Object classification method and device, equipment, storage medium and program product

Abstract

The invention relates to an object classification method and device, equipment, a storage medium and a program product. The method comprises the following steps: acquiring a tactile signal of a to-be-classified object; the tactile signals are acquired by collecting the to-be-classified objects through different sensors; and performing classification processing according to the spatial relationship parameters among the sensors, the tactile signals and a neural network, and determining the category of the to-be-classified object. By adopting the method, the accuracy of the category obtained when the object is classified can be improved.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to an object classification method, apparatus, device, storage medium and program product. Background Art

[0002] Humans can generally judge the category of an object based on information such as temperature, weight, texture, and hardness through touch. Robots generally use sensing technology and algorithms to simulate and obtain tactile information such as temperature, weight, texture, and hardness of an object (for example, using different types of sensors to collect different types of data), thereby realizing the judgment and processing of the object category.

[0003] In related technologies, robots generally use different types of sensors to collect data from objects to obtain different types of data, and then use deep learning models (such as Transformer models, etc.) to classify and process different types of data to obtain the category of the object.

[0004] However, the above technology has the problem that the object categories obtained are not accurate enough. Summary of the invention

[0005] Embodiments of the present application provide an object classification method, apparatus, device, storage medium, and program product, which can improve the accuracy of the categories obtained when classifying objects.

[0006] In a first aspect, an embodiment of the present application provides an object classification method, the method comprising:

[0007] Acquire tactile signals of the objects to be classified; the tactile signals are acquired by collecting the objects to be classified through different sensors;

[0008] Classification processing is performed based on the spatial relationship parameters between the sensors, the tactile signals and the neural network to determine the category of the object to be classified.

[0009] In one embodiment, the neural network includes a graph attention network, and the classification process based on the spatial relationship parameters between the sensors, the tactile signals and the neural network to determine the category of the object to be classified includes:

[0010] The spatial relationship parameters between each sensor and each tactile signal are used as the input of the graph attention network, and classification processing is performed in the graph attention network to determine the category of the object to be classified.

[0011] In one embodiment, the spatial relationship parameter includes a first graph structure between the sensors.

[0012] In one of the embodiments, the first graph structure represents the spatial connection relationship between the sensors.

[0013] In one of the embodiments, the first graph structure is an explicit feature graph structure.

[0014] In one embodiment, the spatial relationship parameters between the sensors and the tactile signals are used as inputs of the graph attention network, and classification processing is performed in the graph attention network to determine the category of the object to be classified, including:

[0015] Determine, according to each tactile signal, a second graph structure corresponding to each tactile signal;

[0016] The first graph structure and the second graph structure are input into the graph attention network, and classification processing is performed in the graph attention network to determine the category of the object to be classified.

[0017] In one embodiment, the step of determining the second graph structure corresponding to each tactile signal according to each tactile signal includes:

[0018] Correlation analysis is performed on each tactile signal to determine a second graph structure corresponding to each tactile signal.

[0019] In one embodiment, the neural network further includes an embedding layer, and the correlation analysis process of each tactile signal to determine the second graph structure corresponding to each tactile signal includes:

[0020] Input each tactile signal into the embedding layer for correlation analysis and processing to determine a correlation matrix corresponding to each tactile signal; the correlation matrix includes the connection relationship and correlation degree between each tactile signal and other tactile signals;

[0021] determining each target tactile signal from each tactile signal according to a correlation matrix;

[0022] A second graph structure is constructed according to the connection relationship and correlation degree between the target tactile signals.

[0023] In one embodiment, the step of determining each target tactile signal from each tactile signal according to the correlation matrix includes:

[0024] According to the correlation matrix, each target tactile signal that meets a correlation threshold is selected from the tactile signals.

[0025] In one of the embodiments, the second graph structure is an implicit feature graph structure.

[0026] In one embodiment, the step of obtaining a tactile signal of an object to be classified includes:

[0027] Acquire tactile signals of the object to be classified; the tactile signals are acquired by collecting the object to be classified through different sensors.

[0028] In one embodiment, the training method of the neural network includes:

[0029] Acquire multiple training tactile signal sets; each training tactile signal set includes multiple training tactile signals of sample objects and labeled categories of the sample objects;

[0030] Obtaining spatial relationship parameters corresponding to each training tactile signal set;

[0031] Each training tactile signal set and the corresponding spatial relationship parameter are used as the input of the initial neural network, and the labeled category of each training tactile signal set is used as the reference output of the initial neural network. The initial neural network is trained to obtain a neural network.

[0032] In a second aspect, an embodiment of the present application provides an object classification device, the device comprising:

[0033] A signal acquisition module is used to acquire tactile signals of the object to be classified; the tactile signals are acquired by collecting the object to be classified through different sensors;

[0034] The classification module is used to perform classification processing based on the spatial relationship parameters between the sensors, the tactile signals and the neural network to determine the category of the object to be classified.

[0035] In a third aspect, an embodiment of the present application provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0036] Acquire tactile signals of the objects to be classified; the tactile signals are acquired by collecting the objects to be classified through different sensors;

[0037] Classification processing is performed based on the spatial relationship parameters between the sensors, the tactile signals and the neural network to determine the category of the object to be classified.

[0038] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the following steps are implemented:

[0039] Acquire tactile signals of the objects to be classified; the tactile signals are acquired by collecting the objects to be classified through different sensors;

[0040] Classification processing is performed based on the spatial relationship parameters between the sensors, the tactile signals and the neural network to determine the category of the object to be classified.

[0041] In a fifth aspect, an embodiment of the present application provides a chip, the chip including a programmable logic circuit and / or a program instruction, and when the chip is running, the object classification method provided by the embodiment of the present application can be executed, and the method can be:

[0042] Acquire tactile signals of the objects to be classified; the tactile signals are acquired by collecting the objects to be classified through different sensors;

[0043] Classification processing is performed based on the spatial relationship parameters between the sensors, the tactile signals and the neural network to determine the category of the object to be classified.

