Method and system for target attribute recognition based on three-dimensional force information of mechanical finger tip

Abstract

The application discloses a target attribute recognition method and system based on three-dimensional force information of a mechanical finger tip, and five kinds of silica gel models with different hardness and three kinds of shapes are prepared; a mechanical hand with an optical three-dimensional force sensor is used to collect force data of the gripping experiment of the models, original data are preprocessed by feature dimension reduction, and the data set is divided into a training set and a test set; the data set is classified based on a support vector machine algorithm optimized by an improved cuckoo search; the training set is used to train the support vector machine classifier model, the cuckoo search method is used to optimize the support vector machine model hyperparameters, the test set is input into the classifier model to realize attribute classification, and the accuracy of the real class and the predicted class and the confusion matrix during attribute classification are obtained; and the newly collected data is classified and a three-dimensional force time sequence diagram is drawn, so that the hardness and shape attributes of the target gripped by the mechanical hand can be effectively classified, and the application has higher accuracy compared with other algorithms.

Description

Technical Field

[0001] This invention belongs to the field of robotics and machine learning algorithm technology, specifically relating to a target attribute recognition method and system based on three-dimensional force information of mechanical fingertips. Background Technology

[0002] With the continuous development of artificial intelligence and robotics, the application scope of robots is constantly expanding. In fields such as automated manufacturing, healthcare, and home services, robots need to have the ability to perceive and grasp different objects. To improve the grasping efficiency and success rate of robots, force sensor-based grasping methods are currently used. However, most force sensors are difficult to accurately reflect the shape and hardness of the object being grasped, causing robots to rely on previous experience or multiple attempts to obtain the correct grasping method, resulting in low efficiency. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to provide a target attribute recognition method and system based on three-dimensional force information of mechanical fingertips to address the shortcomings of the prior art. This method and system solve the technical problems of low classification and recognition accuracy and high implementation difficulty caused by the complexity and diversity of target attributes. It can quickly and accurately obtain the hardness and shape attribute information of the target object, thereby improving the flexibility and efficiency of robot grasping.

[0004] The present invention adopts the following technical solution:

[0005] A target attribute recognition method based on three-dimensional force information of a mechanical fingertip includes the following steps:

[0006] A robotic arm was used to perform a grasping experiment on a silicone model, collecting force data to construct a dataset. The dataset was preprocessed, and principal component analysis was used for feature dimensionality reduction. A pre-defined support vector machine (SVM) model was used to train the preprocessed dataset, and 10-fold cross-validation was employed to verify accuracy and prevent overfitting. Dynamic inertia weights and adjustment factors were introduced to improve the cuckoo search method, which was then used to optimize the hyperparameters of the SVM model. The SVM model was exported, including necessary information such as support vectors and segmentation hyperplanes. Matlab's GUI design capabilities were used to process and predict the input data to obtain the relevant attributes of the target object. Matlab's force time-series plotting function was used to achieve target attribute recognition.

[0007] Specifically, a robotic arm equipped with an optical 3D force sensor was used to perform grasping experiments on silicone models. Each of the 15 types of silicone models was grasped 40 times, with each grasp lasting 15 seconds. The force data from the Contactile sensor within 15 seconds was saved to a .csv file using ROS system commands. Matlab was used to read the raw data from the .csv file and save all the data to a file named data.csv. In addition, labels were manually created for each set of data, using the shape and hardness of the model as two sample labels, and the created labels were saved to a file named label.csv. The data.csv file and label.csv file were matched sequentially to complete the construction of the training dataset.

[0008] Furthermore, the silicone model specifically refers to:

[0009] Food-grade silicone solution for manual molding is used. Silicone A liquid and silicone B liquid are mixed in proportion to prepare 5 different hardness and 3 different shapes of silicone solution. Then, the solution is poured into the mold and stirred evenly. After standing and solidifying, it is demolded to obtain the final product.

[0010] Furthermore, the five different hardnesses are: -5 below Shore 0, 0 Shore 0, 5 Shore 5, 10 Shore 10, and 15 Shore 15; the three different shapes are: 1 for cylinder, 2 for sphere, and 3 for cube.

[0011] Specifically, the feature dimensionality reduction method using principal component analysis is as follows:

[0012] The dataset retains 97% of the maximum explained variance and hides the variance of the least important components. After dimensionality reduction, 10 feature components are retained. The dataset after dimensionality reduction is then split according to a ratio of 75% training set and 25% prediction set.

