WO2023065933A1 - 冰箱、冰箱管理系统及冰箱管理系统的控制方法 - Google Patents

冰箱、冰箱管理系统及冰箱管理系统的控制方法 Download PDF

Info

Publication number
WO2023065933A1
WO2023065933A1 PCT/CN2022/120029 CN2022120029W WO2023065933A1 WO 2023065933 A1 WO2023065933 A1 WO 2023065933A1 CN 2022120029 W CN2022120029 W CN 2022120029W WO 2023065933 A1 WO2023065933 A1 WO 2023065933A1
Authority
WO
WIPO (PCT)
Prior art keywords
positioning model
information
refrigerator
target
tags
Prior art date
Application number
PCT/CN2022/120029
Other languages
English (en)
French (fr)
Inventor
彭红亮
刘兆祥
鲍雨锋
Original Assignee
海信冰箱有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Priority claimed from CN202111207745.1A external-priority patent/CN113915840B/zh
Priority claimed from CN202111212660.2A external-priority patent/CN113915841B/zh
Application filed by 海信冰箱有限公司 filed Critical 海信冰箱有限公司
Priority to CN202280052817.6A priority Critical patent/CN117730235A/zh
Publication of WO2023065933A1 publication Critical patent/WO2023065933A1/zh

Links

Images

Classifications

    • F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
    • F25—REFRIGERATION OR COOLING; COMBINED HEATING AND REFRIGERATION SYSTEMS; HEAT PUMP SYSTEMS; MANUFACTURE OR STORAGE OF ICE; LIQUEFACTION SOLIDIFICATION OF GASES
    • F25D—REFRIGERATORS; COLD ROOMS; ICE-BOXES; COOLING OR FREEZING APPARATUS NOT OTHERWISE PROVIDED FOR
    • F25D29/00—Arrangement or mounting of control or safety devices

