data; the step of extracting features from the multi-source information to obtain the first feature vector includes: obtaining point cloud feature information and posture information of the target animal based on the image data; the point cloud feature information includes the three-dimensional coordinates and color of the sampled point cloud; the posture information includes the two-dimensional coordinates of the preset joints of the target animal; inputting the point cloud feature information into a preset point cloud extraction model to obtain the point cloud feature vector; obtaining a posture sequence based on the posture information, historical posture information and the acquisition time of the image data; inputting the posture sequence into a preset temporal neural network model to obtain the posture feature vector; and fusing the point cloud feature vector and the posture feature vector to obtain the image feature vector.

[0006] Optionally, the step of extracting features from multi-source information to obtain a first feature vector includes: processing radar data based on a preset one-dimensional convolutional neural network to obtain a radar feature vector; the radar feature vector is used to characterize the target animal's respiration, heart rate, and micro-motion information; processing infrared thermal imaging data based on a preset two-dimensional convolutional neural network to obtain an infrared feature vector; the infrared feature vector is used to characterize the target animal's body temperature and body temperature distribution.

[0007] Optionally, the step of identifying the current abnormal state of the target animal based on a preset identification rule and a third feature vector includes: inputting the third feature vector into a preset classification model to obtain the probability distribution of the target animal in each preset abnormal state; wherein the training data of the preset classification model is generated based on a baseline model library of the target animal and preset expert rules; the baseline model library stores the normal range of vital signs and normal behavior baseline of the target animal; obtaining a first preset abnormal state; wherein the probability proportion of being in the first preset abnormal state is greater than or equal to a first preset threshold; and determining the current abnormal state of the target animal as the first preset abnormal state.

[0008] Optionally, the preset abnormal states include at least one of the following: fainting, heatstroke, proximity to a dangerous area, anxiety, abnormal excretion, abnormal vomiting, and cold.

[0009] Optionally, the method further includes: in response to the current abnormal state of the target animal, executing an alarm command corresponding to the current abnormal state of the target animal; in response to a false alarm feedback command issued by the user, updating the normal vital signs range and / or normal behavior baseline of the target animal in the baseline model library based on the third feature vector of the target animal at the time of the false alarm.

[0010] On the other hand, this application also provides an air conditioner, comprising: an acquisition module for acquiring multi-source information of a target animal, the multi-source information including radar data, infrared thermal imaging data, and image data; an extraction module for extracting features from the multi-source information to obtain a first feature vector; the first feature vector is composed of radar feature vectors, infrared feature vectors, and image feature vectors of the same dimension; a first calculation module for inputting the first feature vector into a pre-trained attention weight model to obtain dynamic weight coefficients corresponding to the radar feature vector, infrared feature vector, and image feature vector respectively; a second calculation module for performing weighted calculations and re-concatenating the radar feature vector, infrared feature vector, and image feature vector according to the dynamic weight coefficients to obtain a second feature vector; a third calculation module for inputting the second feature vector into a preset spatial channel attention model to obtain a third feature vector; and a recognition module for recognizing the current abnormal state of the target animal based on preset recognition rules and the third feature vector.

[0011] In another aspect, this application also provides another type of air conditioner, including a memory and a processor, wherein the memory stores a computer program and the processor is configured to run the computer program to perform the aforementioned method.

[0012] In another aspect, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the aforementioned method.

[0013] In another aspect, this application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the aforementioned method.

[0014] The embodiments provided in this application assign dynamic weight coefficients to the radar feature vector, infrared feature vector and image feature vector in the first feature vector by using a pre-trained attention weight model, and perform weighted fusion. Then, the fused feature vector is enhanced by a preset spatial channel attention model, so that the final third feature vector can focus on the features related to the abnormal state of the target animal. This allows the third feature vector to more accurately reflect the actual state of the pet, and enables the air conditioner to more accurately identify the current abnormal state of the target animal. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating an air conditioner control method according to some embodiments of this application;

[0016] Figure 2 This is a schematic diagram of the structure of an air conditioner according to some embodiments of this application;

[0017] Figure 3 This is a schematic diagram of the structure of an air conditioner according to other embodiments of this application. Detailed Implementation

[0018] To make the technical solution and beneficial effects of this application more apparent and understandable, a detailed description is provided below by listing specific embodiments. Unless otherwise defined, the technical and scientific terms used in this application have the same meaning as those in the technical field to which this application pertains. The terminology used in the specification of this application is for the purpose of describing specific embodiments only and is not intended to limit this application.

