Automatically adapt the packaging system to the packaging equipment based on voice control

Abstract

The present invention discloses an automatic adaptive packaging system based on voice-controlled packaging equipment, which relates to the field of voice control and comprises: an acquisition module, a database module, a parsing module and a processing module; the acquisition module monitors the sound source in a set area through a microphone array, and obtains target sound information through sound source feature matching; the database module stores data such as control instructions, sound source vectors, and an instruction-scene association matrix; the parsing module obtains an operation instruction set by matching text information with a mapping set; the processing module screens the operation instruction set to obtain a preprocessing set, and performs different operations based on the number of elements in the preprocessing set. The present invention can reduce the probability of executing erroneous instructions and significantly improve the accuracy of voice recognition.

Description

Technical Field

[0001] The present invention relates to the field of voice control technology, and in particular to an automatic adaptation packaging system for packaging equipment based on voice control. Background Art

[0002] In modern industrial production and logistics, packaging operations are a critical step before product circulation. Their efficiency and accuracy are crucial to overall operations. While automated packaging equipment is currently available on the market, it still faces many limitations in terms of operational convenience and intelligence. For example, operators need to manually enter various parameters on the equipment control panel to adapt to packaging tasks of different specifications. This process is cumbersome and prone to errors. Frequent parameter adjustments can significantly reduce work efficiency, especially when packaging a large number of items of varying specifications.

[0003] At the same time, voice control technology has made significant progress in recent years and has been widely used in smart home appliances, mobile devices, in-vehicle systems and other fields, demonstrating its advantages of convenient and efficient interaction. Introducing voice control technology into packaging equipment is expected to bring a new revolution to packaging operations;

[0004] However, existing voice control technologies still have some drawbacks in adapting to packaging equipment:

[0005] When a target device administrator issues a voice command that matches multiple control commands in the system, the multiple control commands cannot be filtered based on the target device's environmental data and operating data, resulting in the execution of the wrong control command;

[0006] How to improve the accuracy of recognizing voice commands is the problem we need to solve. To this end, we now provide an automatic adaptation packaging system based on voice-controlled packaging equipment. Summary of the Invention

[0007] The purpose of the present invention is to provide a packaging system that automatically adapts packaging equipment based on voice control.

[0008] The purpose of the present invention can be achieved by the following technical solution: Automatically adapting a packaging system to a packaging device based on voice control, comprising:

[0009] Acquisition module: monitors each sound source within the target device's set area and receives target sound information;

[0010] Database module: stores control instructions and their corresponding control sets, sound source vectors and management levels of target device managers, control levels of control instructions, mapping sets, and instruction-scenario association matrices;

[0011] Parsing module: processes the received target sound information to obtain text information, compares the text information with the mapping set stored in the database, and obtains the operation instruction set of the target sound information;

[0012] Processing module: Filters the elements in the operation instruction set to obtain a preprocessing set, and performs corresponding operations based on the number of elements in the preprocessing set.

[0013] Preferably, the process of monitoring each sound source within the target device setting area and receiving the target sound information is:

[0014] Set up several collection nodes in the set area of ​​the target device, place microphone arrays on the collection nodes, and collect sound information from various sound sources in the set area;

[0015] When a sound source is detected to have issued a wake-up command, the sound source features of the sound source are extracted, a similarity vector of the sound source features is obtained, and based on the similarity vector, whether the sound source is the target sound source is determined;

[0016] If the sound source is a target sound source, the sound emitted by the sound source is regarded as the target sound information and received; otherwise, the sound information emitted by the sound source is not received.

[0017] Preferably, the process of obtaining the instruction-scene association matrix is:

[0018] A historical collection interval is preset to collect environmental data and operating data of the target device corresponding to each control instruction during the historical collection interval;

[0019] The environmental data and operation data corresponding to each control instruction are processed to obtain a classified control vector. The classified control vector is classified using the K-means clustering analysis algorithm to obtain the implicit environment corresponding to each control instruction.

[0020] The number of occurrences of various control instructions in each implicit environment is obtained to obtain the instruction-scene association matrix.

[0021] Preferably, the process of obtaining the mapping set is:

[0022] Extract all elements from the control set and arrange them in a row as the control permutation. Use the Word2Vec model to map all elements in the control set to the vector space. Obtain the Euclidean distance between the first element and the remaining elements in the control permutation one by one in the vector space.

[0023] A Euclidean distance threshold is preset. If the Euclidean distance between the first element in the control arrangement and an element in the remaining elements is less than the Euclidean distance threshold, the first element is determined to be similar in meaning to the element.

