A robot-based book search method and system

Abstract

The present invention discloses a robot-based book search method and system, which performs semantic analysis on search information and extracts keywords in the search, thereby determining the theme of the book according to the keywords based on a recurrent neural network and a convolutional neural network, quickly generating a candidate book list, and calculating the weight corresponding to each book in the book list as the full radius through historical search results, thereby screening and sorting the book list, making the book search more accurate, so that users can quickly find target books and improve the accuracy of book search; in addition, the robot can be controlled to move quickly to the target position through the position information of the target book, so as to accurately grab the target book, avoid users or administrators with large collections from being unable to quickly obtain the target book, and facilitate the administrator to manage the books in the library, reduce the manpower and material resources for book search, and improve the efficiency of book search.

Description

Technical Field

[0001] The present invention relates to the technical field of robot control, and in particular to a robot-based book searching method and system. Background Art

[0002] At present, the development of libraries is gradually tending towards large-scale, and the task of book management will become more and more onerous. Repetitive work such as cleaning, putting books on and off shelves, moving books, and classifying books consumes a lot of time and energy of administrators, and the contradiction between collection and management mode is becoming increasingly serious. At the same time, when readers borrow books from the library, it is usually difficult to find the books they need quickly and accurately in the face of the huge collection.

[0003] Existing book search methods are generally based on RFID radio frequency technology. Each book needs to be equipped with an RFID tag so as to determine the approximate range of the book based on the RFID radio frequency technology. Although book search can be achieved to a certain extent, due to the large number of books in the library, if each book is labeled, it will consume a lot of manpower and material resources. Summary of the invention

[0004] The present invention provides a robot-based book search method and system to solve the problems of high cost and low efficiency of existing book search technologies.

[0005] In order to solve the above technical problems, an embodiment of the present invention provides a robot-based book search method, comprising:

[0006] Acquire search information input by a target user, perform semantic analysis on the search information according to a preset semantic analysis model, and acquire a plurality of target keywords; wherein the semantic analysis model is used to identify the semantic characteristics of each word in the search information;

[0007] Extracting features from historical data in the library based on a recurrent neural network and a convolutional neural network to obtain a book recognition model; inputting the target keyword into the book recognition model to generate a first candidate book list;

[0008] Acquire the historical search records of the target user, and determine the first book list in the first candidate book list according to the historical search records; at the same time, acquire the user portrait of the target user, and determine the weight value of each book in the first book list according to the user portrait;

[0009] The weight values ​​are used as the radii of the circles corresponding to the books, respectively, to determine the number of intersections generated by the circles corresponding to the books in the first book list, and the books whose intersections reach a preset number are grouped together to generate a second candidate book list; the historical search record includes historical search information, a historical candidate book list and a historical target book;

[0010] In response to the target user's selection action on the second candidate book list, determining a target book and acquiring location information of the target book;

[0011] The robot is controlled to reach the target position and grab the book according to the position information and the library obstacle information.

[0012] The present invention performs semantic analysis on the search information and extracts the keywords in the search, thereby determining the theme of the book according to the keywords based on the recurrent neural network and the convolutional neural network, quickly generating a list of candidate books, and calculating the weight corresponding to each book in the book list as the total radius through the historical search results, thereby screening and sorting the book list, making the book search more accurate, so that the user can quickly find the target book and improve the accuracy of book search; in addition, the robot can be controlled to move quickly to the target position through the position information of the target book, so as to accurately grab the target book, avoid users or administrators with large collections from being unable to quickly obtain the target book, and facilitate the administrator to manage the books in the library, reduce the manpower and material resources for book search, and improve the efficiency of book search.

[0013] Furthermore, the retrieval information input by the target user is obtained, and a semantic analysis operation is performed on the retrieval information according to a preset semantic analysis model to obtain a number of target keywords, specifically:

[0014] Building a dictionary base based on books in the library, wherein the dictionary base includes nouns and participles;

[0015] Acquire search information input by a target user, input the search information into the semantic analysis model, identify segmentations and symbols in the search information, and generate a number of initial keywords;

[0016] The initial keywords are corrected according to the dictionary library to generate a plurality of target keywords.

[0017] Furthermore, the feature extraction of historical data in the library based on the recurrent neural network and the convolutional neural network is performed to obtain a book recognition model, which is specifically:

[0018] Collecting sample data, and dividing the sample data into a training set and a validation set according to a preset ratio; the sample data includes search records of all users stored in the database;

[0019] Building a book recognition model based on a long short-term memory network and a convolutional neural network, training the book recognition model according to the training set and verification, and obtaining a trained book recognition model;

[0020] According to singular value decomposition, all linear layers in the trained book recognition model are subjected to low-rank processing, and the book recognition model after low-rank processing is output.