[0019] See Figure 1 Some embodiments of this application provide a method for controlling an air conditioner, the method comprising:

[0020] S101, acquire multi-source information about the target animal; the multi-source information includes radar data, infrared thermal imaging data, and image data.

[0021] S102, feature extraction is performed on the multi-source information to obtain the first feature vector. The first feature vector is composed of radar feature vector, infrared feature vector, and image feature vector concatenated together. The radar feature vector, infrared feature vector, and image feature vector have the same number of dimensions.

[0022] S103, input the first feature vector into the pre-trained attention weight model to obtain the dynamic weight coefficients corresponding to the radar feature vector, infrared feature vector and image feature vector respectively.

[0023] S104, the radar feature vector, infrared feature vector and image feature vector are weighted and re-concatenated according to the dynamic weighting coefficient to obtain the second feature vector.

[0024] S105, input the second feature vector into the preset spatial channel attention model to obtain the third feature vector.

[0025] S106, based on preset recognition rules, identify the current abnormal state of the target animal according to the third feature vector.

[0026] Understandably, the target animal refers to the animal that the air conditioner uses to identify abnormal behavior. The target animal can be a pet (such as a cat or dog) or livestock (such as a pig, cow, or chicken) from a farm. When there are multiple animals within the air conditioner's detection range, one can be designated as the target animal through user specification or other methods.

[0027] Multi-source information refers to a set of real-time data about a target animal obtained through various types of acquisition devices. Radar data can be acquired using radar in specific frequency bands (e.g., 60 GHz), infrared thermal imaging data can be acquired using infrared cameras, and image data can be acquired using visible light cameras. The first feature vector refers to the original feature vector extracted from the multi-source information. Specifically, the radar feature vector, extracted from radar data, is used to characterize the target animal's respiratory rate, heart rate, and micro-motion information (such as limb twitching, curling, and trembling). The infrared feature vector, extracted from infrared data, is used to characterize the surface temperature distribution of the target animal. The image feature vector, extracted from image data, is used to characterize the target animal's outline, joint posture, movement trajectory, and activity area.

[0028] A pre-trained attention weight model is a neural network model that takes a first feature vector as input and outputs three dynamic weight coefficients between 0 and 1 based on the data quality and scene information of each modal feature (radar feature vector, infrared feature vector, and image feature vector) in the first feature vector. For example, when the image feature vector shows that the target animal is obstructed by a blanket, resulting in low reliability of the visible light image data, the weight of the image feature vector can be reduced. Or, when the infrared feature vector shows that the target animal is at risk of heatstroke, the weight of the infrared feature vector can be increased. The attention weight model can be trained based on historical data of the first feature vector with scene and data quality labels. A pre-set spatial channel attention model is a pre-set attention model that combines spatial attention models and channel attention models, such as the CBAM module (Convolutional Block Attention Module). Among them, the spatial attention model can focus on key parts of the pet in the spatial dimension, such as the chest cavity (the core area for judging breathing and heart rate) and limbs (the core area for judging posture), and suppress interfering factors such as fur and background furniture. The channel attention model can strengthen the channel weights of core features such as vital signs, body surface temperature, and abnormal movement patterns from the channel dimension, while weakening the influence of irrelevant channels such as respiratory micro-movements.

[0029] The sequential weighted calculation and re-concatenation process involves first weighting each modal feature based on the dynamic weight coefficients output by the pre-trained attention weight model, and then re-concatenating the features to obtain the second feature vector. The preset recognition rule refers to a pre-defined mapping relationship or algorithm for obtaining the current abnormal state of the target animal based on the third feature vector.

[0030] The aforementioned air conditioning control method assigns dynamic weight coefficients to the radar feature vector, infrared feature vector, and image feature vector in the first feature vector using a pre-trained attention weight model, and performs weighted fusion. Subsequently, a preset spatial channel attention model is used to enhance the fused feature vector, so that the final third feature vector can focus on features related to the abnormal state of the target animal. This allows the third feature vector to more accurately reflect the actual state of the pet, enabling the air conditioner to more accurately identify the current abnormal state of the target animal.