[0013] Specifically, the kernel function of the support vector machine model is a Gaussian function with a kernel scale of 21 and a bounding box constraint level of 1. It adopts a one-to-one multi-class classification method and standardizes the training data. The ten-fold cross-validation method is used to divide the dataset into ten folds and estimate the accuracy of each fold.

[0014] Specifically, the improved cuckoo search method is used to optimize the hyperparameters of the support vector machine model as follows:

[0015] S501. Initialize the support vector machine model, preset the number of bird nests, the maximum number of iterations T, the probability of parasitic nests being discovered p, the optimization dimension, and the hyperparameters of the support vector machine model;

[0016] S502. Calculate the individual fitness corresponding to each nest and select the optimal nest location X at this stage. best Add an adjustment factor to enhance the algorithm's global search capability;

[0017] S503. Update the bird's nest location and compare the fitness of the new and old bird's nests. If the calculation result is better, replace the old solution.

[0018] S504. Generate a random number r and compare it with the probability p of the bird's nest being discovered. If r > p, generate a new solution to replace the old solution.

[0019] S505. Recalculate the fitness of all nests and repeatedly update the optimal nest location X. best Output X until the preset maximum number of iterations T is reached. best The corresponding hyperparameter calculation values ​​are used as the optimal hyperparameter solution for the prediction model.

[0020] Furthermore, in step S502, the adjustment factor τ is:

[0021]

[0022] Where, τ min With τ max Let τ represent the maximum and minimum values ​​of the adjustment factor τ, where t and T represent the current iteration number and the maximum iteration number, respectively, and σ is a constant.

[0023] Furthermore, in step S503, the Bird's Nest is updated as follows:

[0024]

[0025] in, This represents the position of the a-th bird's nest after t iterations. Let represent the point-to-point multiplication sign, β represent the compensation control quantity, and Levy(λ) represent the search path.

[0026] Secondly, embodiments of the present invention provide a target attribute recognition system based on three-dimensional force information of a mechanical fingertip, comprising:

[0027] The data acquisition module uses a robotic arm to perform grasping experiments on a silicone model, collecting force data to build a dataset.

[0028] The preprocessing module preprocesses the dataset constructed by the acquisition module, using principal component analysis to reduce the dimensionality of features.

[0029] The training module uses a pre-defined support vector machine model to train the dataset preprocessed by the preprocessing module, and employs ten-fold cross-validation to verify accuracy and prevent overfitting.

[0030] The optimization module introduces dynamic inertia weights and adjustment factors to improve the cuckoo search method, and uses the improved cuckoo search method to optimize the hyperparameters of the support vector machine model obtained from the training module.

[0031] The identification module exports the support vector machine model obtained from the optimization module, including necessary information such as support vectors and segmentation hyperplanes. It uses Matlab's GUI design function to process and predict the input data to obtain the relevant attributes of the target object. It uses Matlab's force time series plotting function to realize target attribute identification.

[0032] Compared with the prior art, the present invention has at least the following beneficial effects:

[0033] A target attribute recognition method based on three-dimensional force information from robotic fingertips employs principal component analysis preprocessing and a support vector machine classification algorithm based on an improved cuckoo search optimization. This method effectively and accurately classifies the hardness and shape attributes of targets grasped by robotic hands, achieving higher accuracy compared to other algorithms. Furthermore, the GUI program provided by this invention can classify newly acquired data and plot time-series graphs of three-dimensional forces. It is simple to operate and easy to use, readily applicable to practical production, and provides effective support and guidance for robotic hand grasping tasks. This invention is of great significance for machines to simulate human attribute recognition, achieve force and tactile perception, and advance the development of intelligent machines.

[0034] Furthermore, gripping experiments were conducted on the fabricated silicone model, and the collected force data and corresponding sample labels were saved as a training dataset. This will provide useful data support for subsequent machine learning models, enabling them to recognize and grasp objects of various shapes and hardness.

[0035] Furthermore, using food-grade silicone molding solution to create silicone models allows for control over the hardness and shape of the models, resulting in 15 different hardnesses and shapes, providing diverse options for subsequent experiments. Simultaneously, the produced silicone models are completely identical in size, ensuring the accuracy of the experimental results.