[0044] In a sixth aspect, an embodiment of the present application provides a computer program product, including a computer program, characterized in that when the computer program is executed by a processor, the object classification method provided by the embodiment of the present application is implemented, and the method can be:

[0045] Acquire tactile signals of the objects to be classified; the tactile signals are acquired by collecting the objects to be classified through different sensors;

[0046] Classification processing is performed based on the spatial relationship parameters between the sensors, the tactile signals and the neural network to determine the category of the object to be classified.

[0047] The object classification method, device, equipment, storage medium and program product described above obtain tactile signals of multiple channels acquired by using different sensors to collect objects to be classified, and classify and process the objects according to the spatial relationship parameters between the sensors, the tactile signals and the neural network to determine the category of the objects to be classified. In this method, since the spatial relationship parameters between the sensors that collect the tactile signals are added to the object classification process, the neural network can extract more and more informative information from the tactile signals for object classification, thereby improving the accuracy of object classification. At the same time, through the spatial position parameters of the tactile signals and the sensors that collect the tactile signals, even in the case of limited tactile signals, a certain degree of information can still be obtained from the connection between the spatial position parameters of the sensors, thereby helping to alleviate data sparsity and small sample problems, while providing more reliable object classification performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 is an internal structure diagram of a computer device in one embodiment;

[0049] Figure 2 is a flow chart of an object classification method in one embodiment;

[0050] Figure 3 is an example diagram of the layout of sensors in another embodiment;

[0051] Figure 4 is a flow chart of an object classification method in another embodiment;

[0052] Figure 5 is a flow chart of an object classification method in another embodiment;

[0053] Figure 6 is a flow chart of an object classification method in another embodiment;

[0054] Figure 7 A structural example diagram of an object classification process in another embodiment;

[0055] Figure 8 FIG. 4 is a structural block diagram of an object classification device in an embodiment. DETAILED DESCRIPTION

[0056] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0057] The object classification method provided in the embodiment of the present application can be applied to a computer device, which can be a terminal or a server. Taking the terminal as an example, its internal structure diagram can be as follows: Figure 1 As shown. The computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, an object classification method is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a key, trackball or touchpad set on the computer device housing, or an external keyboard, touchpad or mouse, etc.

[0058] Those skilled in the art will understand that Figure 1 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0059] Humans can generally judge the category of an object based on information such as temperature, weight, texture, and hardness through touch. Although robots cannot directly judge the type or category of an object through touch like humans, by using advanced perception technology and algorithms, such as temperature sensors, pressure sensors, and tactile sensors, robots can simulate and obtain tactile information such as the temperature, weight, texture, and hardness of an object, thereby judging and processing the category of the object. This provides robots with broader possibilities and application prospects in various application fields, such as mechanical processing, automated operations, and medical care.

[0060] In traditional technology, robots generally use different types of sensors to collect data from objects to obtain different types of data, and then use deep learning models (such as Transformer models, LSTM long short-term memory networks, etc.) to classify and process different types of data to obtain the category of the object. However, this technology cannot help robots better process and understand multiple types of tactile information from multiple channels like the human brain, and classify objects based on the tactile information. In other words, there is a problem in the traditional technology that the object categories obtained are not accurate enough.

[0061] Based on the above-mentioned traditional technology, an embodiment of the present application provides an object classification method, which classifies the categories of objects through multi-channel tactile signals and combines the spatial relationship parameters between sensors that collect the multi-channel tactile signals, so as to improve the accuracy of object classification in this way.

[0062] It should be noted that the beneficial effects brought about by the embodiments of the present application or the technical problems solved are not limited to this one, but may also include other implicit or related problems. For details, please refer to the description of the following embodiments.

[0063] The technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems are described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0064] In one embodiment, Figure 2 As shown, an object classification method is provided, which is applied to Figure 1 Taking the computer device in the example as an example, the method may include the following steps:

[0065] S202, obtaining tactile signals of the object to be classified; the tactile signals are obtained by collecting the object to be classified through different sensors.

[0066] The object to be classified may be any object that needs to be classified, such as fruits, vegetables, animals, plants, medicines, equipment, etc.

[0067] When a robot is used to classify objects to be classified that are touched or grasped, multiple sensors (also called tactile sensors) can be arranged on the robot's manipulator, so that when the manipulator touches or grasps the objects to be classified, it can touch or grasp the objects to be classified through these multiple sensors to collect tactile signals of the objects to be classified.

[0068] The sensors here may include pressure sensors, temperature sensors, humidity sensors, bending sensors, etc. Among them, the pressure sensor can collect the pressure tactile signal received when touching or grasping the object to be classified, the temperature sensor can collect the temperature tactile signal generated when touching or grasping the object to be classified, the humidity sensor can collect the humidity tactile signal generated when touching or grasping the object to be classified, and the bending sensor can collect the bending tactile signal generated when touching or grasping the object to be classified. Of course, other types of sensors can also be included, and the several types given here are only examples and are not limited to this.

[0069] The multiple sensors here can be multiple sensors of the same type, such as multiple pressure sensors; or multiple sensors of different types, such as multiple pressure sensors and multiple bending sensors. In addition, the arrangement positions of the multiple sensors and the sizes of the sensors can be set according to the actual position and size of the manipulator, for example, multiple pressure sensors of the same type but different sizes can be arranged on a manipulator.

[0070] Specifically, the object to be classified can be touched or grasped by a manipulator, and then the sensors on the manipulator can collect tactile signals when touching or grasping the object to be classified, and the obtained tactile signals can be recorded as tactile signals of the object to be classified. As an optional embodiment, obtaining the tactile signals of the object to be classified in this step may include obtaining the tactile signals of the object to be classified; the tactile signals are obtained by collecting the object to be classified through different sensors. In other words, since multiple sensors are used for signal collection, each sensor can collect the tactile signals of one channel, and therefore tactile signals of multiple channels can be obtained here, that is, tactile signals of multiple channels of the object to be classified are obtained. In addition, the tactile signals obtained here may be analog signals.