Definitions

  • the present disclosure relates to the technical field of refrigerators, and in particular to a refrigerator, a refrigerator management system, and a control method for the refrigerator management system.
  • Refrigerator is one of the necessary home appliances in the family. With the development of intelligence, it can provide users with more convenient and faster services. For example, with the quickening pace of life, users often forget the location of the ingredients stored in the refrigerator, and need to rummage in the refrigerator, which is very inconvenient to use. Therefore, refrigerators with ingredients management functions are chosen by more and more consumers.
  • a refrigerator in one aspect, includes multiple tags, reader-writers, controllers and cabinets.
  • the box includes a plurality of storage compartments.
  • the reader is set in the box and is configured to: send radio frequency signals to multiple tags.
  • the plurality of tags are located in a plurality of storage rooms, and each tag stores at least one type of food material information, and the tag is configured to: receive a radio frequency signal, and obtain the position of the radio frequency signal at the location of the tag according to the radio frequency signal Signal strength information, sending signal strength information and ingredient information to the reader, the controller is coupled to the reader, and the controller is configured to: respond to an ingredient location instruction of a target ingredient, and obtain the target ingredient from the reader
  • the signal strength information of the target tag corresponding to the ingredient information of the target tag; according to the signal strength information of the target tag, the location model is used to obtain the location information of the target ingredient; wherein, the location model is based on the signal strength information of multiple tags and the location of multiple tags
  • the information is pre-trained.
  • a refrigerator management system in another aspect, includes a plurality of refrigerators and a server, the server is coupled to the plurality of refrigerators, and is configured to provide positioning models to the plurality of refrigerators.
  • the refrigerator management system includes multiple refrigerators and servers.
  • Each of the plurality of refrigerators includes a plurality of tags, readers, controllers, and cabinets.
  • the box body includes a plurality of storage chambers.
  • the reader-writer is set in the box; multiple tags are located in multiple storage rooms, each of the multiple tags stores food information of at least one food material, and the controller is coupled to the reader-writer.
  • the control method of the refrigerator management system includes: the controller responds to an ingredient positioning instruction of a target ingredient, and acquires signal strength information of a target tag corresponding to the ingredient information of the target ingredient from a reader. According to the signal strength information of the target tag, the controller adopts the positioning model to obtain the position information of the target food material; wherein, the positioning model is pre-trained based on the signal strength information of multiple tags and the position information of multiple tags.
  • a non-transitory computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a refrigerator management system, the refrigerator management system executes the refrigerator management system control method One or more steps in .
  • a computer program product including a computer program, wherein, when the computer program is executed by a refrigerator management system, the refrigerator management system executes one or more of the refrigerator management system control methods. steps.
  • FIG. 1 is a schematic diagram of a refrigerator management system according to some embodiments
  • Fig. 2A is a structural diagram of a refrigerator according to some embodiments.
  • FIG. 2B is a schematic diagram of a refrigerator according to some embodiments.
  • FIG. 3A is a schematic diagram of another refrigerator according to some embodiments.
  • FIG. 3B is a schematic diagram of yet another refrigerator according to some embodiments.
  • Fig. 4 is a schematic diagram of a positioning model according to some embodiments.
  • Fig. 5 is a schematic diagram of another positioning model according to some embodiments.
  • Fig. 6 is a schematic diagram of training a positioning model according to some embodiments.
  • Figure 7A is a schematic diagram of a display according to some embodiments.
  • Figure 7B is a schematic diagram of another display according to some embodiments.
  • Figure 7C is a schematic diagram of yet another display according to some embodiments.
  • Fig. 8 is a flowchart of a control method of a refrigerator management system according to some embodiments.
  • Fig. 9 is a flowchart of another control method of a refrigerator management system according to some embodiments.
  • first and second are used for descriptive purposes only, and cannot be understood as indicating or implying relative importance or implicitly specifying the quantity of indicated technical features. Thus, a feature defined as “first” and “second” may explicitly or implicitly include one or more of these features. In the description of the embodiments of the present disclosure, unless otherwise specified, "plurality” means two or more.
  • the expressions “coupled” and “connected” and their derivatives may be used.
  • the term “connected” may be used in describing some embodiments to indicate that two or more elements are in direct physical or electrical contact with each other.
  • the term “coupled” may be used when describing some embodiments to indicate that two or more elements are in direct physical or electrical contact.
  • the terms “coupled” or “communicatively coupled” may also mean that two or more elements are not in direct contact with each other, but yet still co-operate or interact with each other.
  • the embodiments disclosed herein are not necessarily limited by the context herein.
  • At least one of A, B and C has the same meaning as “at least one of A, B or C” and both include the following combinations of A, B and C: A only, B only, C only, A and B A combination of A and C, a combination of B and C, and a combination of A, B and C.
  • a and/or B includes the following three combinations: A only, B only, and a combination of A and B.
  • the term “if” is optionally interpreted to mean “when” or “at” or “in response to determining” or “in response to detecting,” depending on the context.
  • the phrases “if it is determined that " or “if [the stated condition or event] is detected” are optionally construed to mean “when determining ! or “in response to determining ! depending on the context Or “upon detection of [stated condition or event]” or “in response to detection of [stated condition or event]”.
  • the refrigerator In daily life, the refrigerator has become one of the necessary household appliances in every family, and the food positioning function of the refrigerator is valued by more and more users.
  • the refrigerator is equipped with a reader-writer, which can predict the position and partition of the food material in the refrigerator by receiving the data of the UHF radio frequency identification (Radio Frequency Identification, RFID) tag attached to the food material, so as to realize the Intelligent management of ingredients.
  • UHF radio frequency identification Radio Frequency Identification, RFID
  • the storage location of the food identified based on the RFID tag is generally limited to which storage room of the refrigerator the food is stored in, and the specific location of the food in the storage room cannot be further identified. Then, when the storage room has a large space, when the user needs to find the ingredients, he still needs to rummage through the storage room one by one, which is time-consuming and laborious. It may also cause the air-conditioning of the refrigerator to leak out, resulting in increased power consumption, thereby reducing the user experience. Spend.
  • the refrigerator management system 3 includes a plurality of refrigerators 1 and a server 2 . Multiple refrigerators 1 are respectively coupled to the server 2 .
  • server 2 may be a cloud server.
  • the server 2 is configured to provide the positioning model to a plurality of refrigerators 1 .
  • an initial positioning model is pre-stored in the server 2, and the server 2 can train the initial positioning model.
  • the server 2 can receive the training data sent by multiple refrigerators 1 coupled thereto (for example, the training data is included in the following training sample set), and the server 2 can perform the training data sent by the multiple refrigerators 1 according to the training data.
  • the pre-stored initial localization model is used for training.
  • the server 2 may send the trained positioning model to each refrigerator 1, so that each refrigerator 1 uses the positioning model to locate the ingredients.
  • the multiple refrigerators 1 coupled to the server 2 may be of the same type, for example, the multiple refrigerators 1 have the same model, or the multiple refrigerators 1 have the same function.
  • the positioning model trained by using the training data sent by multiple refrigerators 1 of the same type can be applied to multiple refrigerators 1 of this type.
  • the present disclosure does not specifically limit the number of multiple refrigerators 1, for example, the number of multiple refrigerators 1 can be as large as possible, so that the server 2 can ignore some errors in the process of training the positioning model, thereby improving the accuracy of the positioning model.
  • the present disclosure provides a refrigerator 1 .
  • the refrigerator 1 includes a box body 10 , multiple tags 20 , a reader-writer 30 and a controller 40 .
  • the box body 10 includes a plurality of storage rooms 11 , and the plurality of storage rooms 11 may include a storage room 111 , a storage room 112 and a storage room 113 .
  • the storage chamber 111 may be a refrigerating chamber for refrigerated storage of food;
  • the storage chamber 112 may be a variable temperature chamber for storing food at a variable temperature;
  • the storage chamber 113 may be a freezing chamber for freezing and storing food.
  • the plurality of tags 20 may be RFID tags, or near field communication (Near Field Communication, NFC) tags.
  • the label 20 may include a label body detachably fixed on the food material.
  • the label body can be configured as a sealing clip, which can be clamped on both sides of the food material.
  • the label body may include a sticking part, and the sticking part may stick the label body on the food material.
  • the tag 20 further includes an antenna, and when the tag 20 is an RFID tag, the antenna is an RFID antenna.
  • the label 20 fixed on the food can be put in or taken out together.
  • the label 20 may further include a memory (for example, the memory may be a memory chip), and the memory is used for storing label information of the label, and the label information includes ingredient information.
  • each label 20 may store ingredient information of multiple ingredients, and may also store ingredient information of one ingredient, where one ingredient may be one ingredient or multiple ingredients.
  • the ingredient information stored in different tags 20 may be at least partly the same, or completely different.
  • the ingredient information on the label 20 may include, for example, the type of the ingredient, storage date, expiration date and other information.
  • the reader-writer 30 is arranged in the box body 10 , and the reader-writer 30 is configured to send radio frequency signals to a plurality of tags 20 .
  • the reader-writer 30 includes a read-write host 31 and multiple antennas 32 .
  • the read-write host 31 is coupled with multiple antennas 32 , the read-write host 31 can be set on the box body 10 , and the multiple antennas 32 can be set in each partition of the refrigerator 1 .
  • the refrigerator 1 can be divided into multiple partitions, and at least one antenna 32 is provided in each partition. It should be noted that the refrigerator 1 can be divided into various partitions according to the distribution manner of the storage rooms. For example, the refrigerator 1 can be divided into three partitions, the three partitions correspond to the three storage rooms 11 of the refrigerator 1 (i.e. the storage room 111, the storage room 112 and the storage room 113), and each storage room 11 is provided with at least An antenna 32.
  • the at least one antenna 32 may be located in the inner liner of the storage room 11, and the tag 20 may be located in a space outside the inner liner of the storage room.
  • multiple antennas 32 may be respectively arranged in the storage room 111 , the storage room 112 and the storage room 113 , and the number of antennas 32 in each storage room may be the same or different.
  • the number of antennas 32 in each storage room may be the same or different.
  • FIG. 3A when there are 8 antennas, 4 antennas may be installed in the storage room 111 , and 2 antennas may be installed in the storage room 112 and the storage room 112 respectively.
  • the read-write host 31 sends a radio frequency signal to each tag 20 by controlling the antenna 32, or the antenna 32 returns the radio frequency signal sent by each tag 20 to the read-write host 31, thereby realizing the read-write host 31 and multiple Communication between tags 20 .
  • the radio frequency signal emitted by the antenna 32 can activate the tag 20, and the activated tag 20 transmits the radio frequency signal through the antenna 32, and the data is read and collected by the read-write host 31, thereby obtaining the information of each food material in the refrigerator 1. related data.
  • Each tag 20 is configured to: receive the radio frequency signal sent by the reader-writer 30, obtain the signal strength information of the radio-frequency signal at the location of the tag 20 according to the radio-frequency signal, send the signal strength information to the reader-writer 30 and store in The ingredient information on the label 20.
  • the tag 20 receives radio frequency signals sent by each antenna 32 , and since the positions of the antennas 32 are different, the signal strength information of the radio frequency signals sent by each antenna 32 received by the tag 20 may also be different.
  • the tag 20 calculates the signal strength information of the radio frequency signal at the location of the tag 20 according to the radio frequency signals sent by multiple antennas 32 , and then sends the signal strength information and the food material information stored on the tag 20 to the read-write host 31 through the antenna 32 .
  • the read-write host 31 reads the signal strength information and food information of the tag 20 from the antenna 32 .
  • Labels 20 are related to information such as the humidity of their respective environments. For example, the farther the distance between the antenna 32 and the tag 20, the weaker the signal strength of the radio frequency signal sent by the antenna 32 at the position of the tag 20; For a long time, the signal strength of the radio frequency signal sent by the antenna 32 is weaker at the position of the tag 20; The signal strength becomes weaker.
  • the signal strength information of the radio frequency signal received by the tag 20 from each antenna 32 may be different, the signal strength information may reflect the location information of the tag 20 .
  • the location information of the label 20 is determined, the location information of the food material corresponding to the label 20 is also determined.
  • the controller 40 is coupled to the reader-writer 30, and is configured to: in response to an ingredient positioning instruction of a target ingredient, obtain from the reader-writer 30 the signal strength information of the target tag corresponding to the ingredient information of the target ingredient; The signal strength information of the tag is used to obtain the location information of the target ingredients by using the positioning model.
  • the ingredient positioning instruction is a condition for the controller 40 to locate the ingredient, and the controller 40 can only locate the target ingredient if the positioning condition is met.
  • the ingredient positioning instruction may be triggered by the outside world.
  • the user can input an instruction for locating food about a certain food through the display 50 of the refrigerator 1; A material positioning command about a certain food material is sent to the refrigerator 1 .
  • the locating instruction may also be a food locating instruction automatically triggered by the refrigerator 1 .
  • the controller 40 automatically triggers an ingredient positioning instruction after a preset positioning period.
  • the controller 40 can be set to trigger a food locating instruction every 2 days for sorting out the position information of the food in the refrigerator 1 .
  • the controller 40 acquires the signal strength information of the target tag corresponding to the ingredient information of the target ingredient from the read-write host 31 in response to an ingredient location instruction of the target ingredient.
  • a target ingredient refers to any ingredient that needs to be located.
  • a target label i.e. any label 20
  • the controller 40 responds to the ingredient location command of the target ingredient, identifies the target tag corresponding to the ingredient information from the read-write host 31 according to the ingredient information in the ingredient location instruction, and acquires the signal strength information of the target tag.
  • the eight antennas in the refrigerator 1 simultaneously send out radio frequency signals.
  • the target tag 21 can receive the radio frequency signals from the 8 antennas, and calculate the signal strength information of the 8 antennas at the location of the target tag 21 .
  • the target tag 21 will obtain the signal strength information of each antenna, and send the signal strength information and the food material information on the target tag 21 back to the read-write host 31 through the eight antennas.
  • each antenna 32 is configured to: send radio frequency signals to a plurality of tags 20, and receive corresponding signal strength information and ingredient information from a plurality of tags 20; the read-write host 31 is configured to: obtain information from a plurality of Signal strength information and ingredient information corresponding to multiple tags 20 of the antenna 32 .
  • the read-write host 31 obtains the signal strength information and food information of a plurality of tags 20 in the refrigerator 1, therefore, when the controller 40 acquires the signal strength information of the target tag from the reader-writer 30, the read-write host 31 It is necessary to find the target tag 21 among multiple tags 20 according to the ingredient information of the target ingredient, and the controller 40 acquires the signal strength information corresponding to the target tag 21 from the read-write host.
  • the controller 40 After the controller 40 acquires the signal strength information of the target tag 21, it inputs the signal strength information into the positioning model, and the positioning model can output the position information of the target tag (ie, the target ingredient).
  • the location model is pre-trained based on signal strength information of multiple tags and location information of multiple tags.
  • the server 2 is configured to: determine at least two tags to be trained among the tags 20 in each of the plurality of refrigerators 1, and the at least two tags to be trained are distributed at preset intervals. In at least one storage room 11 of the refrigerator 1.
  • the preset interval between two adjacent labels to be trained may be the same.