[0031] Furthermore, the multi-source information acquired in the aforementioned air conditioning control method includes radar data, infrared thermal imaging data, and image data. Radar data can penetrate the target animal's fur, clothing, and thin obstructions; infrared thermal imaging data is unaffected by natural light. Therefore, radar data and infrared thermal imaging data can compensate for the weaknesses of image data, such as susceptibility to obstructions and light effects. This allows the acquired multi-source information to more comprehensively reflect the target animal's current state, thereby improving the accuracy of the air conditioning system in identifying abnormal states in the target animal.

[0032] Furthermore, in the first feature vector of the aforementioned air conditioning control method, the radar feature vector, infrared feature vector, and image feature vector have the same number of dimensions, so that they can be weighted according to the dynamic weight coefficients, thereby reducing the difficulty of data fusion.

[0033] Optionally, the third feature vector has fewer dimensions than the second feature vector. This reduces the computational load required to identify the current abnormal state of the target animal based on the third feature vector, thereby reducing the computational cost of the air conditioner.

[0034] As a concrete example, the radar feature vector, infrared feature vector, and image feature vector can all have a dimension of 64, the first and second feature vectors have a dimension of 192, and the third feature vector has a dimension of 96. The attention weight model can consist of a fully connected layer and a sigmoid activation function, where the fully connected layer has an input dimension of 192 and an output dimension of 3.

[0035] In some embodiments, the multi-source data further includes the acquisition times of radar data, infrared sensor data, and image data. Step S101, the step of acquiring multi-source data of the target animal, includes: aligning the acquired radar data, infrared sensor data, and image data according to their acquisition times. This allows the radar data, infrared sensor data, and image data to characterize the target animal's state at the same time, facilitating subsequent data fusion and state recognition.

[0036] In some embodiments, step S102, which extracts features from multi-source information to obtain a first feature vector, includes: obtaining point cloud feature information and pose information of the target animal based on image data. The point cloud feature information includes the three-dimensional coordinates and color of the sampled point cloud. The pose information includes the two-dimensional coordinates of preset joints of the target animal. The point cloud feature information is input into a preset point cloud extraction model to obtain a point cloud feature vector. A pose sequence is obtained based on the pose information, historical pose information, and the acquisition time of the image data. The pose sequence is input into a preset temporal neural network model to obtain a pose feature vector. The point cloud feature vector and the pose feature vector are fused to obtain an image feature vector.

[0037] Specifically, point cloud feature information refers to the feature information of several sampled point clouds obtained from image data through a 3D reconstruction algorithm and point cloud sampling. The number of sampled point clouds can be preset, such as 128. The 3D coordinates of the sampled point clouds can reflect the position, volume, and shape of the target animal in 3D space; the color of the sampled point clouds can reflect different parts represented by the point clouds. After obtaining the point cloud feature information, a point cloud feature vector can be extracted from the point cloud feature information using a preset point cloud extraction model (such as the PointNet model).

[0038] Posture information refers to the two-dimensional coordinates of preset joints of a target animal extracted from image data using a pre-defined posture estimation algorithm. The number of preset joints can be pre-set based on the target animal's 3D model; for example, 16 preset joints could be used for canines. After obtaining the posture information, it can be combined with historical posture information and the image data acquisition time to obtain a posture sequence. This posture sequence includes multiple posture information sets and their corresponding acquisition times. The length of the posture sequence can be pre-set; for example, it could include 10 sets of posture information sets plus acquisition times. The posture sequence can be processed using a pre-defined temporal neural network model (such as an LSTM (Long Short-Term Memory) model) to obtain a posture feature vector. This posture feature vector reflects the dynamic changes in the two-dimensional coordinates of the preset joints within the time period covered by the posture sequence.

[0039] Finally, the obtained point cloud feature vector and pose feature vector can be concatenated and input into a fully connected layer to perform dimensionality reduction and fusion of the point cloud feature vector and pose feature vector to obtain the image feature vector.

[0040] The aforementioned air conditioning control method obtains an image feature vector by fusing point cloud feature vectors and attitude feature vectors. This method can fuse the three-dimensional shape of the target animal in a static state with the dynamic changes of the two-dimensional coordinates of preset joint points, making the output image feature vector more comprehensively represent the actual image of the target animal, thereby improving the accuracy of subsequent identification of the target animal's current abnormal state.

[0041] Furthermore, in the aforementioned air conditioning control method, point cloud feature vectors and attitude feature vectors are fused to obtain image feature vectors. The advantage of this approach is that it can reduce the number of dimensions of the output image feature vectors, remove redundant features, and improve the representational ability of the image feature vectors.