[0021] Furthermore, before inputting the target keyword into the book recognition model, the method further includes: optimizing the book recognition model based on the historical search records of the target user to obtain an optimized book recognition model, specifically:

[0022] Acquire a book recognition model, and adjust characteristic parameters of the book recognition model according to the number of target keywords, wherein the characteristic parameters include the number of convolutional layers, convolutional layer parameters, and long short-term memory network parameters;

[0023] Calculate the accuracy between the historical search results and the corresponding historical target books according to the historical search records; and optimize and update the book recognition model according to the accuracy to obtain an optimized book recognition model;

[0024] The target keyword is input into the optimized book recognition model to generate a first candidate book list.

[0025] Furthermore, the target keyword is input into the optimized book recognition model to generate a first candidate book list, specifically:

[0026] Input the target keyword into a long short-term memory network to extract a first feature vector set; and input the first feature vector set into a convolutional neural network to generate a candidate feature vector set;

[0027] The candidate feature vector set is normalized according to a normalization algorithm, the prediction probability of each candidate feature vector in the candidate feature vector set is calculated, and a first candidate book list is generated.

[0028] Furthermore, after determining the target book in response to the target user's selection action on the second candidate book list, the method further includes:

[0029] The search information, the first candidate book list and the historical target books of this search are packaged into sample data according to a preset format and stored in the database;

[0030] Calculating the accuracy of the book recognition model according to the historical search records and the accuracy calculation algorithm; the historical search results include the number of historical searches, the historical candidate book list and the prediction probability corresponding to the historical candidate book list;

[0031] updating the book recognition model according to a preset accuracy threshold and the accuracy of the book recognition model;

[0032] The accuracy calculation algorithm is specifically:

[0033]

[0034] Among them, n is the number of historical search predictions, p (i,max) is the probability value with the maximum predicted probability during the i-th search, p (i,x) is the probability value corresponding to the target book during the i-th search, and x is the ranking position of the target book in the first candidate book list during the i-th search.

[0035] Furthermore, the position information includes the operation point position information and the target position information; and the robot is controlled to reach the target position to complete the book grabbing according to the position information and the library obstacle information, specifically:

[0036] Planning a robot movement path according to the operation point location information and the obstacle information of the library, and controlling the robot to reach a first target point according to the robot movement path;

[0037] A second target point is determined according to the target position information, and a first operation sequence of the robot working arm is generated according to the second target point and the robot position information; and the target point sequence is corrected according to the obstacle information to generate a second operation sequence, and the robot's working arm is controlled to move to the second target point and to grab the target book according to the second operation sequence.

[0038] Furthermore, planning the robot movement path according to the operation point location information and the obstacle information of the library specifically includes:

[0039] Obtaining obstacle information of the library and first real-time position information of the robot, wherein the point cloud information includes position data of all obstacle information in the library;

[0040] Determine a first target point according to the position information of the operation point, and plan a movement path of the robot according to the position data of all obstacle information in the library and the first real-time position information of the robot; when the obstacle is a book, set the safety distance between the robot and the position data as a first distance threshold; when the obstacle is not a book, set the safety distance between the robot and the position data as a second distance threshold;

[0041] The robot is controlled to reach a first target point according to the robot movement path.

[0042] Further, determining a second target point according to the target position information, and generating a first operation sequence of the robot working arm according to the second target point and the robot position information, specifically, is:

[0043] Acquire the second real-time position information of the robot and the real-time position information of each joint of the working arm, and generate the coordinate mapping relationship between each joint of the working arm and the robot according to the second real-time position information and the real-time position information of the working arm;

[0044] A second target point is determined according to the target position information, and a first operation sequence of the operating arm is generated according to the second target point and the coordinate mapping relationship.

[0045] In a second aspect, the present invention provides a robot-based book search system, comprising: a semantic analysis module, a recognition module, a weight distribution module, a reordering module, a target confirmation module and a job control module;

[0046] The semantic analysis module is used to obtain the search information input by the target user, perform semantic analysis on the search information according to a preset semantic analysis model, and obtain a number of target keywords; wherein the semantic analysis model is used to identify the semantic characteristics of each word in the search information;

[0047] The recognition module is used to extract features from historical data in the library based on a recurrent neural network and a convolutional neural network to obtain a book recognition model; the target keyword is input into the book recognition model to generate a first candidate book list;

[0048] The weight allocation module is used to obtain the historical search records of the target user, and determine the first book list in the first candidate book list according to the historical search records; at the same time, obtain the user portrait of the target user, and determine the weight value of each book in the first book list according to the user portrait;