  • the space of the storage room 11 may be divided into multiple areas of the same size and size, adjacent areas border each other or have the same interval, and each label to be trained is set in one area. In this way, the distance between adjacent labels to be trained can be reduced. For example, the distance between any adjacent labels to be trained is basically the same.
  • the space of the storage room 111 can be divided into 9 areas of equal size, and a label 22 to be trained is placed in the center of each area.
  • the location information of multiple areas divided in the storage room 111 can be recorded as P1-0, P1-1, ..., P1-n respectively, where n is the number of areas divided by the storage room 111, and n is An integer greater than or equal to 1.
  • Place n to-be-trained labels 22 in n areas respectively, and the position information of the n areas is respectively the position information of the first to n-th to-be-trained labels.
  • the server 2 is configured to: acquire a training sample set, the training sample set includes signal strength information corresponding to at least two tags to be trained and location information of at least two tags to be trained; according to the training sample set, Perform model training on the initial positioning model to obtain a positioning model.
  • All the antennas 32 in each storage room 11 of the refrigerator 1 can send radio frequency signals to the tags 22 to be trained.
  • the eight antennas in the storage room 111 , the storage room 112 and the storage room 113 simultaneously send radio frequency signals to the tags 22 to be trained.
  • each tag 22 to be trained receives the radio frequency signal sent by each antenna 32, it calculates the signal strength information of each radio frequency signal at its respective position, and passes the signal strength information and the food material information stored on the tag 22 to be trained through all The antenna 32 returns to the read-write host 31.
  • the read-write host 31 obtains the signal strength information of each tag 22 to be trained and the food information stored on each tag 22 to be trained, and according to the food information stored on the tag 22 to be trained, the signals belonging to the same tag 22 to be trained.
  • the intensity information is sent to the controller 40 .
  • the controller 40 will also send the position information P1-i (0 ⁇ i ⁇ n, i is an integer) corresponding to the label 22 to be trained and all the signal strength information corresponding to the label 22 to be trained to the server 2 as the initial positioning model training training sample set.
  • the server 2 can collect multiple data on the signal strength information of the same tag to be trained, and each data can be used as training data for the positioning model to improve the accuracy of the positioning model training.
  • the initial positioning model can be realized by Convolutional Neural Networks (CNN).
  • CNN Convolutional Neural Networks
  • the initial positioning model can be implemented by a lightweight convolutional neural network, which can perform autonomous learning and can extract the most effective features of the location information of the label to be trained for representation.
  • the initial localization model (and the localization model) include a zero-filling layer, at least one Inception structure layer, a pooling layer, and a fully connected layer.
  • the Inception structure layer is composed of multiple convolutions, and at least one Inception structure layer may include two Inception structure layers, that is, a first Inception structure layer and a second Inception structure layer.
  • Fig. 4 is a schematic diagram of a positioning model provided by an embodiment of the present disclosure.
  • the positioning model includes a first input layer 41, which is connected with a first zero-value filling layer 42 after the first input layer 41, and is connected with a first Inception structure 43 after the first zero-value filling layer 42, and the first Inception structure 43 is connected to the second Inception structure 44, the second Inception structure 44 is connected to the pooling layer 45, and the pooling layer is connected to the fully connected layer 46.
  • first zero-fill layer 42 may be implemented by general convolution.
  • the general convolution is used to zero-fill the edge of the feature matrix to increase the height and width of the feature matrix.
  • the first zero-filling layer 42 can also be connected to a convolutional layer, which can be a convolution including a plurality of 1 ⁇ 1 convolution kernels (which can be referred to as 1 ⁇ 1 convolution ), the convolutional layer is used to increase the dimension of the feature matrix obtained by general convolution, so as to increase the number of channels of the feature matrix.
  • the first Inception structure 43 and the second Inception structure 44 may be two Inception structures with the same structure, wherein each Inception structure may be composed of four branches.
  • the first branch of each Inception structure performs a 1 ⁇ 1 convolution operation on the feature matrix; the second branch first performs a 1 ⁇ 1 convolution on the feature matrix, and then uses 3 ⁇ 3 convolution performs convolution operation on the feature matrix after 1 ⁇ 1 convolution processing; the third branch first performs 1 ⁇ 1 convolution on the feature matrix, and then uses 5 ⁇ 5 convolution to perform 1 ⁇ 1 convolution processing The feature matrix is convoluted; the fourth branch first performs 3 ⁇ 3 maximum pooling downsampling on the feature matrix, and then uses 1 ⁇ 1 convolution to perform convolution operation on the feature matrix after 3 ⁇ 3 maximum pooling.
  • the step size of each operation in the four branches of the Inception structure may be 1.
  • the feature matrix output by each branch has the same size, that is, the same height and width (the depth of the feature matrix output by each branch can be different), therefore, the feature matrix output by each branch can be along the depth direction to overlay.
  • a first depth concat (DepthConcat) layer 47 is also included between the first Inception structure 43 and the second Inception structure 44, and the first depth concat layer 47 is used to respectively output the four branches of the first Inception structure 43
  • the feature matrix is spliced according to the depth direction to obtain a complete feature matrix.
  • a second depth superposition layer 48 may also be connected after the second Inception structure 44, and the second depth superposition layer 48 is used to stitch the feature matrices respectively output by the four branches of the second Inception structure 44 according to the depth direction, to get another complete feature matrix.
  • the complete feature matrix output through the second depth stacking layer 48 is input to the pooling layer 45, and the pooling layer 45 performs a global average pooling down-sampling operation on the feature matrix.
  • the feature matrix processed by the pooling layer 45 is input to the fully connected layer 46, and the fully connected layer 46 performs full connection to the feature matrix output by each partition of the refrigerator 1.
  • the three partitions in the refrigerator 1 are The training data is trained to obtain a feature matrix, and after being processed by the fully connected layer 46, a one-dimensional feature matrix is obtained.
  • the one-dimensional feature matrix may correspond to position information of a label to be trained.
  • the server 2 After completing the construction of the initial positioning model, the server 2 uses the obtained training sample set to train the initial positioning model.
  • the server 2 is further configured to: obtain the initial positioning model training data according to the signal strength information corresponding to each label to be trained.
  • the server 2 uses the signal strength information of the label to be trained as the training data of the initial positioning model, for example, may input the signal strength information of the same label to be trained as the training data into the initial positioning model.
  • the signal strength information of 8 antennas at the position of each training tag 22 may be input into the initial positioning model as training data.
  • the server 2 before using the training data to train the initial positioning model, is further configured to: preprocess the training data. Preprocessing may include normalization and dimensionality enhancement.
  • the normalization processing of the training data by the server 2 may include mapping the training data to an interval from 0 to 1, for example, mapping the signal strength information of the 8 antennas at each position of the training label 22 to [0, 1] a certain value in the interval. Normalization processing can improve the convergence speed of the positioning model and shorten the time required for training.
  • the server 2 performs dimension-increasing processing on the normalized training data.
  • one-dimensional training data can be expanded to two dimensions to generate two-dimensional data.
  • the server 2 can save the two-dimensional data in a comma-separated value (Comma-Separated Values, CSV) file, and each line of data in the CSV file can be regarded as a feature map, and the CSV file The number of rows can be used as the number of feature maps.
  • CSV comma-separated Values
  • the server 2 can convert the training data into the form of a feature map to input into the initial positioning model for training, and represent the signal strength information in the form of a feature map, which is more convenient for the identification of the initial positioning model.
  • the training process of the initial positioning model shown in FIG. 4 is exemplarily described by taking a to-be-trained label including 8 signal strength information as an example. It should be noted that the training data of each label to be trained can be trained using the following steps.
  • the preprocessed training data is input to the first input layer 41, and the first input layer 41 converts the preprocessed training data into a feature matrix with a size of 8 ⁇ 1 ⁇ 1 to facilitate subsequent convolution operations .
  • the data processed by the first input layer 41 is called the first feature matrix.
  • the first feature matrix is input to the first zero-value filling layer 42, and the first zero-value filling layer 42 uses general convolution to fill the edge of the first feature matrix with zero values to increase the first feature matrix height and width.
  • the upper, lower, left, and right sides of the first feature matrix with a size of 8 ⁇ 1 ⁇ 1 can be filled with zero values in two rows and two columns respectively, so as to obtain a feature matrix with a size of 12 ⁇ 5 ⁇ 1.
  • the second feature matrix with a size of 12 ⁇ 5 ⁇ 6 is obtained after the feature matrix with a size of 12 ⁇ 5 ⁇ 1 is subjected to convolution operations with six convolution kernels with a size of 1 ⁇ 1.
  • the first zero-value filling layer 42 increases the size and number of channels of the first feature matrix.
  • the second feature matrix is input into the first Inception structure 43, and the second feature matrix is processed by four branches of the first Inception structure 43 respectively.
  • the first branch is processed by 1 ⁇ 1 convolution, it outputs a feature matrix of size 12 ⁇ 5 ⁇ 6
  • the second branch is processed by 1 ⁇ 1 convolution kernel and 3 ⁇ 3 convolution, it outputs a size of 12 ⁇ 5 ⁇ 6 feature matrix
  • the third branch is processed by 1 ⁇ 1 convolution kernel 5 ⁇ 5 convolution, and outputs a feature matrix of size 12 ⁇ 5 ⁇ 6
  • the fourth branch is processed by 3 ⁇ 3 maximum pooling and 1 ⁇ 1 convolution
  • a feature matrix of size 12 ⁇ 5 ⁇ 6 is output.
  • the feature matrices output by the four branches are input to the first depth superposition layer 47, and the first depth superposition layer 47 stitches the four feature matrices in the depth direction to obtain the third feature matrix.
  • the third feature matrix is input to the second Inception structure 44, and the processing process of the second Inception structure 44 is similar to that of the first Inception structure 43, and will not be repeated here.
  • the output feature matrices of the four branches of the second Inception structure 44 are input to the second depth superposition layer 48, and the second depth superposition layer 47 splices the feature matrices output by the four branches according to the depth direction to obtain a size of 12 ⁇ 5 ⁇ 48
  • the fourth characteristic matrix of is input to the second Inception structure 44, and the processing process of the second Inception structure 44 is similar to that of the first Inception structure 43, and will not be repeated here.
  • the output feature matrices of the four branches of the second Inception structure 44 are input to the second depth superposition layer 48, and the second depth superposition layer 47 splices the feature matrices output by the four branches according to the depth direction to obtain a size of 12 ⁇ 5 ⁇ 48
  • the fourth characteristic matrix of is input to the second Inception structure 44, and the processing process
  • the fourth feature matrix is input into the pooling layer 45, and the pooling layer 46 uses 3 ⁇ 1 average pooling on the fourth feature matrix to perform global average pooling downsampling on the fourth feature matrix to obtain 6 ⁇ A fifth characteristic matrix of size 2 ⁇ 48.
  • the sixth step is to input the fifth feature matrix output after training the training data of the labels to be trained in the three partitions of the refrigerator 1 to the fully connected layer 46 for fully connected, and output the training result.
  • the training result corresponds to the location information of the label to be trained.
  • the initial positioning model also includes a classifier (SoftMax), which maps the output of the fully connected layer to the [0,1] interval, thereby achieving multi-classification, and further obtaining the partition corresponding to the label to be trained and The specific location information located in the partition.
  • SoftMax classifier
  • the initial positioning model may also include a zero-filling layer, at least one convolutional layer, at least one pooling layer, at least one fully connected layer, and a classifier layer.
  • at least one convolutional layer may include two convolutional layers
  • at least one pooling layer may include two pooling layers
  • at least one fully connected layer may include two fully connected layers.
  • Fig. 5 is a schematic diagram of the structure of another positioning model provided by an embodiment of the present disclosure.
  • the initial localization model (and localization model) may also include a second input layer 51, a second zero-filling layer 52, a first convolutional layer 53, a first pooling layer 54, a second convolutional layer 55, the second pooling layer 56, the first fully connected layer 57, the second fully connected layer 58 and the classifier 59.
  • the convolutional layer ie, the first convolutional layer 53 and the second convolutional layer 55
  • the convolutional layer can be composed of three parts: convolution operation (Convolution), batch normalization (Batch-Normalization) and activation function Leaky ReLU.
  • Convolution convolution
  • Batch-Normalization batch normalization
  • activation function Leaky ReLU activation function Leaky ReLU
  • the convolution kernel is composed of initialization weights and biases, and the convolution operation is implemented by the convolution kernel in the form of a sliding window.
  • the convolution operation can be implemented by formula (1):
  • mi is the feature map after the convolution operation
  • xi :i+w-1 is the data in the sliding window corresponding to the convolution kernel
  • W is the weight information
  • b is a constant with a small value.
  • the data can be mapped through batch normalization, and the data x and the corresponding mapped data Satisfy the following formula (2):
  • ⁇ r and The mean and variance of all data x from statistics, ⁇ is a small number set to prevent the error of dividing by 0, and in the embodiment of the disclosure, ⁇ is set to The randomly distributed position data is converted into data according to the normal distribution, so that the data distribution of the input network is closer, which is more conducive to the iterative optimization of the network.
  • the embodiment of the present disclosure adopts the Leaky ReLU activation function to overcome the phenomenon of diffuse gradient, and the activation function Leaky ReLU satisfies the relationship described in formula (3):
  • p can be set to 0.04 to ensure that the gradient is not zero in the negative region.
  • the pooling layers can sample or aggregate information on a locally related set of data.
  • the pooling layer may be a max pooling layer, namely a first max pooling layer and a second max pooling layer.
  • max pooling can select the largest one among locally related data.
  • the fully connected layer ie, the first fully connected layer and the second fully connected layer
  • the server 2 before the server 2 uses the positioning model shown in FIG. 5 for training, it needs to perform preprocessing on the data to be trained, and the preprocessing includes normalization processing and dimension increasing processing.
  • the normalization processing is similar to the normalization processing in the positioning model shown in FIG. 4 , and will not be repeated here.
  • the normalized one-dimensional training data can be trained and transformed into three-dimensional data, and the three-dimensional training data can be one-hot encoded. Then input the preprocessed training data into the localization model shown in Figure 5 for model training.
  • FIG. 5 and FIG. 6 taking a training label in FIG. 3B including 8 pieces of signal strength information as an example, the training process of the initial positioning model shown in FIG. 5 is exemplarily described.
  • the preprocessed training data is input into the second input layer 51 of the positioning model to obtain a feature matrix with a size of 8 ⁇ 1 ⁇ 1. It should be noted that the processing process of the second input layer 51 is the same as that of the first input layer 41 , and will not be repeated here.
  • the feature matrix with a size of 8 ⁇ 1 ⁇ 1 is input to the second zero-value filling layer 52 to perform edge zero-value filling processing.
  • the upper, lower, left, and right of the feature matrix can be filled with zero values in one row and one column respectively to obtain a feature matrix with a size of 10 ⁇ 3 ⁇ 1 as shown in FIG. 6 .
  • the feature matrix of size 10 ⁇ 3 ⁇ 1 is input to the first convolutional layer 53, for example, after convolution operation of 32 convolution kernels of size 8 ⁇ 1, a feature matrix of size 9 ⁇ 2 ⁇ 32 is obtained Matrix; after the first pooling layer 54, for example, after 3 ⁇ 1 maximum pooling, the feature matrix with a size of 9 ⁇ 2 ⁇ 32 is converted into a feature matrix with a size of 4 ⁇ 1 ⁇ 32.
  • the fourth step is to input a feature matrix of size 4 ⁇ 1 ⁇ 32 to the second convolutional layer 55, for example, after convolution operation of ten convolution kernels of size 4 ⁇ 1, a feature matrix of size 4 ⁇ 1 ⁇ 10 is obtained Matrix; after being processed by the second pooling layer 56, for example, after 3 ⁇ 1 maximum pooling, a feature matrix with a size of 4 ⁇ 1 ⁇ 10 as shown in FIG. 6 is obtained.
  • the fifth step is to input the 4 ⁇ 1 ⁇ 10 feature matrix obtained by the second maximum pooling layer into the first fully connected layer 57 to obtain a 40 ⁇ 1 two-dimensional feature matrix, and then pass through the second fully connected layer 58, A 3 ⁇ 1 two-dimensional feature matrix as shown in Figure 6 is obtained.
  • the sixth step is to classify the feature matrix with a size of 3 ⁇ 1 through the classifier 59 to obtain the training result.
  • the positioning model is trained through a large amount of training data, and the large amount of training data includes signal strength information of multiple tags collected multiple times.
  • the larger the amount of training data and the more times of training the higher the accuracy of the obtained positioning model.