[0042] It is understood that in the above embodiments, the number of dimensions of the point cloud feature vector, pose feature vector, and image feature vector can be preset according to the actual required accuracy. For example, the point cloud feature vector and pose feature vector can both be set to 64, and the number of dimensions of the fused image feature vector can also be set to 64.

[0043] In some embodiments, step S102, which involves extracting features from multi-source information to obtain a first feature vector, includes: processing radar data based on a preset one-dimensional convolutional neural network to obtain a radar feature vector. The radar feature vector is used to characterize the target animal's respiration, heart rate, and micro-motion information. Infrared thermal imaging data is also processed based on a preset two-dimensional convolutional neural network to obtain an infrared feature vector. The infrared feature vector is used to characterize the target animal's body temperature and body temperature distribution.

[0044] Specifically, one-dimensional convolutional neural networks can efficiently process one-dimensional data such as radar data through local convolution. Therefore, the aforementioned air conditioning control method uses a pre-defined one-dimensional convolutional neural network to process radar data and obtain radar feature vectors, which can effectively improve the processing efficiency of radar data.

[0045] As a specific example, the radar data can be a single-channel time-series signal of length 256. The pre-defined one-dimensional convolutional neural network can be a three-layer one-dimensional convolutional neural network. Each layer of the one-dimensional convolutional neural network sequentially performs one-dimensional convolution, batch normalization, ReLU activation, and max-pooling downsampling operations. The kernel sizes of the three convolutional layers are 3, 5, and 3, respectively, with a stride of 1, and the number of output channels are 32, 64, and 128, respectively. The pre-defined one-dimensional convolutional neural network ultimately outputs a 64-dimensional radar feature vector through a fully connected layer. It is understood that in other embodiments, the pre-defined one-dimensional convolutional neural network can also have other structures.

[0046] Similarly, two-dimensional convolutional neural networks are suitable for processing two-dimensional spatial data such as infrared thermal imaging data. Their advantage lies in their ability to automatically learn local patterns and global structures of body temperature distribution by sliding the convolutional kernel along the spatial dimension, thereby capturing the spatial correlation of body temperature changes. Therefore, the aforementioned air conditioning control method, using a pre-defined two-dimensional convolutional neural network to process infrared thermal imaging data and obtain infrared feature vectors, can effectively improve the processing efficiency of infrared thermal imaging data.

[0047] As a specific example, the input infrared thermal imaging data can be a 64×64 single-channel temperature matrix. The preset two-dimensional convolutional neural network can be a four-layer two-dimensional convolutional network, with each layer sequentially performing two-dimensional convolution, batch normalization, ReLU activation, and max pooling downsampling operations. The kernel size of the four convolutional layers is 3×3, with a stride of 1, and the number of output channels is 16, 32, 64, and 64 respectively. The preset two-dimensional convolutional neural network finally outputs a 64-dimensional infrared feature vector through a global average pooling layer. It is understood that in other embodiments, the preset two-dimensional convolutional neural network can also have other structures.

[0048] Furthermore, it is understood that in the above embodiments, the number of dimensions for both the radar feature vector and the infrared feature vector is set to 64, which is exemplary. In other embodiments, the number of dimensions for the radar feature vector and the infrared feature vector can also be set to other values, such as 128.

[0049] In some embodiments, step S106, based on preset recognition rules and according to a third feature vector, identifies the current abnormal state of the target animal, including: inputting the third feature vector into a preset classification model to obtain the probability distribution of the target animal being in each preset abnormal state. The training data of the preset classification model is generated based on a baseline model library of the target animal and preset expert rules. The baseline model library stores the normal range of vital signs and normal behavioral baselines of the target animal. A first preset abnormal state is obtained. The probability percentage of being in the first preset abnormal state is greater than or equal to a first preset threshold. The current abnormal state of the target animal is determined as the first preset abnormal state.

[0050] Understandably, a predefined classification model refers to a neural network model with classifier functionality, which can generate a probability distribution vector based on the input third feature vector to represent the probability of the target animal being in each predefined abnormal state. The number of dimensions of this probability distribution vector is equal to the number of predefined abnormal states.