[0049] The reordering module is used to use the weight value as the radius of the circle corresponding to the book, determine the number of intersections generated by the circles corresponding to the books in the first book list, collect the books whose number of intersections reaches a preset number, and generate a second candidate book list; the historical search record includes historical search information, a historical candidate book list and a historical target book;

[0050] The target confirmation module is used to determine the target book in response to the target user's selection action on the second candidate book list, and obtain the location information of the target book;

[0051] The operation control module is used to control the robot to reach the target position and complete book grabbing according to the position information and library obstacle information. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 A schematic diagram of a flow chart of a robot-based book search method provided in an embodiment of the present invention;

[0053] Figure 2 A schematic structural diagram of a robot-based book search system provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0054] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0055] Embodiment 1

[0056] Please refer to Figure 1 , Figure 1 A flowchart of a robot-based book search method provided in an embodiment of the present invention includes steps 101 to 106, which are as follows:

[0057] Step 101: obtaining search information input by a target user, performing semantic analysis on the search information according to a preset semantic analysis model, and obtaining a number of target keywords; wherein the semantic analysis model is used to identify the semantic characteristics of each word in the search information;

[0058] In this embodiment, the step of obtaining the search information input by the target user and performing semantic analysis on the search information according to a preset semantic analysis model to obtain a number of target keywords is as follows:

[0059] Building a dictionary base based on books in the library, wherein the dictionary base includes nouns and participles;

[0060] Acquire search information input by a target user, input the search information into the semantic analysis model, identify segmentations and symbols in the search information, and generate a number of initial keywords;

[0061] The initial keywords are corrected according to the dictionary library to generate a plurality of target keywords.

[0062] In this embodiment, a dictionary library is constructed according to the books in the library, and the dictionary library includes a stop dictionary, a noun dictionary, a forward dictionary, a reverse dictionary and a core dictionary. The stop dictionary is used to store stop words. Stop words are used to assist the word segmentation tool in segmenting words. They are manually pre-stored in the stop word list. They generally have no practical meaning and are mostly modal particles, conjunctions or prepositions, such as also, ah, again, than, with, but. The noun dictionary is used to store commodity nouns, character nouns and exclusive nouns, etc. Commodity nouns include brand nouns, model nouns and / or category nouns. When the commodity is a mobile terminal, the commodity name can be apple, iphone, iphone4s, Samsung, Huawei, Honor 6, nokia110. The forward dictionary is used to store auxiliary words with positive semantics, such as how much money, price, quotation, contract, new function, and physical store. The reverse dictionary is used to store auxiliary words with reverse semantics, such as flower screen, talk about water, not high, how to think, and tutorial. The core dictionary uses the core dictionary of the word segmentation tool IKAnalyzer. Words in the stop dictionary, basic dictionary, forward dictionary and reverse dictionary are also added to the core dictionary.

[0063] In this embodiment, a semantic analysis model is constructed, the semantic analysis model includes a target detection model, and the semantic analysis model is trained according to the corpus. The target detection model is used to perform semantic analysis on short sentences or long sentences, identify the semantic characteristics of each word in the search information, and identify and classify the symbols and words in the search information to eliminate the symbols and stop words in the short sentences or long sentences and generate initial keywords.

[0064] In this embodiment, the target user needs to log in on the search system page and enter the search information before searching for books. After the target user logs in, the system can obtain the user portrait and historical search records of the target user.

[0065] In this embodiment, the search information input by the target user is obtained, the detection information search information is input into the semantic analysis model, the part-of-speech category and symbols of each phrase in the search information are identified, and stop words and symbols in the search information are proposed to separate each individual phrase and generate a number of initial keywords.

[0066] In this embodiment, the initial keywords are further refined by performing index comparison in the dictionary library based on the initial keywords, and the keywords that are not separated in the initial keywords are divided, thereby generating a plurality of target keywords.

[0067] Step 102: extracting features from historical data in the library based on a recurrent neural network and a convolutional neural network to obtain a book recognition model; inputting the target keyword into the book recognition model to generate a first candidate book list;

[0068] In this embodiment, the feature extraction of historical data in the library based on the recurrent neural network and the convolutional neural network is performed to obtain a book recognition model, which is specifically:

[0069] Collecting sample data, and dividing the sample data into a training set and a validation set according to a preset ratio; the sample data includes search records of all users stored in the database;

[0070] Building a book recognition model based on a long short-term memory network and a convolutional neural network, training the book recognition model according to the training set and verification, and obtaining a trained book recognition model;

[0071] According to singular value decomposition, all linear layers in the trained book recognition model are subjected to low-rank processing, and the book recognition model after low-rank processing is output.

[0072] In this embodiment, the search records of all users are retrieved from the library database, and the search records include search information, corresponding candidate vector sets and target books.