[0036] Furthermore, setting up silicone models with different hardness and shapes aims to test the recognition performance of this method under various conditions, thereby evaluating its practicality and applicability. Specifically, by testing target objects with different hardnesses, different material properties and real-world application scenarios can be simulated, such as identifying the hardness variations of fruits, foods, or object surfaces with varying hardness. By testing target objects with different shapes, different real-world application scenarios can be simulated, such as identifying workpieces or products of different shapes, as well as irregular shape variations on object surfaces. Therefore, setting up target objects with different hardnesses and shapes helps to verify the generalization ability and applicability of this method, as well as increase its reliability and stability in practical applications.

[0037] Furthermore, the purpose of using Principal Component Analysis (PCA) is to reduce the dimensionality of multidimensional data, extract important feature information, and make the data easier to process and analyze. Specifically, in this patent, the collected robotic hand grasping experiment dataset is multidimensional, including force information of the robotic fingertips at different times for target objects of different shapes and hardness. In order to better identify the attributes of the target objects, it is necessary to extract the most important features from this multidimensional data for subsequent training and classification. The principle of the Principal Component Analysis (PCA) method is to map the original data to a new coordinate system through a linear transformation, so that the transformed data exhibits the maximum variance in the new coordinate system, that is, to extract the most important feature information in the data. Specifically, the Principal Component Analysis (PCA) method performs eigenvalue decomposition on the covariance matrix of the data to obtain a set of eigenvectors and corresponding eigenvalues. Projecting the original data onto the new coordinate system formed by these eigenvectors yields the dimensionality-reduced data. The dimensionality-reduced data still retains most of the information of the original data, but by selecting the most representative principal components, the influence of redundant information is reduced, improving the classification accuracy and generalization ability of the model.

[0038] Furthermore, a refined Gaussian support vector machine model was used to train the dataset. To prevent overfitting, a 10-fold cross-validation method was used to estimate the accuracy per fold, ultimately obtaining the accuracy for hardness and shape classification. The ROC curve of the model was also plotted to evaluate the classifier performance.

[0039] Furthermore, employing the cuckoo search method for hyperparameter optimization allows for global search to find the optimal solution, thereby improving the prediction accuracy and stability of the support vector machine model. In addition, introducing dynamic inertia weights and adjustment factors enhances the algorithm's adaptability and global search capability, contributing to improved convergence speed and optimization accuracy.

[0040] Furthermore, the purpose of setting the adjustment factor τ is to further improve the efficiency and performance of the search algorithm, preventing it from getting trapped in local optima too early and thus unable to continue searching for better solutions. The adjustment factor τ is a parameter that gradually decreases with the number of searches. Its function is to reduce the speed of flock flight and position updates in the algorithm, thereby increasing the number of explorations near better solutions and improving the algorithm's global search capability. Specifically, when the adjustment factor τ is large, the flock flight speed and position update speed are fast, making it easy to skip potential high-quality solutions during the search. Conversely, when the adjustment factor τ is small, the flock flight speed and position update speed are slow, which, while allowing for better exploration of local optima, results in lower efficiency when searching for the global optimum. Therefore, by gradually adjusting the size of the adjustment factor τ, a balance between the algorithm's global and local search capabilities can be maintained during the search process, ensuring that the algorithm does not get trapped in local optima during the global search while simultaneously allowing for deeper searches near local optima.

[0041] Furthermore, the main purpose of nest updating is to avoid falling into local optima during the search process, thereby increasing the probability of a global search and improving the algorithm's optimization performance. Specifically, in the cuckoo search algorithm, nest updating involves exchanging birds from worse nests with birds from better nests, making the search process more global. This avoids getting stuck in local optima and increases the algorithm's diversity. In addition, nest updating can increase the algorithm's search speed. Because worse nests often fail to provide useful information during the search, exchanging them with better nests allows the search process to converge faster.

[0042] It is understandable that the beneficial effects of the second aspect mentioned above can be found in the relevant descriptions in the first aspect mentioned above, and will not be repeated here.

[0043] In summary, this invention solves the problem of accurately perceiving the hardness and shape of objects in traditional grasping methods. The support vector machine model optimized by the improved cuckoo search method has better global search and adaptive capabilities, thereby improving the classification accuracy for nonlinear and high-dimensional data and providing more reliable classification results for robotic gripper force control. It can be applied to fields such as robot grasping and intelligent logistics, improving the operational precision and efficiency of robots and has broad application prospects.