  • the server 2 may use test data to test the obtained positioning model.
  • the test data may be signal strength information of any tag in multiple refrigerators 1 .
  • the server 2 sends the trained positioning model to each refrigerator 1, and the refrigerator 1 stores the positioning model after acquiring the positioning model, and uses the positioning model to locate the ingredients.
  • the controller 40 inputs the acquired signal strength information of the target tag into the positioning model, and the positioning model is similar to the training process of the positioning model in the above-mentioned embodiments (such as the positioning model of FIG. 4 or FIG. 5 ). Steps (not described in detail here), after processing the signal strength information, output the location information of the target label (ie, the target ingredient). In other examples, after the controller 40 inputs the acquired signal strength information of the target tag into the positioning model, the positioning model may directly output the position information of the target tag (ie, the target ingredient) corresponding to the signal strength information.
  • the refrigerator 1 pre-stores the positioning model sent by the server 2, and uses the positioning model to obtain the position information of the tag 20 through the signal strength of the tag 20 corresponding to the same ingredient information.
  • the position information is the specific location information of the food corresponding to the label 20 in the storage room 11, and the location information can facilitate the user to obtain the storage location of the food more quickly.
  • the refrigerator management system 3 provided by the embodiment of the present disclosure has the same beneficial effects as the refrigerator 1 .
  • the server 2 can complete the construction of the initial positioning model, and use the training sample set to train the initial positioning model to obtain the positioning model, and send the positioning model to multiple refrigerators 1. It is used to locate the ingredients in the refrigerator 1. This positioning model can more accurately locate the ingredients.
  • the refrigerator 1 further includes a display 50 coupled to the controller 40 and configured to: acquire spatial information of multiple storage rooms; display a first interface, the first interface Including space information of a plurality of storage rooms; receiving position information of target ingredients from the controller 40; displaying a second interface, the second interface including position information of the target ingredients in the space of the corresponding storage room.
  • a display 50 coupled to the controller 40 and configured to: acquire spatial information of multiple storage rooms; display a first interface, the first interface Including space information of a plurality of storage rooms; receiving position information of target ingredients from the controller 40; displaying a second interface, the second interface including position information of the target ingredients in the space of the corresponding storage room.
  • the controller 40 can send the location information of the target ingredient obtained in the positioning model to the display 50, and the display 50 displays the location corresponding to the location information, making it easier for the user to obtain the specific location of the target ingredient.
  • the space information of each storage room 11 of the refrigerator 1 may be pre-stored in the display 50, and the space information may be displayed through the first interface.
  • This spatial information includes a spatial distribution map or spatial distribution data of each storage room 11 .
  • the display 50 may respectively display the storage room 111 , the storage room 112 , and the three-dimensional model diagram or three-dimensional dimension value corresponding to the storage room 112 .
  • FIG. 7A is a schematic diagram of a first interface displayed on a display 50 according to an embodiment of the present disclosure. As shown in FIG. 7A , the display 50 displays a three-dimensional model corresponding to the storage room 111 .
  • the display 50 After the display 50 receives the location information of the target ingredient sent from the controller 40, it can display a second interface, the second interface includes marking the location of the target ingredient in the three-dimensional model diagram of the storage room corresponding to the location information of the target ingredient.
  • FIG. 7B is a schematic diagram of a second interface displayed by the display 50 according to an embodiment of the present disclosure. As shown in FIG. 7B , the display 50 displays the marks of the positions of the target ingredients in the corresponding three-dimensional model diagram of the storage room 111 .
  • the location information when the location information includes at least two locations, different marking methods can be used to mark, so that the user can clearly obtain a certain location of the storage room where the target ingredient is located.
  • the controller 40 is further configured to: after obtaining the location information of the target ingredient, determine the accuracy rate of the location information of the target ingredient; within a preset time, use a positioning model to obtain the location of various target ingredients information; determine the accuracy of the positioning model according to the accuracy of the location information of various target ingredients.
  • the display is further configured to: display a third interface, where the third interface includes the accuracy rate of the location information of the target ingredient.
  • the controller 40 After the controller 40 obtains the location information of the target ingredient, it can further obtain the accuracy rate of the location information. In some examples, the controller 40 can obtain the accuracy rate of the location information of the target ingredient through the display 50 .
  • the display 50 can also display a feedback window, which is used to provide the user with an option to select whether the position is accurate.
  • the accuracy rate of the position information is obtained, and the controller 40 obtains the accuracy rate of the position information.
  • FIG. 7C is a schematic diagram of a third interface displayed on a display 50 according to an embodiment of the present disclosure.
  • the display 50 displays a feedback window, and the feedback window includes an option indicating whether the location information selected by the user is accurate or not , the controller 40 can obtain the accuracy rate of the location information through the user's selection of yes or no. For example, when the user selects yes, the controller 40 may acquire the accuracy rate of this positioning as 1, and when the user selects no, the controller 40 may acquire the accuracy rate of this positioning as 0.
  • the display 50 may display the third interface every time the controller 40 finishes positioning the target ingredient.
  • the controller 40 may locate multiple target ingredients within a preset time, such as a first preset time, and acquire location information of the multiple target ingredients.
  • a preset time such as a first preset time
  • the embodiment of the present disclosure does not limit the first preset time.
  • the first preset time may be 7 days or 30 days.
  • the multiple target ingredients may be the ingredients queried by the user within the first preset time period, or the ingredients whose accuracy rate has been reported by the user.
  • the controller 40 further obtains the accuracy rate of the positioning model according to the accuracy rate of the position information of each target ingredient.
  • the accuracy rate of the positioning model may be an average value of the accuracy rates of the location information of the multiple target ingredients.
  • the accuracy of the positioning model can reflect the accuracy of the positioning model for the positioning of ingredients.
  • the controller 40 is further configured to: receive a model update parameter from the server 2; update the positioning model according to the model update parameter to obtain an updated positioning model.
  • the updated location model is used to obtain updated location information of a target ingredient.
  • the positioning model may also be updated after a preset time, for example, a second preset time.
  • a preset time for example, a second preset time.
  • the second preset time and the first preset time may be the same or different.
  • the second preset time is 5 days or 7 days.
  • the server 2 trains the stored positioning model with the training data obtained from each refrigerator 1 associated with it to obtain an updated positioning model, thereby obtaining parameters of the updated positioning model. And, every second preset time, the server 2 will send the model update parameters to each refrigerator 1 .
  • the refrigerator 1 After the refrigerator 1 receives the model updating parameters sent by the server 2, it updates the positioning model according to the model updating parameters, updates the stored positioning models, and obtains the updated positioning models.
  • the updated location model is used to obtain the updated location information of various target ingredients; according to the accuracy rate of the updated location information of the various target ingredients, determine The accuracy of the updated positioning model; determine the relationship between the accuracy of the updated positioning model and the accuracy of the positioning model; if the accuracy of the updated positioning model is greater than or equal to the accuracy of the positioning model, keep the updated positioning model; if the accuracy rate of the updated positioning model is lower than the accuracy rate of the positioning model, the positioning model is obtained from the server 2 again.
  • the multiple target ingredients positioned using the updated positioning model may be the same as or different from the multiple target ingredients positioned using the positioning model in the above embodiment, and the multiple target ingredients may be refrigerator 1 Any ingredient in , or any ingredient that the user needs to locate.
  • the preset positioning model in the refrigerator 1 is the first positioning model
  • the updated positioning model is the second positioning model using model update parameters.
  • the refrigerator 1 obtains the accuracy rate of the second positioning model
  • the The accuracy rate of the second positioning model is compared with the accuracy rate of the first positioning model. It should be noted that the accuracy rate of obtaining the second positioning model by the refrigerator 1 is similar to the process of obtaining the accuracy rate of the first positioning model by the refrigerator 1 in the above implementation, which will not be repeated here.
  • the refrigerator 1 can retain the updated positioning model (that is, the second positioning model), and use the second positioning model to continue to obtain the position of the target ingredient information.
  • the refrigerator 1 When the accuracy rate of the second positioning model is lower than the accuracy rate of the first positioning model, the refrigerator 1 will reacquire the first positioning model from the server 2, and use the first positioning model to continue to obtain the position information of the target food item. In some examples, the refrigerator 1 may delete the second positioning model after reacquiring the first positioning model.
  • the refrigerator 1 uses the server 2 to continuously acquire updated positioning models to locate the ingredients, which can ensure that the accuracy of the positioning model in the refrigerator 1 remains at a high level, and further improve the accuracy of ingredient positioning.
  • Fig. 8 is a method for controlling a refrigerator management system provided by an embodiment of the present disclosure.
  • the refrigerator management system may be the refrigerator management system 3 in the above-mentioned embodiment.
  • the refrigerator management system includes multiple refrigerators and servers, and the multiple refrigerators may be A plurality of refrigerators 1 in the above-mentioned embodiment.
  • the refrigerator includes multiple tags, readers, controllers and cabinets, as shown in Figure 8, the control method of the refrigerator management system may include the following steps:
  • Step 81 the controller 40 acquires the signal strength information of the target tag corresponding to the ingredient information of the target ingredient from the reader 30 in response to an ingredient location instruction of a target ingredient.
  • Step 82 the controller 40 obtains the location information of the target food material by using a positioning model according to the signal strength information of the target tag.
  • the positioning model is pre-trained based on signal strength information of multiple tags 20 and location information of multiple tags 20 .
  • the controller 40 after the controller 40 obtains the location information of the target ingredient, it determines the accuracy rate of the location information of the target ingredient; within a preset time, the controller 40 uses the positioning model to obtain the locations of various target ingredients Information; the controller 40 determines the accuracy of the positioning model according to the accuracy of the location information of various target ingredients.
  • Fig. 9 provides another control method of a refrigerator management system according to an embodiment of the present disclosure. As shown in Fig. 9, the control method of the refrigerator management system may include the following steps:
  • Step 91 the server 2 sends the model update parameters to the controller 40 .
  • Step 92 the controller 40 receives the model update parameters from the server 2 .
  • Step 93 the controller 40 updates the positioning model according to the model update parameters, and obtains the updated positioning model.
  • step 94 the controller 40 uses the updated location model to obtain updated location information of a target ingredient.
  • Step 95 within a preset time, the controller 40 uses the updated positioning model to obtain updated position information of various target ingredients.
  • Step 96 the controller 40 determines the accuracy rate of the updated positioning model according to the accuracy rate of the updated position information of various target ingredients.
  • step 97 the controller 40 judges whether the accuracy rate of the updated positioning model is lower than the accuracy rate of the positioning model.
  • step 98 If the accuracy rate of the updated positioning model is less than the accuracy rate of the positioning model, perform step 98; if the accuracy rate of the updated positioning model is greater than or equal to the accuracy rate of the positioning model, perform step 99.
  • Step 98 the controller 40 acquires the positioning model from the server 2 .
  • step 99 the controller 40 retains the updated positioning model.
  • control method of the refrigerator management system further includes: each antenna 32 sends a radio frequency signal to a plurality of tags 20, and receives corresponding signal strength information and ingredient information from a plurality of tags 20; Signal strength information and ingredient information corresponding to multiple tags 20 of multiple antennas 32 .
  • control method of the refrigerator management system further includes: the display 50 acquires the spatial information of the multiple storage rooms 11; the display 50 displays a first interface, and the first interface includes the spatial information of the multiple storage rooms 11; the display 50 Receive the location information of the target ingredient sent from the controller 40 ; display a second interface, the second interface includes the location information of the target ingredient in the corresponding space in the storage room 11 .
  • control method of the refrigerator management system further includes: the display 50 displays a third interface, where the third interface includes the accuracy rate of the location information of the target ingredients.
  • the server 2 determines at least two tags to be trained among the plurality of tags 20 in each refrigerator 1, and the at least two tags to be trained are distributed in at least one storage compartment 11 of the refrigerator 1 at preset intervals;
  • the refrigerator 1 is one of multiple refrigerators coupled to the server 2;
  • the control method of the refrigerator management system further includes: the server 2 obtains a training sample set, and the training sample set includes signal strength information corresponding to at least two tags to be trained and the location information of at least two labels to be trained; the server 2 performs model training on the initial positioning model according to the training sample set to obtain the positioning model; the server 2 sends the positioning model to multiple refrigerators.
  • the initial positioning model includes a zero-filling layer, at least one Inception structure layer, a pooling layer, and a fully connected layer;
  • the server 2 performs model training on the initial positioning model according to the training sample set, including: the server 2 performs model training on the initial positioning model according to each The signal strength information corresponding to each label to be trained is obtained to obtain the training data of the initial positioning model; the server 2 preprocesses the training data; the server 2 sequentially inputs the preprocessed training data into the zero value filling layer and at least one Inception structure Layer, pooling layer and fully connected layer to obtain the processing results of the training data; where the processing results of the training data correspond to the position information of the labels to be trained.
  • the server 2 performs preprocessing on the training data, including: the server 2 performs normalization processing on the training data; the server 2 performs dimension increasing processing on the normalized training data.
  • Some embodiments of the present disclosure provide a computer-readable storage medium (for example, a non-transitory computer-readable storage medium), on which a computer program is stored, and when the computer program is executed by a refrigerator management system, the refrigerator management system executes The control method of the refrigerator management system as described in any one of the above-mentioned embodiments.
  • a computer-readable storage medium for example, a non-transitory computer-readable storage medium
  • the above-mentioned computer-readable storage medium may include, but is not limited to: a magnetic storage device (for example, a hard disk, a floppy disk, or a magnetic tape, etc.), an optical disk (for example, a CD (Compact Disk, a compact disk), a DVD (Digital Versatile Disk, Digital Versatile Disk), etc.), smart cards and flash memory devices (for example, EPROM (Erasable Programmable Read-Only Memory, Erasable Programmable Read-Only Memory), card, stick or key drive, etc.).
  • Various computer-readable storage media described in this disclosure can represent one or more devices and/or other machine-readable storage media for storing information.
  • Some embodiments of the present disclosure also provide a computer program product.
  • the computer program product comprises a computer program stored on a non-transitory computer readable storage medium. Wherein, when the computer program is executed by the refrigerator management system, the refrigerator management system executes the control method of the refrigerator management system as described in the above-mentioned embodiments.
  • Some embodiments of the present disclosure also provide a computer program.
  • the computer program is stored on a non-transitory computer readable storage medium.
  • the refrigerator management system is made to execute the control method of the refrigerator management system as described in the above-mentioned embodiments.