[0051] A baseline model library refers to a data knowledge base established for a target animal. The normal vital signs range of the target animal refers to the normal range of physiological signs such as heart rate, body temperature range, and respiratory rate. The normal behavioral baseline refers to the typical posture and behavioral patterns of the target animal under normal conditions. The baseline model library can be obtained in various ways. For example, within 24 hours after air conditioning initialization, the vital signs range and behavioral patterns of the target animal in a healthy state can be collected to form the baseline model library. The healthy state can be manually calibrated by the user. Preset expert rules refer to a set of multiple expert rules that define how various preset abnormal states change physiological and behavioral characteristics. For example, heatstroke will cause the target animal to experience an increase in body temperature, rapid breathing, and agitation.

[0052] The first preset threshold is a threshold value used to compare the probability proportion of preset abnormal states. When the probability proportion of a certain preset abnormal state is greater than or equal to the first preset threshold, it indicates that the target animal has a high probability of being in that preset abnormal state. It can be understood that the first preset threshold is a preset value and can be set according to the user's requirements for recognition sensitivity; for example, the first preset threshold can be set to 0.8.

[0053] Specifically, the pre-defined classification model needs to be trained and iterated using training data labeled with each pre-defined abnormal state. The training method can be supervised learning or reinforcement learning, and there are no restrictions here. However, due to the significant individual differences among different animals, the pre-defined classification model obtained using historical data from other animals often fails to accurately determine the current abnormal state of the target animal.

[0054] Therefore, in the aforementioned air conditioning control method, a baseline model library of the target animal and preset expert rules are used to generate training data. Specifically, the normal range of vital signs and the baseline of normal behavior of the target animal are used as the basic data. By analyzing the data variation patterns of different preset abnormal states in the preset expert rules, these basic data are either adjusted upwards or downwards. Finally, the label of the preset abnormal state is added to obtain the training data, which is then used to train the preset classification model. This allows the preset classification model to learn the individual differences of the target animal, resulting in a more accurate probability distribution output by the preset classification model and improving the accuracy of subsequent identification.

[0055] Furthermore, the aforementioned air conditioning control method, through a preset classification model, can simultaneously output the probability distribution of the target animal in each preset abnormal state. The probability distribution can simultaneously present the probability of the target animal in different preset abnormal states, which can more comprehensively reflect the current state of the target animal, thereby further improving the accuracy of identification.

[0056] As a specific example, the preset classification model can consist of a fully connected layer and a Softmax classifier. The fully connected layer is used to perform non-linear transformation and feature integration on the third feature vector, and the Softmax classifier is used to output the probability distribution vector. It is understood that the specific structure of the preset classification model described above is exemplary, and in other embodiments, the preset classification model may also have other structures.

[0057] Furthermore, it is understood that in the aforementioned air conditioning control method, identifying a first preset abnormal state with a probability proportion greater than or equal to a first preset threshold as the current abnormal state of the target animal is exemplary. In other embodiments, the air conditioning control method can also identify the abnormal state of the target animal from the probability distribution in other ways. For example, the probability value of each preset abnormal state can be compared with its corresponding preset probability threshold to determine the current abnormal state of the target animal. The advantage of this approach is that different preset probability thresholds can be set according to the danger level of the preset abnormal state, thereby balancing recognition sensitivity and recognition accuracy. Specifically, for high-risk abnormal states such as syncope, the corresponding preset probability threshold is smaller, thereby improving the recognition sensitivity for such high-risk abnormal states; for ordinary abnormal states such as cold, the corresponding preset probability threshold is larger to reduce the probability of false recognition.

[0058] In some embodiments, the preset abnormal state includes at least one of fainting, heatstroke, proximity to a dangerous area, anxiety, abnormal bowel movements, abnormal vomiting, and cold. This enables the air conditioner to identify multiple different abnormal states of the target animal.

[0059] Understandably, "near a danger zone" refers to areas where the target animal's current location is close to electrical outlets, sharp furniture edges, medicine storage areas, or other areas that could pose a threat to the animal's safety. Danger zones can be automatically designated by the air conditioner according to preset rules, or manually designated by the user; no restrictions are placed here.

[0060] In some embodiments, the air conditioning control method further includes: in response to the current abnormal state of the target animal, executing an alarm command corresponding to the current abnormal state of the target animal; and in response to a false alarm feedback command issued by the user, updating the normal vital signs range and / or normal behavior baseline of the target animal in the baseline model library based on the third feature vector of the target animal at the time of the false alarm.