[0024] The control array is updated by removing from the control array all elements that are similar in meaning to the first element and the first element in the remaining elements;

[0025] The first element in the control arrangement is used as an element in the mapping set, and an association relationship is established between the element and all elements with similar meanings and the control instructions corresponding to the element itself; and so on, the operation is performed on the updated control arrangement until there are no elements in the control arrangement, thereby obtaining a mapping set.

[0026] Preferably, the process of obtaining the operation instruction set is:

[0027] Preprocessing and text conversion of target sound information to obtain text information;

[0028] Perform word segmentation on the text information to obtain the corresponding word segmentation set, and match the filtering control instruction set corresponding to each element in the word segmentation set from the database through the mapping set;

[0029] An intersection operation is performed on all the obtained screening control instruction sets to obtain an operation instruction set of the target sound information.

[0030] Preferably, the process of screening the elements in the operation instruction set to obtain the preprocessing set is:

[0031] The control level of each element in the operation instruction set and the management level of the manager corresponding to the target sound information are obtained, and a preprocessing set is obtained based on the comparison result of the control level of the element in the operation instruction set and the management level of the manager corresponding to the target sound information.

[0032] Preferably, based on the number of elements in the preprocessing set, the process of performing the corresponding operation is:

[0033] D1: If the number of elements in the preprocessing set is 0, the user is given voice feedback: "There is no such operation or the user does not have sufficient permissions";

[0034] D2: If the number of elements in the preprocessing set is 1, then execute the operation corresponding to the element;

[0035] D3: If the number of elements in the preprocessing set is greater than 1, the environmental data and operating data of the target device at the current time point are collected and processed to obtain the scenario probability vector at the current time point. The correction factor of each element in the preprocessing set is obtained through the scenario probability vector and the instruction-scenario association matrix, and the operation corresponding to the element with the largest correction factor is executed.

[0036] Compared with the prior art, the present invention has the following beneficial effects:

[0037] The present invention maps the control sets corresponding to all control instructions into a vector space through the Word2Vec model, and obtains the mapping set in the vector space by threshold comparison, so that the packaging system can recognize different expressions of the same semantics, thereby improving the fault tolerance and recognition accuracy of voice instructions;

[0038] The present invention converts target sound information into text information, processes the text information to obtain a corresponding word segmentation set, matches the mapping set to a screening control instruction set of each element in the word segmentation set, and finally takes the intersection of all screening control instruction sets as an operation instruction set of the target sound information. A preprocessing set is further obtained by authority comparison. When there are multiple elements in the preprocessing set, the environmental data and operation data of the target device at the current time point are collected, and the correction factor of each element in the preprocessing set is obtained by combining the instruction-scenario association matrix. The preprocessing set is further screened and the instruction is executed based on the size of the correction factor, which significantly improves the accuracy of voice instruction recognition of the packaging device. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments described in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0040] Figure 1 This is a schematic diagram of the present invention. DETAILED DESCRIPTION

[0041] like Figure 1 As shown, the voice-controlled packaging device automatically adapts to the packaging system, including:

[0042] Acquisition module: used to monitor each sound source within the target device's set area and receive target sound information;

[0043] Specifically, the center point of the target device is obtained, the monitoring range r is preset, and a circle is drawn with the center point as the center and the monitoring range r as the radius, thereby obtaining the set area of ​​the target device;

[0044] Set up several collection nodes in the set area of ​​the target device, place microphone arrays on the collection nodes, and collect sound information from various sound sources in the set area;

[0045] When a sound source is detected to have issued a wake-up command, the sound source features of the sound source are extracted, a similarity vector of the sound source features is obtained, and based on the similarity vector, whether the sound source is the target sound source is determined;

[0046] Specifically, a wake-up command is a command or signal used to activate a target device from a dormant, standby, or inactive state to an operational active state. Examples include the "Xiao Ai" command on Xiaomi speakers and the "Hello, Xiao Di" command on BYD's car system.

[0047] Specifically, the sound source features include: Mel-frequency cepstral coefficients, linear prediction cepstral coefficients, fundamental frequency, and formant;

[0048] Specifically, the sound source features are combined into a vector in a certain order to obtain a sound source vector, the cosine similarity between the sound source vector and all the sound source vectors stored in the database is calculated one by one, and the obtained cosine similarity values ​​are combined to obtain a similarity vector of the sound source feature;

[0049] A similarity threshold is preset. If there is an element in the similarity vector that is greater than or equal to the similarity threshold, the sound source corresponding to the sound source feature is determined to be the target sound source. Otherwise, the sound source corresponding to the sound source feature is determined not to be the target sound source.