[0071] For example, see Figure 3The example diagram shown is of sensors arranged on a manipulator. Multiple sensors may be arranged at various positions such as the fingers and palm of the manipulator. The multiple sensors may include pressure sensors, bending sensors, temperature sensors, humidity sensors, etc. of different sizes, so as to collect tactile signals of different channels of the object to be classified through these different types of sensors.

[0072] S204, performing classification processing according to the spatial relationship parameters between the sensors, the tactile signals and the neural network to determine the category of the object to be classified.

[0073] The neural network may be a network composed of a single network, a network obtained by combining multiple different networks, or a network obtained by combining multiple identical networks. For example, the neural network here may include a graph convolutional neural network GCN, a graph attention network GAT, etc. The neural network may be pre-trained, for example, by training multi-channel sample tactile signals with labeled categories and spatial relationship parameters between sensors that collect sample tactile signals.

[0074] In this step, when multiple sensors are pre-deployed on the manipulator, the spatial relationship parameters between the deployed sensors can be obtained based on the deployment position information, etc. The spatial relationship parameters here mainly represent the spatial relationship parameters of each sensor in the real physical space, such as the spatial position of each sensor and the spatial relative position relationship between each sensor, or the spatial connection relationship between each sensor, etc.

[0075] After obtaining the spatial relationship parameters between the sensors, the tactile signals of multiple channels can also be collected by the sensors. After that, the spatial relationship parameters between the sensors can be combined with the tactile signals of multiple channels of the object to be classified, and the tactile signals of multiple channels can be classified and processed by combining the neural network. In this way, the neural network can better obtain the local structure and global structure and other features between the tactile signals through the spatial relationship parameters between the sensors during the classification process. More features are obtained and the spatial relationship features are combined. Therefore, more accurate classification of the object to be classified can be achieved, and a more accurate category of the object to be classified can be obtained.

[0076] The category of the object to be classified obtained here actually refers to what kind of object the object to be classified is, such as an apple, a cat, a dog, etc.

[0077] In the above object classification method, tactile signals of multiple channels are obtained by using different sensors to collect objects to be classified, and the categories of objects to be classified are determined based on the spatial relationship parameters between the sensors, the tactile signals and the neural network. In this method, since the spatial relationship parameters between the sensors that collect the tactile signals are added to the object classification process, the neural network can extract more and more informative information from the tactile signals for object classification, thereby improving the accuracy of object classification. At the same time, through the spatial position parameters of the tactile signals and the sensors that collect the tactile signals, even in the case of limited tactile signals, a certain degree of information can still be obtained from the connection between the spatial position parameters of the sensors, which helps to alleviate data sparsity and small sample problems, while providing more reliable object classification performance.

[0078] The above embodiments mentioned that the neural network can classify multi-channel tactile signals in combination with the spatial relationship parameters between the sensors. The following describes the specific classification process using the neural network when the neural network includes a graph attention network.

[0079] In another embodiment, another object classification method is provided. Based on the above embodiment, the above S204 may include the following steps:

[0080] In step A, the spatial relationship parameters between the sensors and the tactile signals are used as inputs of the graph attention network, and classification processing is performed in the graph attention network to determine the category of the object to be classified.

[0081] In this step, the neural network includes a graph attention network, which can be denoted as GAT. After obtaining the spatial relationship parameters between the sensors and the tactile signals, the spatial relationship parameters and the tactile signals can be directly input into the graph attention network for classification processing, or the tactile signals can be preprocessed and then input into the graph attention network together with the spatial relationship parameters for classification processing, or the spatial relationship parameters and the tactile signals can be preprocessed and then input into the graph attention network together for classification processing, or the spatial relationship parameters and the tactile signals can be preprocessed and then input into the graph attention network together for classification processing, or other methods can be used to input the spatial relationship parameters between the sensors and the tactile signals as inputs to the graph attention network. In short, the two parts of information, namely the spatial relationship parameters and the tactile signals, can be input into the graph attention network. The preprocessing can be, for example, the process of generating a graph structure.

[0082] In the process of classification by the graph attention network, the graph attention network can learn the weights between different tactile signal channels in advance, so that different weights can be assigned to each tactile signal based on the importance or relevance of each tactile signal during the application process, so that the network pays more attention to more important or more relevant tactile signals, and obtains a more accurate category of the object to be classified. In addition, the above-mentioned graph attention network can also perform feature extraction, feature fusion, convolution, pooling, normalization and other processing on spatial relationship parameters and each tactile signal, and finally obtain the category of the object to be classified.

[0083] That is to say, the above-mentioned graph attention network can automatically learn the weights between each node (such as tactile signal) and its neighbors during the classification process, so that each node can adaptively aggregate the information of neighboring nodes according to its importance. This adaptive feature aggregation method can better adapt to the relationship and importance between different nodes, thereby improving the accuracy of subsequent classification. At the same time, the graph attention network combines the spatial relationship parameters between each sensor and the information between each tactile signal for adaptive aggregation, which is very useful for identifying complex tactile patterns or understanding dynamic changes in tactile sequences, and can help robots better perceive and adapt to complex environmental information, and improve their autonomous decision-making and execution capabilities. At the same time, even in the case of limited data, the classification method in this embodiment can still obtain a certain degree of information from the connection between nodes, which also helps to alleviate data sparsity and small sample problems, and provide more reliable classification performance.

[0084] In this embodiment, the neural network includes a graph attention network, and the spatial relationship parameters between each sensor and each tactile signal can be used as input of the graph attention network to perform classification processing and obtain the category of the object to be classified. Here, the spatial relationship parameters of each sensor and each tactile signal are classified and processed by the graph attention network, which can further improve the classification performance and obtain a more accurate classification result.