[0073] In this embodiment, a book recognition module is constructed based on a long short-term memory network and a convolutional neural network, and the output of the long short-term memory network is connected to the input of the convolutional neural network.

[0074] In this embodiment, the book recognition model is iteratively trained according to the training set until a preset condition is met, and the iteration is stopped to complete the book recognition model training.

[0075] In this embodiment, all the predecessors in the trained book recognition model are decomposed into the product of several sub-matrices, and the singular values ​​of the predecessors are calculated according to the product of the sub-matrices, so that the singular values ​​are screened according to the retention formula; then, a second matrix is ​​generated according to the screened singular values, and the linear layer is updated according to the second matrix to complete the construction of the book recognition model.

[0076] In this embodiment, the linear layer in book recognition is first extracted, and singular value decomposition SVD is performed on the linear layer respectively, and the appropriate number of singular values ​​k is selected according to the retention formula to reconstruct the second matrix M~ for the singular values, and the second matrix M~ is used to replace the original linear layer M, thereby reducing the volatility of the model prediction and the number of model parameters by discarding a small number of unimportant parameters that are prone to cause interference errors.

[0077] As a specific example of an embodiment of the present invention, a linear fully connected layer is represented as a matrix M, whose shape is m*n, and SVD decomposition is used to decompose it into the product of three matrices M=USV T, where U is the left singular vector matrix, S is the singular value diagonal matrix, V is the right singular vector matrix, U and V are m*r matrices, S is an r*r matrix, and the stage singular value matrix is ​​obtained S~. According to the retention formula, the first k main singular values ​​are retained, and the approximate matrix M~ replaces the original linear layer. The retention formula is specifically:

[0078]

[0079] In this embodiment, m, n, and r are all matrix dimensions, indicating the size of each SVD matrix.

[0080] In this embodiment, before the target keyword is input into the book recognition model, the method further includes: optimizing the book recognition model based on the historical search records of the target user to obtain an optimized book recognition model, specifically:

[0081] Acquire a book recognition model, and adjust characteristic parameters of the book recognition model according to the number of target keywords, wherein the characteristic parameters include the number of convolutional layers, convolutional layer parameters, and long short-term memory network parameters;

[0082] Calculate the accuracy between the historical search results and the corresponding historical target books according to the historical search records; and optimize and update the book recognition model according to the accuracy to obtain an optimized book recognition model;

[0083] The target keyword is input into the optimized book recognition model to generate a first candidate book list.

[0084] In this embodiment, after the user logs in, a book recognition model is obtained, and characteristic parameters of the book recognition model are adjusted according to the number of target keywords. The characteristic parameters include the number of convolutional layers, convolutional layer parameters, and long short-term memory network parameters.

[0085] In this embodiment, the accuracy between the historical search results and the corresponding historical target books is calculated according to the historical search records, and the book recognition model is optimized and updated according to the accuracy.

[0086] In this embodiment, the accuracy calculation algorithm is specifically:

[0087]

[0088] Among them, n is the number of historical search predictions, p (i,max) is the probability value with the maximum predicted probability during the i-th search, p (i,x) is the probability value corresponding to the target book during the i-th search, and x is the ranking position of the target book in the first candidate book list during the i-th search.

[0089] In this embodiment, when the accuracy is lower than a preset first accuracy threshold, the search record of the user is used as a data sample, and the sample is expanded, and the book recognition model is trained with the expanded sample.

[0090] In this embodiment, the target keyword is input into the optimized book recognition model to generate a first candidate book list, specifically:

[0091] Input the target keyword into a long short-term memory network to extract a first feature vector set; and input the first feature vector set into a convolutional neural network to generate a candidate feature vector set;

[0092] The candidate feature vector set is normalized according to a normalization algorithm, the prediction probability of each candidate feature vector in the candidate feature vector set is calculated, and a first candidate book list is generated.

[0093] In this embodiment, the long short-term memory network includes two LSTM (Long Short-Term Memory Network) layers.

[0094] In this embodiment, the book recognition model includes an input layer, two LSTM (Long Short-Term Memory) layers, a convolutional neural network layer, and an output layer.

[0095] In this embodiment, the keyword is passed to the first LSTM layer through the input layer, the output of the first LSTM layer is used as the input of the second LSTM layer, and the output of the second LSTM layer is used as the input of the convolutional neural network layer. Convolution and pooling operations are performed in the convolutional neural network, and then the final prediction is performed through the fully connected layer.

[0096] In this embodiment, the output layer includes a Dense layer, whose output dimension is 20, and Sigmoid is used as the activation function, and the output result is 20 first candidate books.