[0061] The aforementioned air conditioning control method, by executing an alarm command corresponding to the current abnormal state of the target animal when the air conditioner detects that the target animal is in an abnormal state, can remind the user to deal with the abnormal situation of the target animal in a timely manner, thereby ensuring the health of the target animal.

[0062] Furthermore, the aforementioned air conditioning control method also updates the normal vital signs range and / or normal behavior baseline of the target animal in the baseline model library based on the third feature vector of the target animal at the time of the false alarm when the user issues a false alarm feedback command. This allows the preset classification model to iterate based on the updated baseline model library, so as to avoid misidentifying the abnormal state of the target animal again in subsequent identification, thereby reducing the probability of false alarms and improving the user experience.

[0063] refer to Figure 2 This application also provides an air conditioner 100 in some embodiments. The air conditioner 100 includes an acquisition module 110, an extraction module 120, a first calculation module 130, a second calculation module 140, a third calculation module 150, and a recognition module 160. The acquisition module 110 acquires multi-source information about a target animal. The multi-source information includes radar data, infrared thermal imaging data, and image data. The extraction module 120 extracts features from the multi-source information to obtain a first feature vector. The first feature vector is constructed by concatenating radar feature vectors, infrared feature vectors, and image feature vectors of the same dimension. The first calculation module 130 inputs the first feature vector into a pre-trained attention weight model to obtain dynamic weight coefficients corresponding to the radar feature vector, infrared feature vector, and image feature vector, respectively. The second calculation module 140 performs weighted calculations and re-concatenates the radar feature vector, infrared feature vector, and image feature vector according to the dynamic weight coefficients to obtain a second feature vector. The third calculation module 150 inputs the second feature vector into a preset spatial channel attention model to obtain a third feature vector. The identification module 160 is used to identify the current abnormal state of the target animal based on a preset identification rule and a third feature vector.

[0064] The various modules in the aforementioned air conditioner can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of a computer device in hardware form or independently of it, or stored in the air conditioner's memory in software form, so that the processor can invoke and activate the corresponding operations of each module. It should be noted that the above module division is illustrative and represents only a logical functional division; in actual implementation, other division methods may be used.

[0065] Based on the aforementioned embodiments of the air conditioning control method, in another embodiment provided in this application, an air conditioner is provided, the internal structure of which can be shown in the diagram below. Figure 3As shown, the air conditioner includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The database stores data. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network. When the computer program is activated by the processor, it implements an air conditioner control method.

[0066] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the air conditioner to which the present application is applied. A specific air conditioner may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0067] Based on the aforementioned embodiments of the air conditioning control method, in another embodiment provided in this application, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is opened by a processor, it implements the steps in the above-described method embodiments.

[0068] Based on the aforementioned embodiments of the air conditioning control method, in another embodiment provided in this application, a computer program product is provided, including a computer program that, when activated by a processor, implements the steps in the above-described method embodiments.

[0069] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims. The terms "comprising," "including," or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, product, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, product, or apparatus. Without further limitation, the presence of other identical or equivalent elements in a process, method, product, or apparatus that includes said elements is not excluded. For example, the use of terms such as "first," "second," etc., is to denote names and does not indicate any specific order.

[0070] When used herein, the singular forms of “a,” “an,” and “the” may also include the plural forms unless the context clearly indicates otherwise. It should also be understood that the terms “comprising / including” or “having,” etc., specify the presence of the stated features, wholes, steps, operations, components, parts, or combinations thereof, but do not preclude the possibility of the presence or addition of one or more other features, wholes, steps, operations, components, parts, or combinations thereof. Meanwhile, in this specification, the term “and / or” includes any and all combinations of the associated listed items.

[0071] In the description of this application, the references to terms such as "some embodiments," "other embodiments," "ideal embodiments," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative descriptions of the above terms do not necessarily refer to the same embodiments or examples.

[0072] It is understood that the various embodiments of the methods described in this specification are presented in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on its differences from other embodiments. Relevant details can be found in the descriptions of other method embodiments.

[0073] It should be understood that the above embodiments are exemplary and are not intended to encompass all possible implementations included in the claims. Various modifications and changes can be made to the above embodiments without departing from the scope of this application. Similarly, the various technical features of the above embodiments can be arbitrarily combined to form other embodiments of this application that may not be explicitly described. Therefore, the above embodiments only illustrate several implementations of this application and do not limit the scope of protection of this patent application.