[0085] The above embodiments mentioned that the spatial relationship parameters can be used as the input of the graph attention network. Usually, the graph attention network is a new type of neural network based on the graph structure. So what is the relationship between the spatial relationship parameters and the graph structure? The following embodiments illustrate this.

[0086] In another embodiment, based on the above embodiment, the above spatial relationship parameter includes a first graph structure between the sensors.

[0087] In this embodiment, the spatial relationship parameters of each sensor may include a graph structure between the sensors, which is recorded as a first graph structure. The first graph structure can be, for example, Figure 3The example diagram of the deployment of various sensors on the robot shown in the figure, the first graph structure may include multiple nodes and the connection relationship between the nodes, each node may represent a sensor, and the node information of each node may include the location of the corresponding sensor.

[0088] At the same time, as an optional embodiment, the above-mentioned first graph structure represents the spatial connection relationship between each sensor, that is, the first graph structure can also include the spatial connection relationship between each sensor. The spatial connection relationship here indicates that the connected sensors are associated, which can characterize the physical adjacency relationship between each sensor / node.

[0089] In addition, as an optional embodiment, the first graph structure is an explicit feature graph structure. During classification processing, the first graph structure can be input into the graph attention network, so that when the graph attention network classifies each tactile signal, it can process the importance and correlation between each tactile signal based on the physical adjacency relationship between each sensor in the first graph structure, that is, the explicit feature graph structure, so as to better obtain the local structural features and global structural features between each tactile signal, obtain more and richer features, and be more conducive to obtaining accurate classification results.

[0090] In this embodiment, the spatial relationship parameters between the sensors may include a graph structure that characterizes the spatial connection relationship between the sensors. The graph structure is input into the graph attention network to assist in tactile signal classification. The graph attention network can better obtain the local structural features and global structural features between the tactile signals, and the features obtained are more numerous and richer, which is more conducive to obtaining accurate classification results.

[0091] The following embodiment describes a specific process of classifying the above-mentioned spatial relationship parameters of each sensor and each tactile signal based on the first graph structure of each sensor.

[0092] In another embodiment, another object classification method is provided. Based on the above embodiment, Figure 4 As shown, the above step A may include the following steps:

[0093] S302: Determine, according to each tactile signal, a second graph structure corresponding to each tactile signal.

[0094] In this step, after obtaining each tactile signal, a graph structure between each tactile signal can be constructed based on a neural network; of course, the graph structure between each tactile signal can also be obtained through other analysis methods, manual methods, etc.; the graph structure between each tactile signal constructed here can be recorded as a second graph structure.

[0095] Here, the second graph structure may also include multiple nodes and the connection relationship between each node. Each node may represent a tactile signal. The node information of each node may include information such as the corresponding tactile signal, the sensor corresponding to the tactile signal, and the weight of the tactile signal. Here, the connection relationship between each node represents the connection relationship between each tactile signal. Usually, connected tactile signals are adjacent signals, and the correlation between them is also relatively strong. Based on this graph structure, more accurate tactile signals and the correlation relationship between tactile signals can be obtained.

[0096] S304: Input the first graph structure and the second graph structure into a graph attention network, perform classification processing in the graph attention network, and determine the category of the object to be classified.

[0097] In this step, after obtaining the first graph structure between each sensor and the second graph structure corresponding to each tactile signal, the two graph structures can be input into the graph attention network. After performing feature extraction, feature fusion, convolution, pooling, normalization and other processing on the two graph structures in the graph attention network, the category corresponding to the object to be classified is obtained.

[0098] Of course, after obtaining the two graph structures, the two graph structures can also be pre-processed and then input into the graph attention network for classification processing to obtain the category corresponding to the object to be classified. The pre-processing here can be, for example, graph structure fusion, redundant graph structure cutting and other processing.

[0099] In addition, as an optional embodiment, the second graph structure is an implicit feature graph structure. That is to say, the spatial connection relationship between the tactile signals corresponding to the second graph structure can be used as an implicit feature of the graph attention network, and the first graph structure is used as an explicit feature. The combination of the two features can provide more features, which facilitates the graph attention network to better achieve classification.

[0100] In this embodiment, the second graph structure between each tactile signal is determined by each tactile signal, and the first graph structure between each sensor and the second graph structure are input into the graph attention network for classification to obtain the category of the object to be classified, so that the graph structure of each tactile signal is determined for subsequent classification, thereby effectively improving the performance of subsequent classification using tactile signals. At the same time, the two graph structures are input into the graph attention network for classification, which can avoid formatting the input data, thereby improving the classification efficiency.

[0101] The following embodiment describes the process of obtaining the second graph structure by the above-mentioned analysis method. Based on the above embodiment, the above S302 may include the following steps:

[0102] Step B: performing correlation analysis on each tactile signal to determine a second graph structure corresponding to each tactile signal.

[0103] In this step, after obtaining each tactile signal, the correlation between the tactile signals can be analyzed, and the correlation here can include the correlation between the tactile signals, whether the spatial positions are connected, similarity, etc. By performing correlation analysis on each tactile signal, the second graph structure between the tactile signals can be determined through the correlation analysis results. In addition, when performing the correlation analysis here, it can be performed through a neural network or through other methods.

[0104] In this embodiment, the second graph structure corresponding to each tactile signal is determined by performing correlation analysis on each tactile signal, which can facilitate accurate and rapid construction of the second graph structure for subsequent classification, thereby improving classification efficiency and accuracy.

[0105] The above embodiments mentioned that a neural network can be composed of multiple different types of networks, and it is also mentioned that a neural network can include a graph attention network. The following embodiments describe how to obtain a more accurate tactile signal and its graph structure through an embedding layer when the neural network also includes an embedding layer.

[0106] In another embodiment, another object classification method is provided. Based on the above embodiment, Figure 5 As shown, the above step B may include the following steps:

[0107] S402, inputting each tactile signal into the embedding layer for correlation analysis and processing, and determining a correlation matrix corresponding to each tactile signal; the correlation matrix includes a connection relationship and a correlation degree between each tactile signal and other tactile signals.