[0050] If the sound source is the target sound source, the sound emitted by the sound source is regarded as the target sound information and received; otherwise, the sound information emitted by the sound source is not received;

[0051] Database module: used to store control instructions and their corresponding control sets, control levels of control instructions, mapping sets, sound source vectors and management levels of target device managers, and instruction-scenario association matrices;

[0052] In detail, the process of obtaining the control set corresponding to the control instruction is as follows: performing word segmentation processing on the control instruction to obtain a word set of the control instruction, and using the word set as the control set of the control instruction;

[0053] The steps to obtain the mapping set are as follows: extract all elements in the control set and arrange them in a row as the control arrangement. Use the Word2Vec model to map all elements in the control set to the vector space. In the vector space, obtain the Euclidean distance between the first element and the remaining elements of the control arrangement one by one. Preset the Euclidean distance threshold. If the Euclidean distance between the first element and an element in the remaining elements is less than the Euclidean distance threshold, then the first element is determined to be similar in meaning to the element.

[0054] All elements with similar meanings to the first element and the first element in the remaining elements of the control array are removed, thereby updating the control array. The first element in the control array is used as an element in the mapping set, and associations are established between the element and all elements with similar meanings to it and the control instructions corresponding to the element itself. Similarly, the same operation is performed on the updated control array until there are no elements in the control array, thereby obtaining a mapping set.

[0055] Specifically, the remaining elements refer to all elements except the first element in the control arrangement;

[0056] Specifically, the control level of control instructions is divided into four levels: S, A, B, C, and D. The management level of the administrator of the target device is also divided into four levels: S, A, B, C, and D, where S>A>B>C>D. If the administrator's management level is A, then he can manage control instructions of levels A, B, C, and D, but cannot control control instructions of level S, and so on.

[0057] In detail, the specific steps to obtain the instruction-scene association matrix are:

[0058] A historical collection interval is preset to collect environmental data and operating data of the target device corresponding to each control instruction during the historical collection interval;

[0059] Specifically, environmental data includes: ambient temperature and humidity, light intensity, noise level, number of people in the target device’s set area, people’s movement speed, time of day, day of the week, and season;

[0060] Operational data includes: device temperature, operating memory usage, operating current, operating voltage, vibration acceleration, latency, packet loss rate, signal strength, and elapsed operating time;

[0061] Specifically, a set time interval before the control instruction is executed is obtained as the collection interval, and collection nodes are set in the collection interval. The average decibel value at each collection node is obtained as the noise level; the average number of people at each collection node is obtained as the number of people in the set area of ​​the target device; and the average movement speed of all people at all collection nodes is obtained as the movement speed of people.

[0062] The environmental data and operation data corresponding to each control instruction are processed to obtain a classified control vector. The classified control vector is classified using the K-means clustering analysis algorithm to obtain the implicit environment corresponding to each control instruction.

[0063] In detail, the environmental data and the operating data are combined in a fixed order to obtain the classified control vector corresponding to each control instruction;

[0064] Specifically, after the classification control vector is classified using the K-means clustering analysis algorithm, a series of clusters are obtained. Each cluster is set to correspond to a hidden environment, and the hidden environments are numbered. The number is represented by j, j = 1, 2...k, and k represents the total number of hidden environments. The hidden environment of the cluster where the classification control vector is located is used as the hidden environment of the control instruction corresponding to the classification control vector;

[0065] Obtain the number of occurrences of various control instructions in each implicit environment and obtain the instruction-scene association matrix;

[0066] Specifically, for a certain implicit environment, the frequency value of each control instruction in the implicit environment is obtained as the correlation between the control instruction and the implicit environment. The correlation between all control instructions and all implicit environments is filled into the matrix to obtain the instruction-scene correlation matrix;

[0067] Specifically, the frequency value of a certain control instruction in a certain implicit environment = the number of the control instructions in the implicit environment / the total number of control instructions in the implicit environment;

[0068] Specifically, the control instructions are numbered, and the number is represented by i, i=1,2...u, where u represents the number of types of control instructions. Each row in the instruction-scenario association matrix represents a control instruction, and each column represents an implicit environment. The element with the subscript ij in the matrix represents the correlation between the control instruction numbered i and the implicit environment numbered j;

[0069] Parsing module: processes the received target sound information to obtain text information, compares the text information with the mapping set stored in the database, and obtains the operation instruction set of the target sound information;