Claims

1. A method for controlling an air conditioner, characterized in that, The method includes: Acquire multi-source information about the target animal; the multi-source information includes radar data, infrared thermal imaging data, and image data. Feature extraction is performed on the multi-source information to obtain a first feature vector; the first feature vector is composed of radar feature vector, infrared feature vector and image feature vector; the radar feature vector, infrared feature vector and image feature vector have the same number of dimensions; The first feature vector is input into a pre-trained attention weight model to obtain the dynamic weight coefficients corresponding to the radar feature vector, the infrared feature vector, and the image feature vector, respectively. The radar feature vector, the infrared feature vector, and the image feature vector are weighted and reassembled according to the dynamic weighting coefficients to obtain the second feature vector. The second feature vector is input into a preset spatial channel attention model to obtain the third feature vector; Based on preset recognition rules, the current abnormal state of the target animal is identified according to the third feature vector.

2. The control method according to claim 1, characterized in that, The multi-source information also includes the acquisition time of the image data; the step of extracting features from the multi-source information to obtain a first feature vector includes: Based on the image data, point cloud feature information and posture information of the target animal are obtained; the point cloud feature information includes the three-dimensional coordinates and color of the sampled point cloud; the posture information includes the two-dimensional coordinates of the preset joints of the target animal. The point cloud feature information is input into a preset point cloud extraction model to obtain point cloud feature vectors; Based on the posture information, historical posture information, and the acquisition time of the image data, a posture sequence is obtained; The attitude sequence is input into a preset temporal neural network model to obtain attitude feature vectors; The point cloud feature vector and the pose feature vector are fused to obtain the image feature vector.

3. The control method according to claim 1, characterized in that, The step of extracting features from the multi-source information to obtain a first feature vector includes: The radar data is processed based on a pre-defined one-dimensional convolutional neural network to obtain the radar feature vector; the radar feature vector is used to characterize the respiration, heart rate and micro-motion information of the target animal. The infrared thermal imaging data is processed based on a preset two-dimensional convolutional neural network to obtain the infrared feature vector; the infrared feature vector is used to characterize the body temperature and body temperature distribution of the target animal.

4. The control method according to any one of claims 1-3, characterized in that, The step of identifying the current abnormal state of the target animal based on the third feature vector according to the preset recognition rules includes: The third feature vector is input into a preset classification model to obtain the probability distribution of the target animal in each preset abnormal state; wherein, the training data of the preset classification model is generated based on the baseline model library of the target animal and preset expert rules; the baseline model library stores the normal range of vital signs and the baseline of normal behavior of the target animal; Obtain a first preset abnormal state; wherein the probability of being in the first preset abnormal state is greater than or equal to a first preset threshold. The current abnormal state of the target animal is determined to be the first preset abnormal state.

5. The control method according to claim 4, characterized in that, The preset abnormal states include at least one of the following: fainting, heatstroke, proximity to a dangerous area, anxiety, abnormal excretion, abnormal vomiting, and cold.

6. The control method according to claim 4, characterized in that, The method further includes: In response to the current abnormal state of the target animal, execute the alarm command corresponding to the current abnormal state of the target animal; In response to a false alarm feedback command issued by the user, the normal vital signs range and / or normal behavior baseline of the target animal in the baseline model library are updated according to the third feature vector of the target animal at the time of the false alarm.

7. An air conditioner, characterized in that, include: The acquisition module is used to acquire multi-source information about the target animal, including radar data, infrared thermal imaging data, and image data. The extraction module is used to extract features from the multi-source information to obtain a first feature vector; the first feature vector is composed of radar feature vectors, infrared feature vectors and image feature vectors of the same dimension. The first calculation module is used to input the first feature vector into a pre-trained attention weight model to obtain dynamic weight coefficients corresponding to the radar feature vector, the infrared feature vector and the image feature vector, respectively. The second calculation module is used to perform weighted calculations and re-concatenate the radar feature vector, the infrared feature vector, and the image feature vector according to the dynamic weighting coefficients to obtain the second feature vector. The third calculation module is used to input the second feature vector into a preset spatial channel attention model to obtain the third feature vector; The identification module is used to identify the current abnormal state of the target animal based on the third feature vector according to the preset identification rules.

8. An air conditioner, comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to run the computer program to perform the method of any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images