Landscapes

  • Engineering & Computer Science (AREA)
  • Chemical & Material Sciences (AREA)
  • Combustion & Propulsion (AREA)
  • Physics & Mathematics (AREA)
  • Mechanical Engineering (AREA)
  • Thermal Sciences (AREA)
  • General Engineering & Computer Science (AREA)
  • Cold Air Circulating Systems And Constructional Details In Refrigerators (AREA)

Abstract

一种冰箱，包括：箱体、读写器、多个标签和控制器。箱体包括多个储藏室。读写器设于箱体内，被配置为向多个标签发送射频信号。多个标签位于多个储藏室中，每个标签存储有至少一种食材的食材信息，每个标签被配置为接收所述射频信号，根据射频信号，得到射频信号在标签所在位置处的信号强度信息，向读写器发送信号强度信息和食材信息。控制器与所述读写器耦接，且被配置为响应于一种目标食材的食材定位指令，从读写器中获取目标食材的食材信息对应的目标标签的信号强度信息；根据目标标签的信号强度信息，采用定位模型，得到目标食材的位置信息；其中，定位模型是基于多个标签的信号强度信息和多个标签的位置信息预先训练得到的。

Description

冰箱、冰箱管理系统及冰箱管理系统的控制方法
本申请要求于2021年10月18日提交的、申请号为202111212660.2的中国专利申请的优先权，以及于2021年10月18日提交的、申请号为202111207745.1的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本公开涉及冰箱技术领域，尤其涉及一种冰箱、冰箱管理系统及冰箱管理系统的控制方法。
背景技术
冰箱作为家庭必备家电之一，随着智能化的发展，能够为用户提供更加方便、快捷的服务。例如，随着生活节奏的加快，用户会经常忘记冰箱内存放的食材的位置，需要在冰箱中翻找，使用很不方便。因此，带有食材管理功能的冰箱被越来越多的消费者选择。
发明内容
一方面，提供一种冰箱。该冰箱包括多个标签、读写器、控制器和箱体。该箱体包括多个储藏室。该读写器，设置于箱体内，被配置为：向多个标签发送射频信号。该多个标签，位于多个储藏室中，每个标签存储有至少一种食材的食材信息，该标签被配置为：接收射频信号，根据射频信号，得到该射频信号在该标签所在位置处的信号强度信息，向读写器发送信号强度信息和食材信息、控制器与读写器耦接，控制器被配置为：响应于一种目标食材的食材定位指令，从读写器中获取目标食材的食材信息对应的目标标签的信号强度信息；根据目标标签的信号强度信息，采用定位模型，得到目标食材的位置信息；其中，定位模型是基于多个标签的信号强度信息和多个标签的位置信息预先训练得到的。
另一方面，提供一种冰箱管理系统。该冰箱管理系统包括多个冰箱以及服务器，该服务器与多个冰箱耦接，且被配置为：向多个冰箱提供定位模型。
又一方面，提供一种冰箱管理系统的控制方法。该冰箱管理系统包括多个冰箱和服务器。多个冰箱中的每个冰箱包括多个标签、读写器、控制器和箱体。箱体包括多个储藏室。读写器设置于箱体内；多个标签位于多个储藏室中，多个标签的每个标签存储有至少一种食材的食材信息，控制器与读写器耦接。该冰箱管理系统的控制方法包括：控制器响应于一种目标食材的食材定位指令，从读写器中获取目标食材的食材信息对应的目标标签的信号强度信息。控制器根据目标标签的信号强度信息，采用定位模型，得到目标食材的位置信息；其中，定位模型是基于多个标签的信号强度信息和多个标签的位置信息预先训练得到的。
再一方面，提供一种非暂态计算机可读存储介质，其上存储有计算机程序，所述计算机程序被冰箱管理系统执行时，使得所述冰箱管理系统执行所述的冰箱管理系统的控制方法中的一个或多个步骤。
又一方面，提供一种计算机程序产品，包括计算机程序，其中，所述计算机程序在被冰箱管理系统执行时，使得所述冰箱管理系统执行所述的冰箱管理系统的控制方法中的一个或多个步骤。
附图说明
为了更清楚地说明本公开中的技术方案，下面将对本公开一些实施例中所需要使用的附图作简单地介绍，显然，下面描述中的附图仅仅是本公开的一些实施例的附图，对于本领域普通技术人员来讲，还可以根据这些附图获得其他的附图。此外，以下描述中的附图可以视作示意图，并非对本公开实施例所涉及的产品的实际尺寸、方法的实际流程、信号的实际时序等的限制。
图1为根据一些实施例的一种冰箱管理系统的示意图；
图2A为根据一些实施例的一种冰箱的结构图；
图2B为根据一些实施例的一种冰箱的示意图；
图3A为根据一些实施例的另一种冰箱的示意图；
图3B为根据一些实施例的又一种冰箱的示意图；
图4为根据一些实施例的一种定位模型的示意图；
图5为根据一些实施例的另一种定位模型的示意图；
图6为根据一些实施例的一种定位模型训练的示意图；
图7A为根据一些实施例的一种显示器的示意图；
图7B为根据一些实施例的另一种显示器的示意图；
图7C为根据一些实施例的又一种显示器的示意图；
图8为根据一些实施例的一种冰箱管理系统的控制方法的流程图；
图9为根据一些实施例的另一种冰箱管理系统的控制方法的流程图。
具体实施方式
下面将结合附图，对本公开一些实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本公开一部分实施例，而不是全部的实施例。基于本公开所提供的实施例，本领域普通技术人员所获得的所有其他实施例，都属于本公开保护的范围。
除非上下文另有要求，否则，在整个说明书和权利要求书中，术语“包括(comprise)”及其其他形式例如第三人称单数形式“包括(comprises)”和现在分词形式“包括(comprising)”被解释为开放、包含的意思，即为“包含，但不限于”。在说明书的描述中，术语“一个实施例(one embodiment)”、“一些实施例(some embodiments)”、“示例性实施例(exemplary embodiments)”、“示例(example)”、“特定示例(specific example)”或“一些示例(some examples)”等旨在表明与该实施例或示例相关的特定特征、结构、材料或特性包括在本公开的至少一个实施例或示例中。上述术语的示意性表示不一定是指同一实施例或示例。此外，所述的特定特征、结构、材料或特点可以以任何适当方式包括在任何一个或多个实施例或示例中。
以下，术语“第一”、“第二”仅用于描述目的，而不能理解为指示或暗示相对重要性或者隐含指明所指示的技术特征的数量。由此，限定有“第一”、“第二”的特征可以明示或者隐含地包括一个或者更多个该特征。在本公开实施例的描述中，除非另有说明，“多个”的含义是两个或两个以上。
在描述一些实施例时，可能使用了“耦接”和“连接”及其衍伸的表达。例如，描述一些实施例时可能使用了术语“连接”以表明两个或两个以上部件彼此间有直接物理接触或电接触。又如，描述一些实施例时可能使用了术语“耦接”以表明两个或两个以上部件有直接物理接触或电接触。然而，术语“耦接”或“通信耦合(communicatively coupled)”也可能指两个或两个以上部件彼此间并无直接接触，但仍彼此协作或相互作用。这里所公开的实施例并不必然限制于本文内容。
“A、B和C中的至少一个”与“A、B或C中的至少一个”具有相同含义，均包括以下A、B和C的组合：仅A，仅B，仅C，A和B的组合，A和C的组合，B和C的组合，及A、B和C的组合。“A和/或B”，包括以下三种组合：仅A，仅B，及A和B的组合。
如本文中所使用，根据上下文，术语“如果”任选地被解释为意思是“当……时”或“在……时”或“响应于确定”或“响应于检测到”。类似地，根据上下文，短语“如果确定……”或“如果检测到[所陈述的条件或事件]”任选地被解释为是指“在确定……时”或“响应于确定……”或“在检测到[所陈述的条件或事件]时”或“响应于检测到[所陈述的条件或事件]”。
本文中“适用于”或“被配置为”的使用意味着开放和包容性的语言，其不排除适用于或被配置为执行额外任务或步骤的设备。
另外，“基于”的使用意味着开放和包容性，因为“基于”一个或多个所述条件或值的过程、步骤、计算或其他动作在实践中可以基于额外条件或超出所述的值。
日常生活中，冰箱已成为每个家庭必备的家电之一，冰箱的食材定位功能被越来越多的用户所看重。通常，冰箱上设置有读写器，读写器通过接收附加在食材上的超高频射频识别(Radio Frequency Identification，RFID)标签的数据，能够预测到该食材在冰箱的位置分区，实现冰箱中食材的智能化管理。
然而，基于RFID标签识别到的食材的存放位置通常仅限于食材存放在冰箱的哪一个储藏室，并不能进一步识别到该食材位于该储藏室的具体位置。那么，当储藏室空间较大时，用户需要找该食材时，还需要在该储藏室内挨个翻找，费时又费力，还可能造成冰箱冷气外泄，引起耗电增高，从而降低了用户的体验度。
为此，本公开实施例提供一种冰箱管理系统。如图1所示，冰箱管理系统3包括多个冰箱1和服务器2。多个冰箱1分别与服务器2耦接。在一些示例中，服务器2可以为云端服务器。
服务器2被配置为向多个冰箱1提供定位模型。
示例性地，服务器2中预先存储有初始定位模型，服务器2可以对初始定位模型进行训练。在一些示例中，服务器2可以接收与其耦接的多个冰箱1发送的训练数据(例如，训练数据包含在下述的训练样本集中)，服务器2根据这多个冰箱1发送的训练数据，对其上预先存储的初始定位模型进行训练。并且，服务器2可以将训练好的定位模型发送给各冰箱1，以使各个冰箱1采用该定位模型对食材进行定位。
在一些示例中，与服务器2耦接的多个冰箱1可以为同一类型的冰箱，例如，该多个冰箱1的型号相同，或者，该多个冰箱1具有相同的功能。采用同一类型的多个冰箱1发送的训练数据所训练出的定位模型，可以适用于该类型的多个冰箱1。本公开对多个冰箱1的数量不作具体限定，例如，多个冰箱1的数量可以尽量多一些，这样，服务器2在训练定位模型的过程中可以忽略掉一些误差，从而提高定位模型的精度。
在一些实施例中，本公开实施例提供一种冰箱1，参照图2A，冰箱1包括箱体10、多个标签20、读写器30和控制器40。
在一些实施例中，参照图2A，箱体10包括多个储藏室11，多个储藏室11可以包括储藏室111、储藏室112和储藏室113。在一些示例中，储藏室111可以为冷藏腔室，用于冷藏储存食物；储藏室112可以为变温腔室，用于变温储存食物；储藏室113可以为冷冻腔室，用于冷冻储存食物。
示例性地，多个标签20可以为RFID标签，也可以为近场通信(Near Field Communication，NFC)标签。在一些示例中，标签20可以包括标签本体，标签本体可拆卸地固定在食材上。例如，标签本体可以被设置为密封夹，该密封夹可夹持在食材的两侧。又例如，标签本体可以包括粘贴部件，该粘贴部件可以将标签本体粘贴在食材上。在另一些示例中，标签20还包括天线，当标签20为RFID标签时，该天线为RFID天线。
需要说明的是，当用户向储藏室11中存放食材或从储藏室11中取出食材时，可以把固定在该食材上的标签20一同放入或取出。
示例性地，标签20还可以包括存储器(例如该存储器可以为存储芯片)，该存储器用于存储标签的标签信息，该标签信息包括食材信息。在一些示例中，每个标签20中可以存储多种食材的食材信息，也可以存储一种食材的食材信息，其中一种食材可以是一个食材，也可以是多个食材。不同标签20存储的食材信息可以至少部分相同，也可以完全不同。标签20上的食材信息例如可以包括食材的种类、存放日期和过期日期等信息。
读写器30设于箱体10内，读写器30被配置为向多个标签20发送射频信号。示例性地，参照图2B，读写器30包括读写主机31和多个天线32。读写主机31与多个天线32耦接，读写主机31可以设置在箱体10上，多个天线32可以设置在冰箱1的各个分区中。
在一些示例中，可以将冰箱1划分为多个分区，每个分区中设置有至少一个天线32。需要说明的是，冰箱1可以按照储藏室的分布方式进行各个分区的划分。例如，可以将冰箱1划分为三个分区，该三个分区分别对应冰箱1的三个储藏室11(即储藏室111、储藏室112和储藏室113)，每个储藏室11内设置的至少一个天线32。例如，该至少一个天线32可以位于该储藏室11的内胆，标签20可以位于该储藏室内胆以外的空间。
例如，多个天线32可以分别设置在储藏室111、储藏室112和储藏室113中，每个储藏室中天线32的数量可以相同，也可以不同。参照图3A，当有8个天线时，储藏室111中可以设置4个天线，储藏室112和储藏室112分别设置2个天线。
示例性地，读写主机31通过控制天线32发送射频信号给每个标签20，或者，天线32将每个标签20发送的射频信号返回给读写主机31，从而实现读写主机31与多个标签20之间的通信。例如，天线32发射出射频信号可以激活标签20，被激活的标签20再将该射频信号通过天线32传出，由读写主机31进行数据的读取和采集，从而获得冰箱1内各食材的相关数据。
每个标签20被配置为：接收读写器30发送的射频信号，根据该射频信号，得到该射频信号在标签20所在位置处的信号强度信息，向读写器30发送信号强度信息和存储在该标签20上的食材信息。
标签20接收各个天线32发送的射频信号，由于各天线32的位置不同，标签20所接收到的每个天线32发送的射频信号的信号强度信息也可能不同。标签20根据多个天线32发送的射频信号，计算该射频信号在标签20所在位置的信号强度信息，然后将该信号强度信息及标签20上存储的食材信息通过天线32发送给读写主机31。读写主机31从天线32中读取标签20的信号强度信息和食材信息。
在一些示例中，每个天线32发送的射频信号在标签20位置处的信号强度信息与该天线32与标签20之间的距离、天线32与标签20之间是否存在障碍物、以及天线32与标签20各自所处环境的湿度等信息有关。例如，天线32与标签20距离越远，该天线32发送的射频信号在标签20位置处的信号强度就越弱；或者，当天线32与标签20之间存在障碍物，或存在的障碍物越多时，该天线32发送的射频信号在标签20位置处的信号强度就越弱；或者，当天线32与标签20各自所处环境的湿度更大时，该天线32发送的射频信号在标签20位置处的信号强度就越弱。由于标签20接收到每个天线32发送的射频信号的信号强度信息可能不同，因此，信号强度信息可以反映标签20的位置信息。当标签20的位置信息确定后，该标签20所对应的食材的位置信息也就确定了。
控制器40与读写器30耦接，且被配置为：响应于一种目标食材的食材定位指令，从读写器30中获取目标食材的食材信息对应的目标标签的信号强度信息；根据目标标签的信号强度信息，采用定位模型，得到目标食材的位置信息。
示例性地，食材定位指令为控制器40进行食材定位的条件，控制器40满足该定位条件，才能进行目标食材的定位。