[0070] Preprocessing and text conversion of target sound information to obtain text information;

[0071] Specifically, the preprocessing includes filtering and denoising; the target sound information after preprocessing is processed using the language model built into the target device to obtain text information;

[0072] Perform word segmentation on the text information to obtain the corresponding word segmentation set, and match the filtering control instruction set corresponding to each element in the word segmentation set from the database through the mapping set;

[0073] In detail, Jieba word segmentation is used to segment text information;

[0074] Obtain the spatial distance between the first element in the word segmentation set and all elements in the mapping set, obtain the corresponding distance vector, preset the spatial distance threshold, and obtain the corresponding screening control instruction set based on the comparison result between the distance vector and the spatial distance threshold;

[0075] Specifically, the first element in the word segmentation set and all elements in the mapping set are mapped to the vector space using the same model, such as the Word2Vec model. The Euclidean distance between the elements in the word segmentation set and the elements in the mapping set is obtained in the vector space as the spatial distance. Alternatively, the cosine similarity between the elements in the word segmentation set and the elements in the mapping set can be obtained in the vector space as the spatial distance, which is not limited in this embodiment.

[0076] Combine all the obtained spatial distances and express them in the form of vectors to obtain the distance vector;

[0077] Extract the elements in the distance vector one by one and compare them with the spatial distance threshold. Mark the elements in the distance vector that are smaller than the spatial distance threshold. After the comparison is completed, use the control instructions corresponding to all marked elements in the distance vector as the filtering control instruction set corresponding to the first element in the word segmentation set.

[0078] Similarly, obtain the filtering control instruction set corresponding to other elements in the word segmentation set;

[0079] Performing an intersection operation on all acquired screening control instruction sets to obtain an operation instruction set for the target sound information;

[0080] Processing module: Filters the elements in the operation instruction set to obtain a preprocessing set, and performs corresponding operations based on the number of elements in the preprocessing set;

[0081] S1: Obtaining the control level of each element in the operation instruction set and the management level of the manager corresponding to the target sound information, and obtaining a preprocessing set based on the comparison result of the control level of the element in the operation instruction set and the management level of the manager corresponding to the target sound information;

[0082] Specifically, if the control level of an element in the operation instruction set is less than or equal to the management level of the manager corresponding to the target sound information, the element is classified into the pre-processing set;

[0083] S2: If the number of elements in the preprocessing set is 0, a voice feedback message "There is no such operation or the user does not have sufficient permissions" is given to the user;

[0084] S3: If the number of elements in the preprocessing set is 1, then execute the operation corresponding to the element;

[0085] S4: If the number of elements in the preprocessing set is greater than 1, the environmental data and operating data of the target device at the current time point are collected and processed to obtain the scenario probability vector at the current time point. The correction factor of each element in the preprocessing set is obtained by using the scenario probability vector and the instruction-scenario association matrix, and the operation corresponding to the element with the largest correction factor is executed;

[0086] Specifically, based on the implicit environments corresponding to each control instruction collected in the historical collection interval, label the classification control vectors corresponding to each control instruction to obtain labeled classification control vectors, and use the labeled classification control vectors to train a convolutional neural network to obtain a scene model;

[0087] Specifically, the training of a convolutional neural network model using labeled data is a well-known technology, and will not be elaborated in this embodiment;

[0088] Process the environmental data and operation data at the current time point to obtain the classification control vector at the current time point, and input the classification control vector at the current time point into the scene model to obtain a scene probability vector;

[0089] Specifically, inputting the classification control vector at the current time point into the scene model will obtain the probability that the current time point belongs to each implicit environment, and combine the probabilities that the current time point belongs to each implicit environment to obtain the scene probability vector corresponding to the current time point;

[0090] Obtain the correction factor of each element in the preprocessing set one by one;

[0091] Specifically, use the formula:

[0092]

[0093] Obtain the correction factor of each element in the preprocessing set, where w represents the element in the preprocessing set, and GL(w丨j) represents the association degree between this element and the implicit environment numbered j, represents the value of the implicit environment numbered j in the scene probability vector;

[0094] The above is only a preferred embodiment of the present invention, and does not impose any form of limitation on the present invention. Although the present invention has been disclosed above with preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some changes or modifications within the scope of the technical solution of the present invention to make equivalent changes to equivalent embodiments. However, as long as it does not depart from the content of the technical solution of the present invention, any modification or equivalent replacement made to the above embodiments based on the technical essence of the present invention still falls within the scope of the technical solution of the present invention.