[0097] Step 103: Obtain the historical search records of the target user, and determine the first book list in the first candidate book list according to the historical search records; at the same time, obtain the user portrait of the target user, and determine the weight value of each book in the first book list according to the user portrait;

[0098] In this embodiment, the historical search records and user portrait of the target user are obtained, the user's field of interest is determined based on the historical search records, the books in the first candidate book list are screened based on the field of interest, the top K books close to the field of interest are retained, and the first book list is generated.

[0099] Step 104: using the weight values ​​as the radius of the circles corresponding to the books, determining the number of intersections generated by the circles corresponding to the books in the first book list, and collecting the books whose number of intersections reaches a preset number to generate a second candidate book list; the historical search record includes historical search information, a historical candidate book list and a historical target book;

[0100] In this embodiment, weights are assigned to the books in the first book list according to the user portrait of the target user, so that the weights are used as the radius of the circle corresponding to the books, the number of intersections generated by the circles corresponding to the books in the first book list is determined, the books whose number of intersections reaches a preset number are grouped together to generate a second candidate book list, wherein the second candidate book list is arranged in descending order according to the number of intersections, and the second candidate book list is pushed to the interactive interface.

[0101] In this embodiment, the historical search data itself contains the books that the user has searched for before. The user's reading field and reading habits can be determined from the books that the user has searched for, so as to rank the first candidate books. The use of user portraits is actually to add a weight to each book in each first book list. The larger the weight, the larger the radius of the circle, the larger the intersection that can be generated by the circle, and the more intersections there are, the greater the probability that the book will be selected. Therefore, the first book list is screened and assembled according to the number of intersections to generate a second candidate book list.

[0102] In this embodiment, the first K books close to the field of interest are retained. After the first book list is generated, a second book list is generated for several books eliminated from the first candidate book list. If the user does not select a target book from the second candidate book list, the second book list is popped up in a small window on the interactive page for the user to select.

[0103] Step 105: In response to the target user's selection action on the second candidate book list, determining a target book and acquiring location information of the target book;

[0104] In this embodiment, after determining the target book in response to the target user's selection action on the second candidate book list, the method further includes:

[0105] The search information, the first candidate book list and the historical target books of this search are packaged into sample data according to a preset format and stored in the database;

[0106] Calculating the accuracy of the book recognition model according to the historical search records and the accuracy calculation algorithm; the historical search results include the number of historical searches, the historical candidate book list and the prediction probability corresponding to the historical candidate book list;

[0107] updating the book recognition model according to a preset accuracy threshold and the accuracy of the book recognition model;

[0108] The accuracy calculation algorithm is specifically:

[0109]

[0110] Among them, n is the number of historical search predictions, p (i,max) is the probability value with the maximum predicted probability during the i-th search, p (i,x) is the probability value corresponding to the target book during the i-th search, and x is the ranking position of the target book in the first candidate book list during the i-th search.

[0111] In this embodiment, after the user selects a book, the search information of this search, the first candidate book list and the historical target books are packaged into sample data according to a preset format and stored in the database to facilitate the background to update the model.

[0112] In this embodiment, the accuracy of the book recognition model is calculated, and if the accuracy is lower than a preset second accuracy threshold, the book recognition model is re-updated according to all data samples in the library database.

[0113] In this embodiment, through two accuracy calculations, the first calculation is set before the user starts searching, and whether personalized update is required is determined based on the first accuracy calculation, and the sample is expanded according to the user's own search record, so as to perform personalized adjustment and training on the user's book recognition model. The second calculation is set after the user's search is completed, and the second accuracy calculation is used to determine whether the book recognition model needs to be globally updated, and the book recognition model is re-updated according to all data samples in the library database to improve the accuracy of the book recognition model.

[0114] Step 106: Control the robot to reach the target position and complete book grabbing according to the position information and library obstacle information.

[0115] In this embodiment, the position information includes the operation point position information and the target position information; the robot is controlled to reach the target position to complete the book grabbing according to the position information and the library obstacle information, specifically:

[0116] Planning a robot movement path according to the operation point location information and the obstacle information of the library, and controlling the robot to reach a first target point according to the robot movement path;

[0117] A second target point is determined according to the target position information, and a first operation sequence of the robot working arm is generated according to the second target point and the robot position information; and the target point sequence is corrected according to the obstacle information to generate a second operation sequence, and the robot's working arm is controlled to move to the second target point and to grab the target book according to the second operation sequence.

[0118] In this embodiment, the visual servo information is converted into the control parameters of the robot, and the spatial position of the robot's mechanical working arm is determined according to the location information of the target book and the obstacle information, so as to control the robot to grasp the book.