[0044] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0045] Figure 1 This is a schematic diagram of a target attribute recognition method based on three-dimensional force information of a mechanical fingertip;

[0046] Figure 2 Optimize the support vector machine flowchart to improve the cuckoo search method;

[0047] Figure 3 This is a graph showing the minimum classification error of the model during the iterative optimization process.

[0048] Figure 4 ROC curves of the support vector machine models before and after optimization using the improved cuckoo search method are shown, where (a) is the preset SVM model and (b) is the ICSSVM model.

[0049] Figure 5 The confusion matrix diagram shows the hardness classification performance of the ICSSVM algorithm and three other algorithms. (a) is the KNN model, (b) is the Naive Bayes model, (c) is the two-layer neural network model, and (d) is the ICSSVM model.

[0050] Figure 6 The diagram shows the confusion matrix of the ICSSVM algorithm and three other algorithms for shape classification. (a) is the KNN model, (b) is the Naive Bayes model, (c) is the two-layer neural network model, and (d) is the ICSSVM model. Detailed Implementation

[0051] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0052] In the description of this invention, it should be understood that the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0053] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0054] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes such combinations. For example, A and / or B can represent three cases: A alone, A and B simultaneously, and B alone. Additionally, the character " / " in this invention generally indicates that the preceding and following objects have an "or" relationship.

[0055] It should be understood that although terms such as first, second, third, etc., may be used in the embodiments of the present invention to describe the preset range, these preset ranges should not be limited to these terms. These terms are only used to distinguish the preset ranges from one another. For example, without departing from the scope of the embodiments of the present invention, the first preset range may also be referred to as the second preset range, and similarly, the second preset range may also be referred to as the first preset range.

[0056] Depending on the context, the word "if" as used here can be interpreted as "when," "when," "in response to determination," or "in response to detection." Similarly, depending on the context, the phrase "if determination" or "if detection (of the stated condition or event)" can be interpreted as "when determination," "in response to determination," "when detection (of the stated condition or event)," or "in response to detection (of the stated condition or event)."

[0057] The accompanying drawings illustrate various structural schematic diagrams according to embodiments disclosed in this invention. These drawings are not to scale, and some details have been enlarged for clarity, and some details may have been omitted. The shapes of the various regions and layers shown in the drawings, as well as their relative sizes and positional relationships, are merely exemplary and may deviate from reality due to manufacturing tolerances or technical limitations. Furthermore, those skilled in the art can design regions / layers with different shapes, sizes, and relative positions as needed.

[0058] This invention provides a target attribute recognition method based on three-dimensional force information from a mechanical fingertip. It employs an improved cuckoo search method to optimize the support vector machine model, resulting in higher accuracy and stability in recognizing target attributes of different shapes and hardnesses. Furthermore, meticulous adjustments to feature selection and parameter optimization shorten the model training time, thus achieving better adaptability to real-world application scenarios. This method is suitable for applications such as robot and intelligent hand recognition of target attributes, and has broad application prospects.

[0059] The relevant experimental equipment includes:

[0060] Food-grade silicone for manual molding, Beijing Yinshi Robotics robotic arm, Contactile optical 3D force sensor with a sampling frequency of 1000Hz, and computer Windows and ROS systems.

[0061] Please see Figure 1 The present invention discloses a target attribute recognition method based on three-dimensional force information of a mechanical fingertip, comprising the following steps:

[0062] S1. Prepare 5 different hardnesses and 3 different shapes of silicone solution using food-grade silicone solution for hand molding in proportion. After standing and solidifying, demold to obtain the model.

[0063] Food-grade silicone molding compound was used to make the silicone models. Silicone A and silicone B were mixed in a specific ratio, poured into the mold, and stirred thoroughly. After standing for 6 hours to solidify, the models were manually demolded. A total of 15 silicone models with different hardnesses and shapes were made: cylinders below 0 degrees, cylinders 0 degrees, cylinders 5 degrees, cylinders 10 degrees, cylinders 15 degrees; spheres below 0 degrees, spheres 0 degrees, spheres 5 degrees, spheres 10 degrees, spheres 15 degrees; cubes below 0 degrees, cubes 0 degrees, cubes 5 degrees, cubes 10 degrees, and cubes 15 degrees. The hardness unit is Shore hardness (degrees). Models of the same shape are identical in size.