Claims

1. Automatically adapt the packaging system to the packaging equipment based on voice control, characterized by: include: Acquisition module: monitors each sound source within the target device's set area and receives target sound information; Database module: stores control instructions and their corresponding control sets, sound source vectors and management levels of target device managers, control levels of control instructions, mapping sets, and instruction-scenario association matrices; Parsing module: processes the received target sound information to obtain text information, compares the text information with the mapping set stored in the database, and obtains the operation instruction set of the target sound information; Processing module: Filters the elements in the operation instruction set to obtain a preprocessing set, and performs corresponding operations based on the number of elements in the preprocessing set; The process of obtaining the instruction-scene association matrix is: A historical collection interval is preset to collect environmental data and operating data of the target device corresponding to each control instruction during the historical collection interval; Processing the environmental data and operating data corresponding to each control instruction to obtain a classified control vector, classifying the classified control vector to obtain the implicit environment corresponding to each control instruction; Obtain the number of occurrences of various control instructions in each implicit environment and obtain the instruction-scene association matrix; Based on the number of elements in the preprocessing collection, the process of performing the corresponding operation is: D1: If the number of elements in the preprocessing set is 0, a voice feedback message "No such operation or insufficient user authority" is given to the user; D2: If the number of elements in the preprocessing set is 1, then execute the operation corresponding to the element; D3: If the number of elements in the preprocessing set is greater than 1, the environmental data and operating data of the target device at the current time point are collected and processed to obtain the probability that the current time point belongs to each implicit environment. The probabilities of the current time point belonging to each implicit environment are combined to obtain the scene probability vector at the current time point. The correction factor of each element in the preprocessing set is obtained through the scene probability vector and the instruction-scene association matrix, and the operation corresponding to the element with the largest correction factor is executed.

2. The automatic adaptive packaging system for packaging equipment based on voice control according to claim 1 is characterized in that: The process of monitoring each sound source within the target device's set area and receiving the target sound information is as follows: Set up several collection nodes in the set area of ​​the target device, place microphone arrays on the collection nodes, and collect sound information from various sound sources in the set area; When a sound source is detected to have issued a wake-up command, the sound source features of the sound source are extracted, a similarity vector of the sound source features is obtained, and based on the similarity vector, whether the sound source is the target sound source is determined; If the sound source is a target sound source, the sound emitted by the sound source is regarded as the target sound information and received; otherwise, the sound information emitted by the sound source is not received.

3. The automatic adaptive packaging system for packaging equipment based on voice control according to claim 2 is characterized in that: The process of obtaining the instruction-scene association matrix uses the K-means clustering analysis algorithm to classify the classification control vectors.

4. The automatic adaptive packaging system for packaging equipment based on voice control according to claim 3 is characterized in that: The process of obtaining the mapping set is: Extract all elements from the control set and arrange them in a row as the control permutation. Use the Word2Vec model to map all elements in the control set to the vector space. Obtain the Euclidean distance between the first element and the remaining elements in the control permutation one by one in the vector space. A Euclidean distance threshold is preset. If the Euclidean distance between the first element in the control arrangement and an element in the remaining elements is less than the Euclidean distance threshold, the first element is determined to be similar in meaning to the element. The control array is updated by removing from the control array all elements that are similar in meaning to the first element and the first element in the remaining elements; The first element in the control arrangement is used as an element in the mapping set, and an association relationship is established between the element and all elements with similar meanings and the control instructions corresponding to the element itself; and so on, the operation is performed on the updated control arrangement until there are no elements in the control arrangement, thereby obtaining a mapping set.

5. The automatic adaptive packaging system for packaging equipment based on voice control according to claim 4 is characterized in that: The process of obtaining the operation instruction set is: Preprocessing and text conversion of target sound information to obtain text information; Perform word segmentation on the text information to obtain the corresponding word segmentation set, and match the filtering control instruction set corresponding to each element in the word segmentation set from the database through the mapping set; An intersection operation is performed on all the obtained screening control instruction sets to obtain an operation instruction set of the target sound information.

6. The automatic adaptive packaging system for packaging equipment based on voice control according to claim 5 is characterized in that: The process of filtering the elements in the operation instruction set to obtain the preprocessing set is as follows: The control level of each element in the operation instruction set and the management level of the manager corresponding to the target sound information are obtained, and a preprocessing set is obtained based on the comparison result of the control level of the element in the operation instruction set and the management level of the manager corresponding to the target sound information.

Images