[0119] In this embodiment, the location information of the target book is extracted through the point cloud information, and the location information includes the operation point information and the obstacle information. The robot moves from the initial position to the operation point in the first posture.

[0120] In this embodiment, planning the robot movement path according to the operation point location information and the obstacle information of the library specifically includes:

[0121] Obtaining obstacle information of the library and first real-time position information of the robot, wherein the point cloud information includes position data of all obstacle information in the library;

[0122] Determine a first target point according to the position information of the operation point, and plan a movement path of the robot according to the position data of all obstacle information in the library and the first real-time position information of the robot; when the obstacle is a book, set the safety distance between the robot and the position data as a first distance threshold; when the obstacle is not a book, set the safety distance between the robot and the position data as a second distance threshold;

[0123] The robot is controlled to reach a first target point according to the robot movement path.

[0124] In this embodiment, since books in the library are easily damaged, the distance between the robot and the obstacle needs to be set for the books. When the obstacle is a book, the safety distance is set to the first distance threshold, and the safety distance is larger at this time to prevent the robot from colliding and damaging the book. When the obstacle is not a book, the full distance is set to the second distance threshold, and the safety distance is smaller at this time.

[0125] In this embodiment, several robots are arranged in the library at different starting points. According to the straight-line distance between the operation point and each robot at the starting point, the robot with the smallest straight-line distance is selected as the robot for grabbing books this time.

[0126] A moving path is planned according to the starting point position and the working point position of the robot, so as to control the robot to reach the first target point.

[0127] In this embodiment, the second target point is determined according to the target position information, and the first operation sequence of the robot working arm is generated according to the second target point and the robot position information, specifically:

[0128] Acquire the second real-time position information of the robot and the real-time position information of each joint of the working arm, and generate the coordinate mapping relationship between each joint of the working arm and the robot according to the second real-time position information and the real-time position information of the working arm;

[0129] A second target point is determined according to the target position information, and a first operation sequence of the operating arm is generated according to the second target point and the coordinate mapping relationship.

[0130] In this embodiment, when the robot moves to the first target point according to the planned moving path, the second real-time position information of the robot and the real-time position information of each joint of the working arm are obtained, the kinematic coordinate system of the robot is restored according to the center point of the robot, and the kinematic parameters of the robot are determined according to the kinematic coordinate system, and the coordinate information of each joint point of the robot's working arm is determined according to the kinematic parameters, thereby establishing a coordinate mapping relationship between each joint of the working arm and the robot.

[0131] In this embodiment, the working arm of the robot includes a support arm and a mechanical claw.

[0132] In this embodiment, the target position information of the target book is determined to determine the second target point, and the first operation sequence of the operating arm is generated according to the second target point and the coordinate mapping relationship, and the first operation sequence is corrected according to the real-time obstacle information to generate the second operation sequence, and the end point of the support arm is controlled to move to the second target point according to the second operation sequence, and the mechanical claw is extended and retracted to grab the book.

[0133] In this embodiment, after grabbing the book, the robot retracts the arm and plans a return route according to the real-time location information and the location information of the target user, and delivers the target book to the target user according to the return route.

[0134] Please refer to Figure 2 , Figure 2 A schematic diagram of the structure of a robot-based book search system provided by an embodiment of the present invention, comprising: a semantic analysis module 201, a recognition module 202, a weight distribution module 203, a reordering module 204, a target confirmation module 205 and a job control module 206;

[0135] The semantic analysis module is used to obtain the search information input by the target user, perform semantic analysis on the search information according to a preset semantic analysis model, and obtain a number of target keywords; wherein the semantic analysis model is used to identify the semantic characteristics of each word in the search information;

[0136] The recognition module is used to extract features from historical data in the library based on a recurrent neural network and a convolutional neural network to obtain a book recognition model; the target keyword is input into the book recognition model to generate a first candidate book list;

[0137] The weight allocation module is used to obtain the historical search records of the target user, and determine the first book list in the first candidate book list according to the historical search records; at the same time, obtain the user portrait of the target user, and determine the weight value of each book in the first book list according to the user portrait;

[0138] The reordering module is used to use the weight value as the radius of the circle corresponding to the book, determine the number of intersections generated by the circles corresponding to the books in the first book list, collect the books whose number of intersections reaches a preset number, and generate a second candidate book list; the historical search record includes historical search information, a historical candidate book list and a historical target book;

[0139] The target confirmation module is used to determine the target book in response to the target user's selection action on the second candidate book list, and obtain the location information of the target book;

[0140] The operation control module is used to control the robot to reach the target position and complete book grabbing according to the position information and library obstacle information.