[0108] The embedding layer can be recorded as an embedding layer, which is a word encoding technology specifically used in the field of NLP (natural language processing). The embedding layer in this embodiment is mainly used to learn and establish the importance and correlation between various tactile signals, and at the same time construct a graph structure between various tactile signals. The embedding layer can also be pre-trained, for example, it can be trained as part of a neural network.

[0109] In this step, after the neural network is trained, the embedding layer therein is also trained accordingly. Then, after obtaining each tactile signal as mentioned above, each tactile signal can be input into the embedding layer, and a corresponding weight is assigned to each tactile signal in the embedding layer. Specifically, the correlation / relevance between each tactile signal can be analyzed through the embedding layer to obtain a correlation matrix of each tactile signal.

[0110] The correlation matrix here includes multiple elements, wherein each row or each column may include multiple elements. For example, each row of the correlation matrix may include multiple elements, each element respectively represents the connection relationship and correlation degree between a tactile signal and itself and other tactile signals. The connection relationship here may represent whether there is a connection relationship between the tactile signal and other tactile signals or the closeness of the connection relationship.

[0111] Taking the tactile signals of three channels as an example, the correlation matrix obtained here can be a matrix of 3 rows and 3 columns, which can include 9 elements. For example, the element of the 1st row and the 2nd column can represent the connection relationship and correlation degree between the first tactile signal and the second tactile signal.

[0112] S404: Determine each target tactile signal from each tactile signal according to the correlation matrix.

[0113] In this step, after obtaining the correlation matrix between the tactile signals, the more important and accurate tactile signals can be selected accordingly. In the specific selection, as an optional embodiment, the target tactile signals that meet the correlation threshold can be selected from the tactile signals according to the correlation matrix.

[0114] The correlation threshold here may be a preset threshold, such as 0.8, and satisfying the correlation threshold here may be that the correlation degree is greater than or equal to the correlation threshold. After obtaining the correlation matrix, the correlation degree between each tactile signal and other tactile signals may be compared with the correlation threshold, and multiple tactile signals with strong correlation are selected and recorded as target tactile signals.

[0115] S406: construct a second graph structure according to the connection relationship and correlation degree between the target tactile signals.

[0116] In this step, after obtaining the target tactile signals with strong correlation, the graph structure between the target tactile signals can be constructed through the connection relationship between the target tactile signals in the correlation matrix, and combined with the degree of correlation between the target tactile signals, which is recorded as the second graph structure. As mentioned above, each node in the second graph structure can represent a tactile signal of a channel, and the node information of each node can include information such as the tactile signal corresponding to the node, the degree of correlation between the node and other nodes, etc.

[0117] In this embodiment, each tactile signal is subjected to correlation analysis through an embedding layer, and each target tactile signal with strong correlation is selected from the obtained correlation matrix to construct a graph structure, so that the correlation between each target tactile signal in the constructed graph structure is strong, so that the graph attention network can be better assisted in classifying each target tactile signal to obtain a better classification result. In addition, the process of selecting each target tactile signal that meets the correlation threshold from each tactile signal through the correlation matrix is ​​simple and intuitive, so it can effectively improve the efficiency of constructing the graph structure of each target tactile signal, thereby improving the classification efficiency.

[0118] The above embodiments mentioned that the neural network can be trained. The following embodiments illustrate the training process of the neural network.

[0119] In another embodiment, another object classification method is provided, such as Figure 6 As shown, based on the above embodiment, the training method of the above neural network may include the following steps:

[0120] S502, obtaining a plurality of training tactile signal sets; each training tactile signal set includes a plurality of training tactile signals of sample objects and labeled categories of the sample objects.

[0121] In this step, multiple sensors can be pre-arranged on the manipulator, and then each sensor can be used to touch or grasp different sample objects to obtain tactile signals collected by each sensor, which are recorded as training tactile signals. Multiple training tactile signals collected by each sensor for the same sample object can be combined together to form a training tactile signal set corresponding to the sample. In other words, each sample object can obtain one or more training tactile signal sets, and the category of each sample object is also known in advance, so each training tactile signal set can be labeled with the category of its corresponding sample object, which is recorded as the labeled category. In addition, the sample objects here can be the same object, or they can be different objects.

[0122] Of course, in order to expand the training tactile signal set, each training tactile signal set may be amplified to obtain an amplified training tactile signal set, and the amplification process may include scaling, slicing, noise addition, etc. Afterwards, a sliding window may be used to select a desired training tactile signal set from the amplified training tactile signal set.

[0123] S504: Obtain spatial relationship parameters corresponding to each training tactile signal set.

[0124] In this step, when a plurality of sensors are arranged on the manipulator, the spatial relationship parameters between the sensors can be obtained, and the spatial relationship parameters are the spatial relationship parameters corresponding to the collected training tactile signal set.

[0125] If multiple sensors arranged in the same manner are used to collect signals from each sample object to obtain each training tactile signal set, the spatial relationship parameters corresponding to each training tactile signal set can be the same. Of course, in actual situations, different sensors can be set according to different sample objects, and then different training tactile signal sets and different spatial relationship parameters can be obtained.

[0126] In addition, as mentioned above, the spatial relationship parameters here may include a graph structure that characterizes the spatial connection relationship between the sensors, and each node in the graph structure may include information such as the spatial position of the corresponding sensor and the spatial adjacency relationship with other sensors.

[0127] S506, taking each training tactile signal set and the corresponding spatial relationship parameter as the input of the initial neural network, taking the labeled category of each training tactile signal set as the reference output of the initial neural network, training the initial neural network, and obtaining a neural network.

[0128] In this step, the initial neural network is an untrained neural network, which may include an untrained initial embedding layer and an untrained initial graph attention network.