在一些示例中，该食材定位指令可以是由外界触发的。例如，参照图2B，用户可以通过冰箱1的显示器50，输入关于某一种食材的食材定位指令；又例如，用户也可以通过能够管理冰箱1的终端设备上安装的应用程序(Application，APP)向冰箱1发送关于某一种食材的食材定位指令。
在另一些示例中，该定位指令也可以是冰箱1自动触发的某一种食材定位指令。例如，控制器40在间隔预设的定位周期后，自动触发的食材定位指令。比如，控制器40可以设置每隔2天触发一次食材定位指令，用于对冰箱1内的食材的位置信息进行整理。
例如，控制器40响应于一种目标食材的食材定位指令，从读写主机31中获取该目标食材的食材信息所对应的目标标签的信号强度信息。
在一些示例中，目标食材指的是需要定位的任一食材。目标食材上固定有目标标签(即 任一标签20)，该目标标签上存储有目标食材的食材信息。控制器40响应于目标食材的食材定位指令，根据该食材定位指令中的食材信息，从读写主机31中识别与该食材信息相对应的目标标签，并获取该目标标签的信号强度信息。
例如，参照图3A，当用户将目标标签21放入某一储藏室(如储藏室111)后，冰箱1中的8个天线同时发出射频信号。目标标签21可以接收来自这8根天线发出的射频信号，计算这8根天线在目标标签21所在位置的信号强度信息。目标标签21将获取每根天线的信号强度信息，将该信号强度信息和目标标签21上的食材信息再通过这8个天线发送回读写主机31。
在一些实施例中，每个天线32被配置为：向多个标签20发送射频信号，接收来自多个标签20对应的信号强度信息和食材信息；读写主机31被配置为：获取来自多个天线32的多个标签20对应的信号强度信息和食材信息。
由于读写主机31中获取到的是冰箱1中多个标签20的信号强度信息和食材信息，因此，当控制器40从读写器30中获取目标标签的信号强度信息时，读写主机31需要根据目标食材的食材信息在多个标签20中查找到目标标签21，控制器40从读写主机中获取该目标标签21对应的信号强度信息。
控制器40获取到目标标签21的信号强度信息后，将该信号强度信息输入到定位模型中，该定位模型能够输出该目标标签(即目标食材)的位置信息。
示例性地，定位模型是基于多个标签的信号强度信息和多个标签的位置信息预先训练得到的。
在一些实施例中，服务器2被配置为：在多个冰箱1中的每个冰箱中的多个标签20中确定至少两个待训练标签，该至少两个待训练标签按照预设间隔分布在冰箱1的至少一个储藏室11内。
在一些示例中，相邻两个待训练标签之间的预设间隔可以相同。在一些示例中，可以将储藏室11的空间划分为多个大小尺寸相同的区域，相邻区域相互接壤或者间隔相同，每个待训练标签设置在一个区域中。这样，可以减小相邻待训练标签之间的距离。例如，任意相邻待训练标签之间的距离基本相同。例如，参照图3B，可以将储藏室111的空间划分为9个等尺寸的区域，在每一个区域的心中位置出放置一个待训练标签22。
需要说明的是，储藏室111的空间划分的区域越多，待训练标签22之间的间隔越小，所训练的定位模型的定位精度就越高。例如，可以将储藏室111中划分的多个区域的位置信息分别记为P1-0、P1-1,…,P1-n，其中，n为储藏室111所划分的区域的个数，n为大于或等于1的整数。在n个区域分别放置n个待训练标签22，该n个区域的位置信息分别是第1个至第n个待训练标签的位置信息。
在一些实施例中，服务器2被配置为：获取训练样本集，该训练样本集包括至少两个待训练标签对应的信号强度信息和至少两个待训练标签的位置信息；根据该训练样本集，对初始定位模型进行模型训练，得到定位模型。
冰箱1的每个储藏室11中的所有天线32可以向待训练标签22发出射频信号。如图3B所示，储藏室111、储藏室112和储藏室113中的8个天线同时向各待训练标签22发出射频信号。每个待训练标签22接收到每根天线32发出的射频信号后，计算每个射频信号在各自位置处的信号强度信息，并且将该信号强度信息以及待训练标签22上存储的食材信息通过所有天线32返回给读写主机31。读写主机31获取到每个待训练标签22的信号强度信息以及每个待训练标签22上存储的食材信息，根据待训练标签22上存储的食材信息，将属于同一个待训练标签22的信号强度信息发送给控制器40。
控制器40还会将待训练标签22对应的位置信息P1-i(0≤i≤n，i为整数)和待训练标签22对应的所有信号强度信息发送给服务器2，作为初始定位模型训练的训练样本集。
在一些示例中，服务器2可以对同一待训练标签的信号强度信息采集多次数据，每次 数据都可以作为定位模型的训练数据，以提高定位模型训练的精度。
在进行初始定位模型训练之前，服务器2需要完成初始定位模型的搭建。初始定位模型可以通过卷积神经网络(Convolutional Neural Networks,CNN)来实现。例如，初始定位模型可以通过轻量级卷积神经网络实现，轻量级卷积神经网络可以进行自主学习，能够提取出待训练标签的位置信息中最有效的特征进行表示。
在一些实施例中，初始定位模型(以及定位模型)包括零值填充层、至少一个Inception结构层、池化层和全连接层。其中，Inception结构层由多个卷积组成，至少一个Inception结构层可以包括两个Inception结构层，即第一Inception结构层和第二Inception结构层。
图4为本公开实施例提供的一种定位模型的示意图。如图4所示，定位模型包括第一输入层41，第一输入层41后连接一个第一零值填充层42，第一零值填充层42后连接第一Inception结构43，第一Inception结构43后连接第二Inception结构44，第二Inception结构44后连接池化层45，池化层后连接全连接层46。
在一些示例中，第一零值填充层42可以通过泛卷积实现。泛卷积用于对特征矩阵的边缘进行零值填充操作，以增大特征矩阵的高度和宽度。在另一些示例中，第一零值填充层42还可以连接一个卷积层，该卷积层可以为包括多个1×1大小的卷积核的卷积(可以称为1×1卷积)，该卷积层用于对泛卷积得到的特征矩阵进行增维处理，以增大该特征矩阵的通道数。
示例性地，第一Inception结构43和第二Inception结构44可以为两个结构相同的Inception结构，其中，每个Inception结构都可以由四个分支组成。
在一些示例中，如图4所示，每个Inception结构的第一个分支对特征矩阵进行1×1卷积操作；第二个分支先对特征矩阵进行1×1卷积后，再使用3×3卷积对1×1卷积处理后的特征矩阵进行卷积操作；第三个分支先对特征矩阵进行1×1卷积，再使用5×5卷积对1×1卷积处理后的特征矩阵进行卷积操作；第四分支先对特征矩阵进行3×3最大池化下采样处理，再使用1×1卷积对3×3最大池化处理后的特征矩阵进行卷积操作。
在另一些示例中，Inception结构的四个分支中每一步操作的步长可以均为1。这样，每个分支所输出的特征矩阵具有相同的尺寸，即具有相同的高度和宽度(每个分支输出的特征矩阵的深度可以不同)，因而，可以将每个分支输出的特征矩阵沿着深度方向进行叠加。
示例性地，第一Inception结构43和第二Inception结构44之间还包括第一深度叠加(DepthConcat)层47，第一深度叠加层47用于将第一Inception结构43的四个分支分别输出的特征矩阵按照深度方向进行拼接，得到一个完整的特征矩阵。类似的，在第二Inception结构44后也可以连接一个第二深度叠加层48，第二深度叠加层48用于对第二Inception结构44的四个分支分别输出的特征矩阵按照深度方向进行拼接，得到另一个完整的特征矩阵。
然后，将经过第二深度叠加层48输出的完整的特征矩阵输入到池化层45，池化层45对该特征矩阵进行全局平均池化下下采样操作。最后，将池化层45处理后的特征矩阵输入到全连接层46，全连接层46对冰箱1的各分区输出的特征矩阵进行全连接，例如，参照图3B，将冰箱1中三个分区的训练数据训练得到特征矩阵，经过全连接层46处理后得到一个一维特征矩阵。该一维特征矩阵可以对应一个待训练标签的位置信息。
在完成初始定位模型的搭建后，服务器2采用获取到的训练样本集对该初始定位模型进行训练。
在一些实施例中，服务器2还被配置为：根据每个待训练标签对应的信号强度信息，得到初始定位模型训练数据。
服务器2将待训练标签的信号强度信息作为初始定位模型的训练数据，例如，可以将同一待训练标签的信号强度信息作为训练数据输入到初始定位模型中。例如，参照图3B， 可以将8个天线在每个训练标签22位置处的信号强度信息作为训练数据输入到初始定位模型中。
示例性地，利用训练数据对初始定位模型训练之前，服务器2还被配置为：对训练数据进行预处理。预处理可以包括归一化处理和增维处理。
首先，服务器2对训练数据进行归一化处理可以包括将该训练数据映射到0至1区间内，例如，将8个天线在每个训练标签22位置处的信号强度信息分别映射为[0,1]区间的某一数值。归一化处理可以提升定位模型的收敛速度，缩短训练所需要的时间。
然后，服务器2对归一化处理后的训练数据进行增维处理。例如，可以将一维的训练数据扩充到两个维度，生成二维数据。在一些示例中，服务器2可以将该二维数据保存在一个逗号分隔值(Comma-Separated Values，CSV)的文件中，该CSV文件中的每一行数据可以看成一个特征图，该CSV文件的行数可以作为特征图的数量。
因此，经过预处理后，服务器2可以将训练数据转换为特征图的形式以输入到初始定位模型中进行训练，以特征图的形式表征信号强度信息，更加便于初始定位模型的识别。
下面参照图3B和图4，以一个待训练标签包括8个信号强度信息为例，对图4所示的初始定位模型的训练过程进行示例性说明。需要说明的是，每个待训练标签的训练数据都可以采用以下步骤进行训练。
第一步，将预处理后的训练数据输入到第一输入层41，第一输入层41将预处理后的训练数据转换为8×1×1大小的特征矩阵，以方便后续进行卷积操作。例如，将第一输入层41处理后的数据称为第一特征矩阵。
第二步，将第一特征矩阵输入到第一零值填充层42，第一零值填充层42采用泛卷积对第一特征矩阵的边缘使用零值进行填充，以增大第一特征矩阵的高度和宽度。例如，可以对8×1×1大小的第一特征矩阵的上、下、左、右分别填充两行和两列的零值，从而得到12×5×1大小的特征矩阵。在一些示例中，将12×5×1大小的特征矩阵经过6个大小为1×1的卷积核的卷积运算后，得到12×5×6大小的第二特征矩阵。由此，第一零值填充层42增大了第一特征矩阵的尺寸以及通道数。
第三步，将第二特征矩阵输入到第一Inception结构43，第二特征矩阵分别经过第一Inception结构43的四个分支的处理。其中，第一分支经过1×1卷积处理后，输出12×5×6大小的特征矩阵；第二分支经过1×1卷积核3×3卷积处理后，输出12×5×6大小的特征矩阵；第三分支经过1×1卷积核5×5卷积处理后，输出12×5×6大小的特征矩阵；第四分支经过3×3最大池化和1×1卷积处理后，输出12×5×6大小的特征矩阵。在一些示例中，将四个分支输出的特征矩阵输入到第一深度叠加层47，第一深度叠加层47对该四个特征矩阵按照深度方向进行拼接，得到12×5×24大小的第三特征矩阵。
第四步，将第三特征矩阵输入到第二Inception结构44，第二Inception结构44的处理过程与第一Inception结构43类似，此处不再赘述。第二Inception结构44的四个分支的输出的特征矩阵输入到第二深度叠加层48，第二深度叠加层47对四个分支输出的特征矩阵按照深度方向进行拼接，得到12×5×48大小的第四特征矩阵。
第五步，将第四特征矩阵输入到池化层45中，池化层46对第四特征矩阵采用3×1平均池化对第四特征矩阵进行全局平均池化下采样处理，得到6×2×48大小的第五特征矩阵。
第六步，将冰箱1的三个分区的待训练标签的训练数据训练后输出的第五特征矩阵输入到全连接层46进行全连接，输出训练结果。该训练结果与待训练标签的位置信息相对应。
在一些示例中，该初始定位模型还包括分类器(SoftMax)，分类器将全连接层输出结果映射到[0,1]区间内，从而实现多分类，进一步获取到待训练标签对应的分区以及位于分区中的具体位置信息。
在另一些实施例中，初始定位模型还可以包括零值填充层、至少一个卷积层、至少一 个池化层、至少一个全连接层和分类器层。其中，至少一个卷积层可以包括两个卷积层，至少一个池化层可以包括两个池化层，至少一个全连接层可以包括两个全连接层。
图5为本公开实施例提供的另一种定位模型的结构的示意图。如图5所示，初始定位模型(以及定位模型)还可以包括第二输入层51、第二零值填充层52、第一卷积层53、第一池化层54、第二卷积层55，第二池化层56，第一全连接层57、第二全连接层58和分类器59。
在一些示例中，卷积层(即第一卷积层53和第二卷积层55)可以由卷积操作(Convolution)、批量标准化(Batch-Normalization)和激活函数Leaky ReLU三部分构成。卷积操作用于获取位置信息的特征映射，然后通过非线性函数激活该特征映射。
卷积核由初始化权重和偏置组成，卷积运算由卷积核以滑动窗口的形式实现。例如，卷积运算可以通过公式(1)实现：
m i＝F(Wx i:i+w-1+b)   公式(1)；
其中，m i是卷积运算后的特征映射，x i:i+w-1为卷积核对应滑动窗口内的数据，W为权重信息，b为一个数值较小的常量。
当输入数据分布相近，或者分布在较小的范围内时(如0附近)，更有利于函数的迭代优化。为了保证输入的分布相近，可以通过批量标准化进行数据的映射，数据x与对应的映射后的数据
Figure PCTCN2022120029-appb-000001
满足如下公式(2)：
Figure PCTCN2022120029-appb-000002
其中，μ r和
Figure PCTCN2022120029-appb-000003
来自统计的所有数据x的均值和方差，ε是为了防止除0的错误而设置的较小数，本公开实施例中将ε设置为
Figure PCTCN2022120029-appb-000004
将随机分布的位置数据转化成按照正态分布的数据，使输入网络的数据分布较近，更有利于网络的迭代优化。
本公开实施例采用Leaky ReLU激活函数，克服了梯度弥漫的现象，激活函数Leaky ReLU满足如公式(3)所述的关系：
Figure PCTCN2022120029-appb-000005
在一些示例中，可以设定p为0.04，以保证在负区域时的梯度不为0。
池化层(即第一池化层和第二池化层)可以对局部相关的一组数据进行采样或信息聚合。在一些示例中，池化层可以为最大池化层，即第一最大池化层和第二最大池化层。例如，最大池化可以从局部相关数据中选取最大的一个数据。
全连接层(即第一全连接层和第二全连接层)将上述各层中提取出来的特征输入到全连接网络，然后计算分类评估值。
示例性地，服务器2在采用图5所示的定位模型进行训练之前，需要对待训练数据进行预处理，该预处理包括归一化处理和增维处理。其中，归一化处理与图4所示的定位模型中的归一化处理类似，此处不再赘述。在一些示例中，可以将归一化处理后的一维训练数据训练转换为三维数据，并对三维形式的训练数据进行独热编码。然后将经过预处理的训练数据输入到图5所示的定位模型中进行模型训练。