[0141] In this embodiment, the present invention performs semantic analysis on the search information and extracts the keywords in the search, thereby determining the theme of the book according to the keywords based on the recurrent neural network and the convolutional neural network, quickly generating a list of candidate books, and calculating the weights corresponding to each book in the book list as the total radius through the historical search results, thereby screening and sorting the book list, making the book search more accurate, so that users can quickly find the target book and improve the accuracy of book search; in addition, the robot can be controlled to move quickly to the target position through the location information of the target book, so as to accurately grab the target book, avoid users or administrators with large collections from being unable to quickly obtain the target book, facilitate the administrator to manage the books in the library, reduce the manpower and material resources for book search, and improve the efficiency of book search.

[0142] The specific embodiments described above further illustrate the purpose, technical solutions and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. It is particularly pointed out that for those skilled in the art, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the scope of protection of the present invention.

Claims

1. A robot-based book search method, characterized in that: include: Acquire search information input by a target user, perform semantic analysis on the search information according to a preset semantic analysis model, and acquire a plurality of target keywords; wherein the semantic analysis model is used to identify the semantic characteristics of each word in the search information; Extracting features from historical data in the library based on a recurrent neural network and a convolutional neural network to obtain a book recognition model; inputting the target keyword into the book recognition model to generate a first candidate book list; Acquire the historical search records of the target user, and determine the first book list in the first candidate book list according to the historical search records; at the same time, acquire the user portrait of the target user, and determine the weight value of each book in the first book list according to the user portrait; The weight values ​​are used as the radii of the circles corresponding to the books, respectively, to determine the number of intersections generated by the circles corresponding to the books in the first book list, and the books whose intersections reach a preset number are grouped together to generate a second candidate book list; the historical search record includes historical search information, a historical candidate book list and a historical target book; In response to the target user's selection action on the second candidate book list, determining a target book and acquiring location information of the target book; According to the position information and the library obstacle information, the robot is controlled to reach the target position to complete the book grabbing; wherein, the position information includes the operation point position information and the target position information; the control of the robot to reach the target position to complete the book grabbing according to the position information and the library obstacle information is specifically as follows: the robot movement path is planned according to the operation point position information and the library obstacle information, and the robot is controlled to reach the first target point according to the robot movement path; the second target point is determined according to the target position information, and a first operation sequence of the robot working arm is generated according to the second target point and the robot position information; and the target point sequence is corrected according to the obstacle information to generate a second operation sequence, and the robot's working arm is controlled to move to the second target point and the working arm is controlled to grab the target book according to the second operation sequence; The planning of the robot moving path according to the work point location information and the obstacle information of the library specifically includes: obtaining the obstacle information of the library and the first real-time location information of the robot, wherein the obstacle information includes the location data of all obstacle information in the library; determining the first target point according to the work point location information, and planning the robot moving path according to the location data of all obstacle information in the library and the first real-time location information of the robot; when the obstacle is a book, setting the safety distance between the robot and the location data as a first distance threshold; when the obstacle is not a book, setting the safety distance between the robot and the location data as a second distance threshold; and controlling the robot to reach the first target point according to the robot moving path; The method of determining a second target point according to the target position information, and generating a first operation sequence of the robot working arm according to the second target point and the robot position information, specifically comprises: acquiring the second real-time position information of the robot and the real-time position information of each joint of the working arm, and generating a coordinate mapping relationship between the joints of the working arm and the robot according to the second real-time position information and the real-time position information of the working arm; determining the second target point according to the target position information, and generating the first operation sequence of the working arm according to the second target point and the coordinate mapping relationship.

2. The robot-based book search method according to claim 1, characterized in that: The step of obtaining the search information input by the target user and performing semantic analysis on the search information according to a preset semantic analysis model to obtain a number of target keywords is as follows: Building a dictionary base based on books in the library, wherein the dictionary base includes nouns and participles; Acquire search information input by a target user, input the search information into the semantic analysis model, identify segmentations and symbols in the search information, and generate a number of initial keywords; The initial keywords are corrected according to the dictionary library to generate a plurality of target keywords.

3. The robot-based book search method according to claim 1, characterized in that: The feature extraction of historical data in the library based on recurrent neural network and convolutional neural network is used to obtain a book recognition model, which is specifically: Collecting sample data, and dividing the sample data into a training set and a validation set according to a preset ratio; the sample data includes search records of all users stored in the database; Building a book recognition model based on a long short-term memory network and a convolutional neural network, training the book recognition model according to the training set and verification, and obtaining a trained book recognition model; According to singular value decomposition, all linear layers in the trained book recognition model are subjected to low-rank processing, and the book recognition model after low-rank processing is output.