[0129] After obtaining each training tactile signal set, the training tactile signal set can be input into the initial embedding layer to obtain the graph structure corresponding to each training tactile signal set, and then the graph structure corresponding to each sensor and the graph structure of each training tactile signal set obtained here are input into the initial graph attention network to classify each training tactile signal set and obtain the predicted category corresponding to each training tactile signal set. After that, the loss between the predicted category of each training tactile signal set and the corresponding labeled category can be calculated, and the calculated loss can be back-propagated to adjust the parameters of the initial embedding layer and the initial graph attention network to achieve the training of the initial embedding layer and the initial graph attention network, and iteratively execute this training process until the calculated loss is unchanged / stable / less than the threshold, then it is determined that the initial embedding layer and the initial graph attention network are trained well, that is, the initial neural network is trained well, and a trained neural network is obtained, that is, a trained embedding layer and graph attention network are obtained.

[0130] In this embodiment, the initial neural network is trained by using each training tactile signal set and its corresponding spatial relationship parameters, labeling categories, etc. This can make the trained neural network more accurate, thereby improving the accuracy of subsequent classification through the neural network.

[0131] A detailed embodiment is given below to further illustrate the technical solution of the present application. Figure 7The classification structure schematic diagram shown in FIG. 1 shows, based on the above embodiment, the method may include the following steps:

[0132] S1, obtaining tactile signals of the object to be classified; the tactile signals are obtained by collecting the object to be classified through different sensors;

[0133] S2, constructing a first graph structure between the sensors according to the physical adjacency relationship between the sensors;

[0134] S3, inputting each tactile signal into an embedding layer in a neural network for correlation analysis, and determining a correlation matrix corresponding to each tactile signal; the correlation matrix includes a connection relationship and a correlation degree between each tactile signal and other tactile signals;

[0135] S4, selecting target tactile signals satisfying a correlation threshold from the tactile signals according to the correlation matrix;

[0136] S5, constructing a second graph structure (i.e., the learning channel relationship in the graph, where the learning channel relationship is to learn the correlation relationship between the tactile signals of each channel) according to the connection relationship and correlation degree between the target tactile signals;

[0137] S6, inputting the first graph structure and the second graph structure into a graph attention network in the neural network, performing classification processing in the graph attention network, and determining the category of the object to be classified.

[0138] It should be noted that this embodiment mainly uses the graph structure constructed by the multi-channel tactile signals collected by each sensor and the physical adjacency relationship of each sensor in reality as the input of the graph attention network. The embedding layer is used to learn the implicit connection relationship and weight size between the tactile signals of different channels. With the continuous back-propagation training of the neural network, the parameters of the embedding layer will be continuously adjusted to output the most accurate tactile signal and its graph structure. The physical adjacency relationship (prior information) of the sensors passed into the graph attention network in the neural network will also continuously adjust the connection relationship and connection weight during the neural network training. Finally, the relationship information between the tactile signals of each channel learned by the embedding layer is used to aggregate the tactile features using the graph attention network, and finally the tactile features are passed to the fully connected layer to output the tactile signal category.

[0139] It can be seen that in this embodiment, the explicit graph structure is mainly used to characterize the physical adjacency relationship between each sensor, and the implicit graph structure is used to autonomously learn the implicit correlation relationship between each tactile signal during training, and a graph attention network is introduced to learn the weights between different tactile signal channels, so that the graph attention network can better capture the local structure and global structure between each tactile signal, and effectively utilize the structural information between tactile signals of different channels to obtain more and richer features, improve the accuracy of tactile signal classification, and further help the robot better perceive and adapt to complex environmental information, and improve its autonomous decision-making and execution capabilities.

[0140] It should be understood that, although the various steps in the above flow chart are shown in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear description in this article, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a part of the steps in the above figure may include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0141] In one embodiment, Figure 8 As shown, the embodiment of the present application provides an object classification device, including: a signal acquisition module and a classification module, wherein:

[0142] A signal acquisition module is used to acquire tactile signals of the object to be classified; the tactile signals are acquired by collecting the object to be classified through different sensors;

[0143] The classification module is used to perform classification processing based on the spatial relationship parameters between the sensors, the tactile signals and the neural network to determine the category of the object to be classified.

[0144] In another embodiment, another object classification device is provided. Based on the above embodiment, the neural network includes a graph attention network, and the classification module may include:

[0145] The classification unit is used to take the spatial relationship parameters between the sensors and the tactile signals as the input of the graph attention network, perform classification processing in the graph attention network, and determine the category of the object to be classified.

[0146] Optionally, the above-mentioned spatial relationship parameter includes a first graph structure between the sensors.

[0147] Optionally, the first graph structure represents a spatial connection relationship between sensors.

[0148] Optionally, the first graph structure is an explicit feature graph structure.

[0149] In another embodiment, another object classification device is provided. Based on the above embodiment, the classification unit includes:

[0150] a determination subunit, configured to determine, according to each tactile signal, a second graph structure corresponding to each tactile signal;

[0151] The classification subunit is used to input the first graph structure and the second graph structure into the graph attention network, perform classification processing in the graph attention network, and determine the category of the object to be classified.

[0152] In another embodiment, another object classification device is provided. Based on the above embodiment, the above determination subunit is specifically used to perform correlation analysis on each tactile signal to determine the second graph structure corresponding to each tactile signal.

[0153] In another embodiment, another object classification device is provided. Based on the above embodiment, the neural network also includes an embedding layer. The determination subunit is specifically used to input each tactile signal into the embedding layer for correlation analysis and processing, and determine the correlation matrix corresponding to each tactile signal; the correlation matrix includes the connection relationship and the degree of correlation between each tactile signal and other tactile signals; each target tactile signal is determined from each tactile signal according to the correlation matrix; and a second graph structure is constructed according to the connection relationship and the degree of correlation between each target tactile signal.

[0154] Optionally, the second graph structure is an implicit feature graph structure.

[0155] Optionally, the determination subunit is specifically configured to select, according to the correlation matrix, target tactile signals satisfying a correlation threshold from the tactile signals.