下面参照图5和图6，以图3B中一个待训练标签包括8个信号强度信息为例，对图5所示的初始定位模型的训练过程进行示例性说明。
第一步，将经过预处理后的训练数据输入到定位模型的第二输入层51中，得到8×1×1大小的特征矩阵。需要说明的是，第二输入层51和第一输入层41的处理过程相同，此处不再赘述。
第二步，将8×1×1大小的特征矩阵输入到第二零值填充层52，进行边缘零值填充处理。例如，可以在该特征矩阵的上、下、左、右分别填充1行和1列的零值，得到如图6所示的10×3×1大小的特征矩阵。
第三步，将10×3×1大小的特征矩阵输入到第一卷积层53，例如，经过32个8×1大小的卷积核的卷积运算，得到9×2×32大小的特征矩阵；再经过第一池化层54，例如3×1最大池化后，将9×2×32大小的特征矩阵转换为4×1×32大小的特征矩阵。
第四部，将4×1×32大小的特征矩阵输入到第二卷积层55，例如，经过10个4×1大小的卷积核的卷积运算，得到4×1×10大小的特征矩阵；再经过第二池化层56处理后，例如3×1最大池化后，得到如图6所示的4×1×10大小的特征矩阵。
第五步，将第二最大池化层得到的4×1×10大小的特征矩阵输入到第一全连接层57，得到40×1的二维特征矩阵，再经过第二全连接层58，得到如图6所示的3×1的二维特征矩阵。
第六步，将3×1大小的特征矩阵通过分类器59进行分类，得到训练结果。
需要说明的是，定位模型是经过大量的训练数据训练而成的，该大量的训练数据包括多次采集的多个标签的信号强度信息。在一些示例中，训练数据的数据量越大，训练的次数越多，所得到的定位模型的精确越高。
在一些示例中，服务器2在完成对初始定位模型的训练后，还可以采用测试数据对得到的定位模型进行测试。该测试数据可以为多个冰箱1中任一标签的信号强度信息。
在一些实施例中，服务器2将训练得的定位模型发送给各个冰箱1，冰箱1在获取到该定位模型后，对该定位模型进行存储，并利用该定位模型用于对食材的定位。
在一些示例中，控制器40将获取到目标标签的信号强度信息输入到该定位模型中，该定位模型按照与上述实施例中定位模型(如图4或图5的定位模型)的训练过程类似的步骤(此处不再赘述)，对该信号强度信息进行处理后，输出该目标标签(即目标食材)的位置信息。在另一些示例中，控制器40将获取到目标标签的信号强度信息输入到该定位模型后，该定位模型可以直接输出该信号强度信息对应的目标标签(即目标食材)的位置信息。
因此，本公开实施例提供的冰箱1，通过预先存储服务器2发送的定位模型，并采用该定位模型，通过属于同一食材信息对应的标签20的信号强度，得到标签20的位置信息，该位置信息为标签20对应的食材在储藏室11中的具体位置信息，该位置信息能够方便用户更加快速获取到该食材的存放位置。
本公开实施例提供的冰箱管理系统3与冰箱1具有相同的有益效果。另外，本公开实施例提供的冰箱管理系统3中，服务器2能够完成初始定位模型的搭建，并且采用训练样本集对该初始定位模型进行训练，得到定位模型，将该定位模型发送给多个冰箱1，用于对冰箱1中食材的定位，该定位模型能够对食材的定位更加精确。
在一些实施例中，如图2B所示，冰箱1还包括显示器50，显示50与控制器40耦接，被配置为：获取多个储藏室的空间信息；显示第一界面，该第一界面包括多个储藏室的空间信息；接收来自控制器40的目标食材的位置信息；显示第二界面，第二界面包括目标食材在对应的储藏室的空间中的位置信息。
控制器40可以定位模型中得到的目标食材的位置信息发送给显示器50，显示器50对该位置信息对应的位置进行显示，更加便于用户获取到目标食材的具体位置。
在一些示例中，显示器50中可以预先存储有冰箱1的各储藏室11的空间信息，并且可以通过第一界面对该空间信息进行显示。该空间信息包括各储藏室11的空间分布图或空间分布数据。例如，显示器50可以分别显示储藏室111、储藏室112和储藏室112对应的三维模型图或者三维尺寸数值。
图7A为本公开实施例提供的一种显示器50显示的第一界面的示意图，如图7A所示，显示器50显示了储藏室111对应的三维模型图。
当显示器50接收到来自控制器40发送的目标食材的位置信息后，可以显示第二界面，第二界面包括在目标食材的位置信息对应的储藏室的三维模型图中标记出目标食材的位 置。
图7B为本公开实施例提供的一种显示器50显示的第二界面的示意图，如图7B所示，显示器50显示了目标食材的位置在对应的储藏室111的三维模型图中的标记。
在一些示例中，当该位置信息包括至少两个时，可以采用不同的标记方式标记，便于用户能够清楚的获取到目标食材位于哪一储藏室的某一位置。
在一些实施例中，控制器40还被配置为：在得到目标食材的位置信息后，确定目标食材的位置信息的准确率；在预设时间内，采用定位模型，得到多种目标食材的位置信息；根据多种目标食材的位置信息的准确率，确定定位模型的准确率。
在一些实施例中，显示器还被配置为：显示第三界面，第三界面包括所述目标食材的位置信息的准确率。
控制器40在得到目标食材的位置信息后，可以进一步获取该位置信息的准确率。在一些示例中，控制器40可以通过显示器50来获取该目标食材的位置信息的准确率。
例如，显示器50在对目标食材的位置信息对应的位置进行显示时，还可以显示一个反馈窗口，该反馈窗口用于为用户提供选择有关该位置是否准确的选项，用户通过选择该位置是否准确来得到该位置信息的准确率，控制器40获取该位置信息的准确率。
图7C为本公开实施例提供的一种显示器50显示的第三界面的示意图，如图7C所示，显示器50显示了反馈窗口，该反馈窗口包括指示用户选择当前定位的位置信息是否准确的选项，控制器40可以通过用户选择是或者否得到该位置信息的准确率。例如，当用户选择是，控制器40可以获取到本次定位的准确率为1，当用户选择否，控制器40可以获取到本次定位的准确率为0。在一些示例中，控制器40每完成一次目标食材的定位，显示器50可以显示一次第三界面。
示例性地，控制器40在预设时间内，例如第一预设时间内，可以对多种目标食材进行定位，并且获取到该多种目标食材的位置信息。本公开实施例对第一预设时间不作限定例如，第一预设时间可以为7天或者30天。在一些示例中，多种目标食材可以为第一预设时间内，用户查询的食材，或者用户反馈了准确率的食材。
控制器40根据每种目标食材的位置信息的准确率，进一步得到该定位模型的准确率。在一些示例中，定位模型的准确率可以该多种目标食材的位置信息的准确率的平均值。定位模型的准确率可以反映该定位模型对食材定位的精确程度。
在一些实施例中，控制器40还被配置为：接收来自服务器2的模型更新参数；根据该模型更新参数更新定位模型，得到更新后的定位模型。采用更新后的定位模型，得到一种目标食材的更新后的位置信息。
在冰箱1获取到定位模型后，在预设时间后，例如第二预设时间，还可以对该定位模型进行更新。其中，第二预设时间和第一预设时间可以相同，也可以不同。例如，第二预设时间为5天或者7天。
在一些示例中，服务器2通过与其具有关联关系的各个冰箱1中获取的训练数据对其存储的定位模型进行训练，得到更新后的定位模型，从而获取到更新后的定位模型的参数。并且，每隔第二预设时间，服务器2会将该模型更新参数发送给各个冰箱1。
冰箱1接收服务器2发送的模型更新参数后，根据该模型更新参数更新定位模型，对存储的定位模型进行更新，得到更新后的定位模型。
在一些实施例中，在第一预设时间内，采用所更新后的定位模型，得到多种目标食材的更新后的位置信息；根据多种目标食材的更新后的位置信息的准确率，确定更新后的定位模型的准确率；判断更新后的定位模型的准确率与定位模型的准确率大小关系；若更新后的定位模型的准确率大于或等于定位模型的准确率，保留所述更新后的定位模型；若更新后的定位模型的准确率小于定位模型的准确率，则从服务器2重新获取定位模型。
需要说明的是，采用更新后的定位模型进行定位的多种目标食材与上述实施例中采用 定位模型进行定位的多种目标食材可以相同，也可以不同，该多种目标食材可以分别为冰箱1中的任意食材，或者为用户需要进行定位的任意食材。
在一些示例中，冰箱1中预先设置的定位模型为第一定位模型，采用模型更新参数得到更新后定位模型为第二定位模型，当冰箱1获取到第二定位模型的准确率后，将第二定位模型的准确率与第一定位模型的准确率进行比较。需要说明的是，冰箱1获取第二定位模型的准确率与上述实施中冰箱1获取第一定位模型的准确率的过程类似，此处不再赘述。
当第二定位模型的准确率大于或等于第一定位模型的准确率时，冰箱1可以保留更新后的定位模型(即第二定位模型)，并采用第二定位模型继续进行获取目标食材的位置信息。
当第二定位模型的准确率小于第一定位模型的准确率时，冰箱1会从服务器2重新获取第一定位模型，并且采用第一定位模型继续进行获取目标食材的位置信息。在一些示例中，冰箱1可在重新获取到第一定位模型后，可以将第二定位模型进行删除。
冰箱1通过服务器2不断获取更新的定位模型对食材进行定位，可以保证冰箱1中定位模型的准确率保持在较高的水平，进一步提高食材定位的精确度。
图8为本公开实施例提供一种冰箱管理系统的控制方法，该冰箱管理系统可以为上述实施例中的冰箱管理系统3，该冰箱管理系统包括多个冰箱和服务器，该多个冰箱可以为上述实施例中的多个冰箱1。该冰箱包括多个标签、读写器、控制器和箱体，如图8所示，该冰箱管理系统的控制方法可以包括以下步骤：
步骤81，控制器40响应于一种目标食材的食材定位指令，从读写器30中获取目标食材的食材信息对应的目标标签的信号强度信息。
步骤82，控制器40根据目标标签的信号强度信息，采用定位模型，得到目标食材的位置信息。
其中，定位模型是基于多个标签20的信号强度信息和多个标签20的位置信息预先训练得到。
在一些实施例中，控制器40在得到目标食材的位置信息后，确定目标食材的位置信息的准确率；在预设时间内，控制器40采用所述定位模型，得到多种目标食材的位置信息；控制器40根据多种目标食材的位置信息的准确率，确定定位模型的准确率。
图9为本公开实施例提供另一种冰箱管理系统的控制方法，如图9所示，该冰箱管理系统的控制方法可以包括以下步骤：
步骤91，服务器2向控制器40发送模型更新参数。
步骤92，控制器40接收来自服务器2的模型更新参数。
步骤93，控制器40根据模型更新参数，更新定位模型，得到更新后的定位模型。
步骤94，控制器40采用更新后的定位模型，得到一种目标食材的更新后的位置信息。
步骤95，在预设时间内，控制器40采用更新后的定位模型，得到多种目标食材的更新后的位置信息。
步骤96，控制器40根据多种目标食材的更新后的位置信息的准确率，确定更新后的定位模型的准确率。
步骤97，控制器40判断更新后的定位模型的准确率是否小于定位模型的准确率。
若更新后的定位模型的准确率小于定位模型的准确率，执行步骤98；若更新后的定位模型的准确率大于或等于定位模型的准确率，执行步骤99。
步骤98，控制器40从服务器2获取定位模型。
步骤99，控制器40保留更新后的定位模型。
在一些实施例中，该冰箱管理系统的控制方法还包括：每个天线32向多个标签20发送射频信号，接收来自多个标签20对应的信号强度信息和食材信息；读写主机31获取来自多个天线32的多个标签20对应的信号强度信息和食材信息。
在一些实施例中，该冰箱管理系统的控制方法还包括：显示器50获取多个储藏室11的空间信息；显示器50显示第一界面，第一界面包括多个储藏室11的空间信息；显示器50接收来自控制器40发送的目标食材的位置信息；显示第二界面，第二界面包括目标食材在对应的储藏室11中的空间中的位置信息。
在一些实施例中，该冰箱管理系统的控制方法还包括：显示器50显示第三界面，所第三界面包括目标食材的位置信息的准确率。
在一些实施例中，服务器2在每个冰箱1中的多个标签20中确定至少两个待训练标签，至少两个待训练标签按照预设间隔分布在冰箱1的至少一个储藏室11内；其中，冰箱1为与服务器2耦接的多个冰箱中的一个；该冰箱管理系统的控制方法还包括：服务器2获取训练样本集，训练样本集包括至少两个待训练标签对应的信号强度信息和至少两个待训练标签的位置信息；服务器2根据训练样本集，对初始定位模型进行模型训练，得到定位模型；服务器2将定位模型发送给多个冰箱。
在一些实施例中，初始定位模型包括零值填充层、至少一个Inception结构层、池化层和全连接层；服务器2根据训练样本集，对初始定位模型进行模型训练，包括：服务器2根据每个待训练标签对应的信号强度信息，得到初始定位模型的训练数据；服务器2对训练数据进行预处理；服务器2将经过所预处理后的训练数据依次输入到零值填充层、至少一个Inception结构层、池化层和全连接层，得到训练数据的处理结果；其中，训练数据的处理结果与待训练标签的位置信息对应。
在一些实施例中，服务器2对训练数据进行预处理，包括：服务器2对训练数据进行归一化处理；服务器2对归一化处理后的训练数据进行增维处理。
上述冰箱管理系统的控制方法和上述一些实施例所述的冰箱管理系统的有益效果相同，此处不再赘述。
需要说明的是，本公开实施例的附图中以特定顺序描述的各个步骤，并非要求或者暗示必须按照该特定顺序来执行这些步骤，或是必须执行全部所示的步骤才能实现期望的结果。可以对附图中的各步骤进行附加，也可以省略某些步骤，或者将多个步骤合并为一个步骤执行，或者将一个步骤分解为多个步骤执行等。
本公开的一些实施例提供了一种计算机可读存储介质(例如，非暂态计算机可读存储介质)，其上存储有计算机程序，该计算机程序被冰箱管理系统执行时，使得冰箱管理系统执行如上述实施例中任一实施例所述的冰箱管理系统的控制方法。
示例性的，上述计算机可读存储介质可以包括，但不限于：磁存储器件(例如，硬盘、软盘或磁带等)，光盘(例如，CD(Compact Disk，压缩盘)、DVD(Digital Versatile Disk，数字通用盘)等)，智能卡和闪存器件(例如，EPROM(Erasable Programmable Read-Only Memory，可擦写可编程只读存储器)、卡、棒或钥匙驱动器等)。本公开描述的各种计算机可读存储介质可代表用于存储信息的一个或多个设备和/或其它机器可读存储介质。
本公开的一些实施例还提供了一种计算机程序产品。该计算机程序产品包括计算机程序，该计算机程序存储在非暂态计算机可读存储介质上。其中，该计算机程序在被冰箱管理系统执行时，使得冰箱管理系统执行如上述实施例所述的冰箱管理系统的控制方法。
本公开的一些实施例还提供了一种计算机程序。该计算机程序存储在非暂态计算机可读存储介质上。当该计算机程序在被冰箱管理系统执行时，使得冰箱管理系统执行如上述实施例所述的冰箱管理系统的控制方法。
上述计算机可读存储介质、计算机程序产品及计算机程序的有益效果和上述一些实施例所述的冰箱管理系统的控制方法的有益效果相同，此处不再赘述。
本领域技术人员将会理解，本发明的公开范围不限于上述具体实施例，并且可以在不脱离本申请的精神的情况下对实施例的某些要素进行修改和替换。本申请的范围受附权利要求的限制。