[0064] S2. Using a robotic arm from Beijing Yinshi Company, equipped with an optical three-dimensional force sensor from Contactile Company, a grasping experiment was conducted on the silicone model made in step S1. After collecting force data, a dataset was constructed for training the model.

[0065] The silicone models created in step S1 were subjected to grasping experiments using a robotic arm equipped with a Contactile optical 3D force sensor. Each of the 15 models was grasped 40 times, with each grasp lasting 15 seconds. The force data from the Contactile sensor over the 15 seconds was saved to a .csv file using ROS system commands. The data sampling frequency was 1000Hz, with a total of 600 sets. Matlab was used to read the raw data from the .csv file and save all data to a file named data.csv. Furthermore, labels were manually created for each set of data, using the model's shape and hardness as two sample labels. The hardness labels were: -5 (below 0 degrees), 0 (0 degrees), 5 (5 degrees), 10 (10 degrees), 15 (15 degrees) (unit: Shore hardness); the shape labels were: 1 (cylinder), 2 (sphere), 3 (cube). The created labels were saved to a corresponding label.csv file. The data.csv file and label.csv file were then matched sequentially to complete the construction of the training dataset.

[0066] S3. Perform preprocessing steps on the dataset constructed in step S2, including invalid data fragment pruning and three-dimensional data shaping, and use principal component analysis (PCA) to perform feature dimensionality reduction, and split the preprocessed dataset into training set and test set;

[0067] The specific methods for data preprocessing are as follows: cropping invalid data fragments, shaping the data into three-dimensional data, and not making any changes to the frequency or amplitude of the original data in order to retain the effective information to the greatest extent.

[0068] The specific method for feature dimensionality reduction of the data is as follows: Principal component analysis (PCA) is used to perform feature dimensionality reduction on the dataset constructed in step S2: 97% of the maximum explained variance is retained and the variance of the least important component is hidden. After dimensionality reduction, 10 feature components are retained. Then, the dataset after feature dimensionality reduction is split according to the ratio of 75% training set and 25% prediction set.

[0069] S4. Use the preset support vector machine model to initially train the dataset after preprocessing in step S3. Use the ten-fold cross-validation method to verify the accuracy and prevent overfitting, and obtain the classification accuracy and ROC curve of the trained model.

[0070] The model used for training the dataset was a pre-defined fine-grained Gaussian support vector machine model with a Gaussian kernel function, a kernel scale of 21, and a bounding box constraint level of 1. A one-to-one multi-class classification method was employed, and the training data was standardized. Ten-fold cross-validation was then used to prevent overfitting by dividing the dataset into ten folds and estimating the accuracy of each fold. The final validation results showed accuracies of 87.8% for hardness and 88.6% for shape classification, and the ROC curves of the model were plotted.

[0071] S5. Use an improved cuckoo search method to optimize the hyperparameters of the support vector machine model trained in step S4. The optimization of the cuckoo search method introduces dynamic inertia weights and adjustment factors to enhance the algorithm's adaptability and global search capability.

[0072] Please see Figure 2 An improved cuckoo search method is used to optimize the hyperparameters of the preset support vector machine model. Therefore, the problem of finding the optimal solution for the model's hyperparameters is transformed into finding the optimal nest location using the cuckoo search method. The specific process is as follows:

[0073] S501. Initialize the model, pre-setting the number of bird nests, the maximum number of iterations T, the probability of parasitic nests being discovered p, the optimization dimension, and the hyperparameters of the support vector machine model. The hyperparameters include: kernel function, kernel scale, and box constraint level.

[0074] S502. Calculate the individual fitness corresponding to each nest and select the optimal nest location X at this stage. best Add an adjustment factor to enhance the algorithm's global search capability;

[0075] The adjustment factor τ is:

[0076]

[0077] Where, τ min With τ max Let τ represent the maximum and minimum values ​​of the adjustment factor τ, where t and T represent the current iteration number and the maximum iteration number, respectively, and σ is a constant.

[0078] S503. Update the bird's nest location and compare the fitness of the new and old nests. If the calculation result is better, replace the old solution. The bird's nest update uses the following formula:

[0079]

[0080] in, This represents the position of the a-th (a = 1, 2, ..., n) bird's nest after t iterations. The symbol represents point-to-point multiplication, β represents the compensation control quantity, which can be adjusted, and Levy(λ) represents the search path.