[0156] Optionally, the signal acquisition module may include:

[0157] The signal acquisition unit is used to acquire the tactile signal of the object to be classified; the tactile signal is acquired by collecting the object to be classified through different sensors.

[0158] In another embodiment, another object classification device is provided. Based on the above embodiment, the above device may further include a training module, and the training module includes:

[0159] A training set acquisition unit, used to acquire a plurality of training tactile signal sets; each training tactile signal set includes a plurality of training tactile signals of a sample object and a labeled category of the sample object;

[0160] A parameter acquisition unit, used to acquire the spatial relationship parameters corresponding to each training tactile signal set;

[0161] The training unit is used to use each training tactile signal set and the corresponding spatial relationship parameters as the input of the initial neural network, use the labeled category of each training tactile signal set as the reference output of the initial neural network, train the initial neural network, and obtain the neural network.

[0162] For the specific definition of the object classification device, please refer to the definition of the object classification method above, which will not be repeated here. Each module in the above-mentioned object classification device can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0163] In one embodiment, the present application provides a computer device, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the following steps are implemented:

[0164] Acquire tactile signals of the object to be classified; the tactile signals are acquired by collecting the object to be classified through different sensors; and classify and process the object to be classified according to the spatial relationship parameters between the sensors, the tactile signals and the neural network to determine the category of the object to be classified.

[0165] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0166] The spatial relationship parameters between each sensor and each tactile signal are used as the input of the graph attention network, and classification processing is performed in the graph attention network to determine the category of the object to be classified.

[0167] In one embodiment, the spatial relationship parameter includes a first graph structure between the sensors.

[0168] In one embodiment, the first graph structure represents the spatial connection relationship between the sensors.

[0169] In one embodiment, the first graph structure is an explicit feature graph structure.

[0170] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0171] A second graph structure corresponding to each tactile signal is determined according to each tactile signal; the first graph structure and the second graph structure are input into a graph attention network, and classification processing is performed in the graph attention network to determine the category of the object to be classified.

[0172] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0173] Correlation analysis is performed on each tactile signal to determine a second graph structure corresponding to each tactile signal.

[0174] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0175] Each tactile signal is input into the embedding layer for correlation analysis and processing to determine the correlation matrix corresponding to each tactile signal; the above correlation matrix includes the connection relationship and the degree of correlation between each tactile signal and other tactile signals; each target tactile signal is determined from each tactile signal according to the correlation matrix; and a second graph structure is constructed according to the connection relationship and the degree of correlation between each target tactile signal.

[0176] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0177] According to the correlation matrix, each target tactile signal that meets a correlation threshold is selected from the tactile signals.

[0178] In one embodiment, the second graph structure is an implicit feature graph structure.

[0179] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0180] Acquire tactile signals of the object to be classified; the tactile signals are acquired by collecting the object to be classified through different sensors.

[0181] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0182] Acquire multiple training tactile signal sets; each training tactile signal set includes multiple training tactile signals of sample objects and labeled categories of the sample objects; obtain spatial relationship parameters corresponding to each training tactile signal set; use each training tactile signal set and the corresponding spatial relationship parameters as input of the initial neural network, use the labeled categories of each training tactile signal set as reference output of the initial neural network, train the initial neural network, and obtain a neural network.

[0183] In one embodiment, the present application provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the following steps are implemented:

[0184] Acquire tactile signals of the object to be classified; the tactile signals are acquired by collecting the object to be classified through different sensors; and classify and process the object to be classified according to the spatial relationship parameters between the sensors, the tactile signals and the neural network to determine the category of the object to be classified.

[0185] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0186] The spatial relationship parameters between each sensor and each tactile signal are used as the input of the graph attention network, and classification processing is performed in the graph attention network to determine the category of the object to be classified.

[0187] In one embodiment, the spatial relationship parameter includes a first graph structure between the sensors.

[0188] In one embodiment, the first graph structure represents the spatial connection relationship between the sensors.

[0189] In one embodiment, the first graph structure is an explicit feature graph structure.

[0190] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0191] A second graph structure corresponding to each tactile signal is determined according to each tactile signal; the first graph structure and the second graph structure are input into a graph attention network, and classification processing is performed in the graph attention network to determine the category of the object to be classified.

[0192] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0193] Correlation analysis is performed on each tactile signal to determine a second graph structure corresponding to each tactile signal.

[0194] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0195] Each tactile signal is input into the embedding layer for correlation analysis and processing to determine the correlation matrix corresponding to each tactile signal; the above correlation matrix includes the connection relationship and the degree of correlation between each tactile signal and other tactile signals; each target tactile signal is determined from each tactile signal according to the correlation matrix; and a second graph structure is constructed according to the connection relationship and the degree of correlation between each target tactile signal.

[0196] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0197] According to the correlation matrix, each target tactile signal that meets a correlation threshold is selected from the tactile signals.

[0198] In one embodiment, the second graph structure is an implicit feature graph structure.

[0199] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0200] Acquire tactile signals of the object to be classified; the tactile signals are acquired by collecting the object to be classified through different sensors.

[0201] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0202] Acquire multiple training tactile signal sets; each training tactile signal set includes multiple training tactile signals of sample objects and labeled categories of the sample objects; obtain spatial relationship parameters corresponding to each training tactile signal set; use each training tactile signal set and the corresponding spatial relationship parameters as input of the initial neural network, use the labeled categories of each training tactile signal set as reference output of the initial neural network, train the initial neural network, and obtain a neural network.

[0203] In one embodiment, the present application also provides a computer program product comprising instructions, which, when executed on a computer, causes the computer to perform the following steps:

[0204] Acquire tactile signals of the object to be classified; the tactile signals are acquired by collecting the object to be classified through different sensors; and classify and process the object to be classified according to the spatial relationship parameters between the sensors, the tactile signals and the neural network to determine the category of the object to be classified.