4. The robot-based book search method according to claim 3, characterized in that: Before inputting the target keyword into the book recognition model, the method further includes: optimizing the book recognition model based on the historical search records of the target user to obtain an optimized book recognition model, specifically: Acquire a book recognition model, and adjust characteristic parameters of the book recognition model according to the number of target keywords, wherein the characteristic parameters include the number of convolutional layers, convolutional layer parameters, and long short-term memory network parameters; Calculate the accuracy between the historical search results and the corresponding historical target books according to the historical search records; and optimize and update the book recognition model according to the accuracy to obtain an optimized book recognition model; The target keyword is input into the optimized book recognition model to generate a first candidate book list.

5. The robot-based book search method according to claim 4, characterized in that: The step of inputting the target keyword into the optimized book recognition model to generate a first candidate book list is specifically as follows: Input the target keyword into a long short-term memory network to extract a first feature vector set; and input the first feature vector set into a convolutional neural network to generate a candidate feature vector set; The candidate feature vector set is normalized according to a normalization algorithm, the prediction probability of each candidate feature vector in the candidate feature vector set is calculated, and a first candidate book list is generated.

6. The robot-based book search method according to claim 5, characterized in that: After determining the target book in response to the target user's selection action on the second candidate book list, the method further includes: The search information, the first candidate book list and the historical target books of this search are packaged into sample data according to a preset format and stored in the database; Calculating the accuracy of the book recognition model according to the historical search records and the accuracy calculation algorithm; the historical search results include the number of historical searches, the historical candidate book list and the prediction probability corresponding to the historical candidate book list; updating the book recognition model according to a preset accuracy threshold and the accuracy of the book recognition model; The accuracy calculation algorithm is specifically: Among them, n is the number of historical search predictions, p (i,max) is the probability value with the maximum predicted probability during the i-th search, p (i,x) is the probability value corresponding to the target book during the i-th search, and x is the ranking position of the target book in the first candidate book list during the i-th search.

7. A robot-based book search system, characterized in that: include: Semantic analysis module, recognition module, weight allocation module, reordering module, target confirmation module and job control module; The semantic analysis module is used to obtain the search information input by the target user, perform semantic analysis on the search information according to a preset semantic analysis model, and obtain a number of target keywords; wherein the semantic analysis model is used to identify the semantic characteristics of each word in the search information; The recognition module is used to extract features from historical data in the library based on a recurrent neural network and a convolutional neural network to obtain a book recognition model; the target keyword is input into the book recognition model to generate a first candidate book list; The weight allocation module is used to obtain the historical search records of the target user, and determine the first book list in the first candidate book list according to the historical search records; at the same time, obtain the user portrait of the target user, and determine the weight value of each book in the first book list according to the user portrait; The reordering module is used to use the weight value as the radius of the circle corresponding to the book, determine the number of intersections generated by the circles corresponding to the books in the first book list, collect the books whose number of intersections reaches a preset number, and generate a second candidate book list; the historical search record includes historical search information, a historical candidate book list and a historical target book; The target confirmation module is used to determine the target book in response to the target user's selection action on the second candidate book list, and obtain the location information of the target book; The operation control module is used to control the robot to reach the target position to complete the book grabbing according to the position information and the library obstacle information; wherein the position information includes the operation point position information and the target position information; the control of the robot to reach the target position to complete the book grabbing according to the position information and the library obstacle information is specifically: planning the robot movement path according to the operation point position information and the library obstacle information, and controlling the robot to reach the first target point according to the robot movement path; determining the second target point according to the target position information, and generating a first operation sequence of the robot working arm according to the second target point and the robot position information; and correcting the target point sequence according to the obstacle information to generate a second operation sequence, and controlling the robot's working arm to move to the second target point and controlling the working arm to grab the target book according to the second operation sequence; The planning of the robot moving path according to the work point location information and the obstacle information of the library specifically includes: obtaining the obstacle information of the library and the first real-time location information of the robot, wherein the obstacle information includes the location data of all obstacle information in the library; determining the first target point according to the work point location information, and planning the robot moving path according to the location data of all obstacle information in the library and the first real-time location information of the robot; when the obstacle is a book, setting the safety distance between the robot and the location data as a first distance threshold; when the obstacle is not a book, setting the safety distance between the robot and the location data as a second distance threshold; and controlling the robot to reach the first target point according to the robot moving path; The method of determining a second target point according to the target position information, and generating a first operation sequence of the robot working arm according to the second target point and the robot position information, specifically comprises: acquiring the second real-time position information of the robot and the real-time position information of each joint of the working arm, and generating a coordinate mapping relationship between the joints of the working arm and the robot according to the second real-time position information and the real-time position information of the working arm; determining the second target point according to the target position information, and generating the first operation sequence of the working arm according to the second target point and the coordinate mapping relationship.

Images