Claims (20)

  1. 一种冰箱，包括：
    箱体，包括多个储藏室；
    读写器，设置于所述箱体内，被配置为：向多个标签发送射频信号；
    所述多个标签，位于所述多个储藏室中，每个标签存储有至少一种食材的食材信息，所述标签被配置为：接收所述射频信号，根据所述射频信号，得到所述射频信号在所述标签所在位置处的信号强度信息，向所述读写器发送所述信号强度信息和存储在所述标签上的食材信息；
    控制器，与所述读写器耦接，且被配置为：响应于一种目标食材的食材定位指令，从所述读写器中获取所述目标食材的食材信息对应的目标标签的信号强度信息；以及根据所述目标标签的信号强度信息，采用定位模型，得到所述目标食材的位置信息；其中，所述定位模型是基于所述多个标签的信号强度信息和所述多个标签的位置信息预先训练得到的。
  2. 根据权利要求1所述的冰箱，其中，所述控制器，还被配置为：
    在得到所述目标食材的位置信息后，确定所述目标食材的位置信息的准确率；
    在预设时间内，采用所述定位模型，得到多种目标食材的位置信息；
    根据所述多种目标食材的位置信息的准确率，确定所述定位模型的准确率。
  3. 根据权利要求2所述的冰箱，其中，所述控制器，还被配置为：
    接收来自服务器的模型更新参数；
    根据所述模型更新参数，更新所述定位模型，得到更新后的定位模型；
    采用所述更新后的定位模型，得到一种目标食材的更新后的位置信息；
    在所述预设时间内，采用所述更新后的定位模型，得到多种目标食材的更新后的位置信息；
    根据所述多种目标食材的更新后的位置信息的准确率，确定所述更新后的定位模型的准确率；
    判断所述更新后的定位模型的准确率与所述定位模型的准确率大小关系；
    若所述更新后的定位模型的准确率大于或等于所述定位模型的准确率，保留所述更新后的定位模型；
    若所述更新后的定位模型的准确率小于所述定位模型的准确率，则从所述服务器重新获取所述定位模型。
  4. 根据权利要求1-3中任一项所述的冰箱，其中，所述读写器包括一个读写主机和多个天线，一个储藏室中设置有至少一个天线；
    每个天线被配置为：向所述多个标签发送所述射频信号，接收所述多个标签对应的信号强度信息和食材信息；
    所述读写主机被配置为：获取来自所述多个天线的所述多个标签对应的信号强度信息和食材信息。
  5. 根据权利要求1-4中任一项所述的冰箱，还包括：显示器，与所述控制器耦接，且被配置为：
    获取所述多个储藏室的空间信息；
    显示第一界面，所述第一界面包括所述多个储藏室的空间信息；
    接收来自所述控制器的所述目标食材的位置信息；
    显示第二界面，所述第二界面包括所述目标食材在对应的储藏室的空间中的位置信息。
  6. 根据权利要求5所述的冰箱，所述显示器，还被配置为：显示第三界面，所述第三界面包括所述目标食材的位置信息的准确率。
  7. 一种冰箱管理系统，包括：
    多个如权利要求1所述的冰箱；以及
    服务器，与所述多个冰箱耦接，且被配置为：向所述多个冰箱提供所述定位模型。
  8. 根据权利要求7所述的冰箱管理系统，其中，所述服务器被配置为：
    在所述多个冰箱中的每个冰箱中的多个标签中确定至少两个待训练标签，所述至少两个待训练标签按照预设间隔分布在所述冰箱的至少一个储藏室内；
    获取训练样本集，所述训练样本集包括所述至少两个待训练标签对应的信号强度信息和所述至少两个待训练标签的位置信息；
    根据所述训练样本集，对初始定位模型进行模型训练，得到所述定位模型；
    将所述定位模型发送给所述多个冰箱。
  9. 根据权利要求8所述的冰箱管理系统，其中，所述初始定位模型包括零值填充层、至少一个Inception结构层、池化层和全连接层；
    所述服务器，被配置为：
    根据每个待训练标签对应的信号强度信息，得到所述初始定位模型的训练数据；
    对所述训练数据进行预处理；
    将经过预处理后的训练数据依次输入到所述零值填充层、所述至少一个Inception结构层、所述池化层和所述全连接层，得到所述训练数据的处理结果；其中，所述训练数据的处理结果与所述待训练标签的位置信息对应。
  10. 根据权利要求9所述的冰箱管理系统，其中，所述服务器，被配置为：
    对所述训练数据进行归一化处理；
    对所述归一化处理后的训练数据进行增维处理。
  11. 一种冰箱管理系统的控制方法，其中，所述冰箱管理系统包括：多个冰箱和服务器；所述多个冰箱中的每个冰箱包括多个标签、读写器、控制器和箱体，所述箱体包括多个储藏室；所述读写器设置于所述箱体内；所述多个标签位于所述多个储藏室中，所述多个标签的每个标签存储有至少一种食材的食材信息，所述控制器与所述读写器耦接；
    所述方法包括：
    所述控制器响应于一种目标食材的食材定位指令，从所述读写器中获取所述目标食材的食材信息对应的目标标签的信号强度信息；
    所述控制器根据所述目标标签的信号强度信息，采用来自所述服务器的定位模型，得到所述目标食材的位置信息；其中，所述定位模型是基于所述多个标签的信号强度信息和所述多个标签的位置信息预先训练得到的。
  12. 根据权利要求11所述的方法，还包括：
    所述控制器在得到所述目标食材的位置信息后，确定所述目标食材的位置信息的准确率；
    在预设时间内，所述控制器采用所述定位模型，得到多种目标食材的位置信息；
    所述控制器根据所述多种目标食材的位置信息的准确率，确定所述定位模型的准确率。
  13. 根据权利要求12所述的方法，还包括：
    所述控制器接收来自所述服务器的模型更新参数；
    根据所述模型更新参数，所述控制器更新所述定位模型，得到更新后的定位模型；
    所述控制器采用所述更新后的定位模型，得到一种目标食材的更新后的位置信息；
    在所述预设时间内，所述控制器采用所述更新后的定位模型，得到多种目标食材的更新后的位置信息；
    所述控制器根据所述多种目标食材的更新后的位置信息的准确率，确定所述更新后的定位模型的准确率；
    所述控制器判断所述更新后的定位模型的准确率与所述定位模型的准确率大小关系；
    若所述更新后的定位模型的准确率大于或等于所述定位模型的准确率，所述控制器保留所述更新后的定位模型；
    若所述更新后的定位模型的准确率小于所述定位模型的准确率，所述控制器从所述服务器重新获取所述定位模型。
  14. 根据权利要求11-13中任一项所述的方法，其中，所述读写器包括一个读写主机和多个天线，一个储藏室中设置有至少一个天线；所述方法还包括：
    每个天线向所述多个标签发送射频信号，接收来自所述多个标签对应的信号强度信息和食材信息；
    所述读写主机获取来自所述多个天线的所述多个标签对应的信号强度信息和食材信息。
  15. 根据权利要求11-14中任一项所述的方法，其中，所述冰箱还包括显示器，所述方法还包括：
    所述显示器获取所述多个储藏室的空间信息；
    所述显示器显示第一界面，所述第一界面包括所述多个储藏室的空间信息；
    所述显示器接收所述控制器发送的所述目标食材的位置信息；
    显示第二界面，所述第二界面包括所述目标食材在对应的储藏室中的空间中的位置信息。
  16. 根据权利要求15所述的方法，还包括：
    所述显示器显示第三界面，所述第三界面包括所述目标食材的位置信息的准确率。
  17. 根据权利要求11所述的方法，其中，所述服务器在每个冰箱中的多个标签中确定至少两个待训练标签，所述至少两个待训练标签按照预设间隔分布在所述冰箱的至少一个储藏室内；其中，所述冰箱为与所述服务器耦接的多个冰箱中的一个；
    所述方法还包括：
    所述服务器获取训练样本集，所述训练样本集包括所述至少两个待训练标签对应的信号强度信息和所述至少两个待训练标签的位置信息；
    所述服务器根据所述训练样本集，对初始定位模型进行模型训练，得到所述定位模型；
    所述服务器将所述定位模型发送给所述多个冰箱。
  18. 根据权利要求17所述的方法，其中，所述初始定位模型包括零值填充层、至少一个Inception结构层、池化层和全连接层；
    所述服务器根据所述训练样本集，对初始定位模型进行模型训练，包括：
    所述服务器根据每个待训练标签对应的信号强度信息，得到所述初始定位模型的训练数据；
    所述服务器对所述训练数据进行预处理；
    所述服务器将经过预处理后的训练数据依次输入到所述零值填充层、所述至少一个Inception结构层、所述池化层和所述全连接层，得到所述训练数据的处理结果；其中，所述训练数据的处理结果与所述待训练标签的位置信息对应。
  19. 根据权利要求18所述的方法，其中，所述服务器对所述训练数据进行预处理，包括：
    所述服务器对所述训练数据进行归一化处理；
    所述服务器对所述归一化处理后的训练数据进行增维处理。
  20. 一种非暂态计算机可读存储介质，其上存储有计算机程序，所述计算机程序被冰箱系统执行时，使得所述冰箱系统实现如权利要11-19中任一项所述的冰箱管理系统的控制方法。
PCT/CN2022/120029 2021-10-18 2022-09-20 冰箱、冰箱管理系统及冰箱管理系统的控制方法 WO2023065933A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202280052817.6A CN117730235A (zh) 2021-10-18 2022-09-20 冰箱、冰箱管理系统及冰箱管理系统的控制方法

Applications Claiming Priority (4)

Application Number Priority Date Filing Date Title
CN202111207745.1A CN113915840B (zh) 2021-10-18 2021-10-18 一种冰箱及其食材定位方法
CN202111212660.2 2021-10-18
CN202111207745.1 2021-10-18
CN202111212660.2A CN113915841B (zh) 2021-10-18 2021-10-18 一种冰箱及其食材定位方法

Publications (1)

Publication Number Publication Date
WO2023065933A1 true WO2023065933A1 (zh) 2023-04-27

Family

ID=86058794

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/120029 WO2023065933A1 (zh) 2021-10-18 2022-09-20 冰箱、冰箱管理系统及冰箱管理系统的控制方法

Country Status (2)

Country Link
CN (1) CN117730235A (zh)
WO (1) WO2023065933A1 (zh)

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106568293A (zh) * 2016-10-14 2017-04-19 青岛海尔特种电冰箱有限公司 冰箱食物操作的检测方法及冰箱
CN107623724A (zh) * 2017-08-28 2018-01-23 合肥美的智能科技有限公司 冰箱内食材和空间共享的方法及系统、冰箱
CN107798277A (zh) * 2016-09-05 2018-03-13 合肥美的智能科技有限公司 食材识别系统和方法、食材模型训练方法、冰箱和服务器
CN107975996A (zh) * 2016-10-21 2018-05-01 九阳股份有限公司 一种冰箱控制方法
CN108932512A (zh) * 2017-05-22 2018-12-04 九阳股份有限公司 一种冰箱食材录入方法
CN110906640A (zh) * 2018-09-14 2020-03-24 博西华电器（江苏）有限公司 存储设备的物品定位方法及存储设备
CN113915841A (zh) * 2021-10-18 2022-01-11 海信（山东）冰箱有限公司 一种冰箱及其食材定位方法
CN113915840A (zh) * 2021-10-18 2022-01-11 海信（山东）冰箱有限公司 一种冰箱及其食材定位方法

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107798277A (zh) * 2016-09-05 2018-03-13 合肥美的智能科技有限公司 食材识别系统和方法、食材模型训练方法、冰箱和服务器
CN106568293A (zh) * 2016-10-14 2017-04-19 青岛海尔特种电冰箱有限公司 冰箱食物操作的检测方法及冰箱
CN107975996A (zh) * 2016-10-21 2018-05-01 九阳股份有限公司 一种冰箱控制方法
CN108932512A (zh) * 2017-05-22 2018-12-04 九阳股份有限公司 一种冰箱食材录入方法
CN107623724A (zh) * 2017-08-28 2018-01-23 合肥美的智能科技有限公司 冰箱内食材和空间共享的方法及系统、冰箱
CN110906640A (zh) * 2018-09-14 2020-03-24 博西华电器（江苏）有限公司 存储设备的物品定位方法及存储设备
CN113915841A (zh) * 2021-10-18 2022-01-11 海信（山东）冰箱有限公司 一种冰箱及其食材定位方法
CN113915840A (zh) * 2021-10-18 2022-01-11 海信（山东）冰箱有限公司 一种冰箱及其食材定位方法

Also Published As

Publication number Publication date
CN117730235A (zh) 2024-03-19

Similar Documents

Publication Publication Date Title
US11756324B2 (en) Fish biomass, shape, size, or health determination
CN111523621A (zh) 图像识别方法、装置、计算机设备和存储介质
CN110929774B (zh) 图像中目标物的分类方法、模型训练方法和装置
TWI704513B (zh) 用於疾病預測及控制之方法及系統
CN109191455A (zh) 一种基于ssd卷积网络的大田作物病虫害检测方法
CN111507253B (zh) 基于人工智能的陈列物品审核方法、装置
CN114359738B (zh) 一种跨场景鲁棒的室内人数无线检测方法及系统
US10970128B2 (en) Server, air conditioner and method for controlling thereof
CN110234085A (zh) 基于对抗迁移网络的室内位置指纹地图生成方法及系统
CN107360548A (zh) 基于rfid的室内物品定位方法和系统
CN111274972B (zh) 基于度量学习的菜品识别方法及装置
CN110781595B (zh) 能源使用效率pue的预测方法、装置、终端及介质
Kong et al. Pattern mining saliency
CN113915840B (zh) 一种冰箱及其食材定位方法
Kuo et al. Hybrid of artificial immune system and particle swarm optimization-based support vector machine for Radio Frequency Identification-based positioning system
CN112232334B (zh) 一种智能售货商品识别检测方法
CN113915841B (zh) 一种冰箱及其食材定位方法
US20230389474A1 (en) Method for determining a fruit to be harvested and a device for harvesting a fruit
CN109239661A (zh) 一种基于深度q网络的rfid室内定位系统及算法
WO2023065933A1 (zh) 冰箱、冰箱管理系统及冰箱管理系统的控制方法
CN109448024A (zh) 利用深度数据构建约束相关滤波器的视觉跟踪方法、系统
Lee et al. An incremental nonparametric Bayesian clustering-based traversable region detection method
WO2024130078A1 (en) Computer vision methods, systems, and devices for inferring counts of occluded objects
US20220147300A1 (en) Retrieving personalized visual content items in real time for display on digital-content-display devices within a physical space
CA3132106A1 (en) Automated product localization through mobile data analysis

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22882544

Country of ref document: EP

Kind code of ref document: A1

WWE Wipo information: entry into national phase

Ref document number: 202280052817.6

Country of ref document: CN

NENP Non-entry into the national phase

Ref country code: DE