[0205] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0206] The spatial relationship parameters between each sensor and each tactile signal are used as the input of the graph attention network, and classification processing is performed in the graph attention network to determine the category of the object to be classified.

[0207] In one embodiment, the spatial relationship parameter includes a first graph structure between the sensors.

[0208] In one embodiment, the first graph structure represents the spatial connection relationship between the sensors.

[0209] In one embodiment, the first graph structure is an explicit feature graph structure.

[0210] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0211] A second graph structure corresponding to each tactile signal is determined according to each tactile signal; the first graph structure and the second graph structure are input into a graph attention network, and classification processing is performed in the graph attention network to determine the category of the object to be classified.

[0212] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0213] Correlation analysis is performed on each tactile signal to determine a second graph structure corresponding to each tactile signal.

[0214] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0215] Each tactile signal is input into the embedding layer for correlation analysis and processing to determine the correlation matrix corresponding to each tactile signal; the above correlation matrix includes the connection relationship and the degree of correlation between each tactile signal and other tactile signals; each target tactile signal is determined from each tactile signal according to the correlation matrix; and a second graph structure is constructed according to the connection relationship and the degree of correlation between each target tactile signal.

[0216] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0217] According to the correlation matrix, each target tactile signal that meets a correlation threshold is selected from the tactile signals.

[0218] In one embodiment, the second graph structure is an implicit feature graph structure.

[0219] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0220] Acquire tactile signals of the objects to be classified; the tactile signals are acquired by collecting the objects to be classified through different sensors.

[0221] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0222] Acquire multiple training tactile signal sets; each training tactile signal set includes multiple training tactile signals of sample objects and labeled categories of the sample objects; obtain spatial relationship parameters corresponding to each training tactile signal set; use each training tactile signal set and the corresponding spatial relationship parameters as input of the initial neural network, use the labeled categories of each training tactile signal set as reference output of the initial neural network, train the initial neural network, and obtain a neural network.

[0223] It should be noted that the data involved in this application (including but not limited to data used for analysis, data stored, data displayed, etc.) are all data fully authorized by all parties.

[0224] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory, etc. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0225] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0226] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A method for object classification, characterized in that: The method comprises: Acquiring tactile signals of the object to be classified; the tactile signals are acquired by collecting the object to be classified through different sensors; Classification processing is performed according to the spatial relationship parameters between the sensors, the tactile signals and the neural network to determine the category of the object to be classified.

2. The method according to claim 1, characterized in that The neural network includes a graph attention network, and the classification process is performed according to the spatial relationship parameters between the sensors, the tactile signals and the neural network to determine the category of the object to be classified, including: The spatial relationship parameters between the sensors and the tactile signals are used as inputs of the graph attention network, and classification processing is performed in the graph attention network to determine the category of the object to be classified.

3. The method according to claim 2, characterized in that The spatial relationship parameter includes a first graph structure between the sensors.

4. The method according to claim 3, characterized in that The first graph structure represents the spatial connection relationship between the sensors.

5. The method according to claim 3, characterized in that: The first graph structure is an explicit feature graph structure.

6. The method according to claim 3, characterized in that: The step of using the spatial relationship parameters between the sensors and the tactile signals as inputs of the graph attention network, performing classification processing in the graph attention network, and determining the category of the object to be classified includes: Determine, according to each of the tactile signals, a second graph structure corresponding to each of the tactile signals; The first graph structure and the second graph structure are input into the graph attention network, and classification processing is performed in the graph attention network to determine the category of the object to be classified.

7. The method according to claim 6, characterized in that The determining, according to each of the tactile signals, a second graph structure corresponding to each of the tactile signals comprises: Correlation analysis is performed on each of the tactile signals to determine a second graph structure corresponding to each of the tactile signals.

8. The method according to claim 7, characterized in that The neural network further includes an embedding layer, and the performing correlation analysis on each of the tactile signals to determine a second graph structure corresponding to each of the tactile signals includes: Inputting each of the tactile signals into the embedding layer for correlation analysis and processing to determine a correlation matrix corresponding to each of the tactile signals; the correlation matrix includes a connection relationship and a correlation degree between each of the tactile signals and other tactile signals; Determining each target tactile signal from each of the tactile signals according to the correlation matrix; The second graph structure is constructed according to the connection relationship and correlation degree between the target tactile signals.

9. The method according to claim 8, characterized in that The step of determining each target tactile signal from each tactile signal according to the correlation matrix comprises: According to the correlation matrix, each target tactile signal that meets a correlation threshold is selected from the tactile signals.

10. The method according to claim 6, characterized in that The second graph structure is an implicit feature graph structure.

11. The method according to claim 1, characterized in that: The step of obtaining a tactile signal of the object to be classified comprises: Acquire tactile signals of multiple channels of the object to be classified; each of the tactile signals is acquired by collecting the object to be classified through different sensors.

12. The method according to claim 1, characterized in that The training method of the neural network includes: Acquire multiple training tactile signal sets; each of the training tactile signal sets includes multiple training tactile signals of sample objects and labeled categories of the sample objects; Acquiring spatial relationship parameters corresponding to each of the training tactile signal sets; Each of the training tactile signal sets and the corresponding spatial relationship parameters are used as inputs of the initial neural network, and the labeled categories of each of the training tactile signal sets are used as reference outputs of the initial neural network. The initial neural network is trained to obtain the neural network.

13. An object classification device, characterized in that: The device comprises: A signal acquisition module, used to acquire tactile signals of the object to be classified; the tactile signals are acquired by collecting the object to be classified through different sensors; The classification module is used to perform classification processing according to the spatial relationship parameters between the sensors, the tactile signals and the neural network to determine the category of the object to be classified.

14. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 12 are implemented.

15. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 12 are implemented.

16. A chip, characterized in that: The chip comprises a programmable logic circuit and / or program instructions, and the steps of the method according to any one of claims 1 to 12 are implemented when the chip is executed.

17. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 12 are implemented.

Images