[0081] After 300 iterations, the optimized hyperparameters are obtained, which are then substituted into the model to obtain the optimized new model. The prediction accuracy is calculated, and the ROC curve is plotted.

[0082] S504. Generate a random number r and compare it with the probability p of the bird's nest being discovered. If r > p, then generate a new solution to replace the old solution.

[0083] S505. Recalculate the fitness of all nests and repeatedly update the optimal nest location X. best Output X until the preset maximum number of iterations T is reached. best The corresponding hyperparameter calculation values ​​are used as the optimal hyperparameter solution for the prediction model.

[0084] Finally, the optimized support vector machine model (ICSSVM) based on the improved cuckoo algorithm achieved a classification accuracy of 96.2% for hardness and 96.0% for shape. The classification confusion matrix, ROC curve, and minimum classification error during the iteration process were plotted.

[0085] The improvements to the cuckoo search method are reflected in:

[0086] Dynamic inertia weights are introduced so that the value of p is not fixed. The values ​​of p are as follows:

[0087]

[0088] Where newp represents the dynamic discovery probability, satisfying newp∈[0.2, 0.5], ω max and ω min Let represent the maximum and minimum values ​​of the inertia weight ω, respectively, and let r represent a random number within the interval (0, 1). The constant represents the control of weight offset, and frnd() represents a non-symmetric random number.

[0089] Adding an adjustment factor can accelerate convergence and improve the algorithm's adaptability; therefore, the cuckoo's position after adding the adjustment factor is updated as follows:

[0090]

[0091] S6. Export the optimized model obtained in step S5, and use the GUI design function of Matlab to design a program. The functions include: classifying and predicting the shape and hardness of the target object, and drawing three-dimensional force time series graphs of different fingers.

[0092] The GUI program was designed using Matlab. The program's functions include: reading the data to be classified, selecting a classifier model, performing prediction based on the hardness and shape of the grasping target corresponding to the imported data, calculating the accuracy of the 10-fold cross-validation, and plotting the time series graphs of the three-dimensional force at the fingertips of the three mechanical fingers and the resultant force of the three fingers.

[0093] Using Matlab to design a GUI program enables convenient prediction of target object shape and hardness, as well as the plotting of force time series graphs, improving operational convenience and efficiency. Furthermore, comparing the optimized model with other algorithm classifiers verifies its accuracy and usability, providing a basis for practical applications.

[0094] S7. Compare and analyze the optimized model from step S6 with other algorithm classifier models to verify the accuracy and usability of the model.

[0095] In the comparative analysis, other classifier models used included the KNN classifier, the Naive Bayes classifier, and the two-layer neural network classifier, which achieved accuracies of 85.3%, 90.0%, and 88.8% for classifying the hardness of the target object, respectively; and 91.5%, 86.8%, and 90.8% for classifying the shape of the target object, respectively.

[0096] In another embodiment of the present invention, a target attribute recognition system based on three-dimensional force information of a mechanical fingertip is provided. This system can be used to implement the above-mentioned target attribute recognition method based on three-dimensional force information of a mechanical fingertip. Specifically, the target attribute recognition system based on three-dimensional force information of a mechanical fingertip includes an acquisition module, a preprocessing module, a training module, an optimization module, and a recognition module.

[0097] The data acquisition module uses a robotic arm to perform a grasping experiment on a silicone model and collects force data to build a dataset.

[0098] The preprocessing module preprocesses the dataset constructed by the acquisition module, using principal component analysis to reduce the dimensionality of features.

[0099] The training module uses a pre-defined support vector machine model to train the dataset preprocessed by the preprocessing module, and employs ten-fold cross-validation to verify accuracy and prevent overfitting.

[0100] The optimization module introduces dynamic inertia weights and adjustment factors to improve the cuckoo search method, and uses the improved cuckoo search method to optimize the hyperparameters of the support vector machine model obtained from the training module.

[0101] The identification module exports the support vector machine model obtained from the optimization module, including necessary information such as support vectors and segmentation hyperplanes. It uses Matlab's GUI design function to process and predict the input data to obtain the relevant attributes of the target object. It uses Matlab's force time series plotting function to realize target attribute identification.

[0102] In another embodiment of the present invention, a terminal device is provided, comprising a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions to achieve a corresponding method flow or corresponding function. The processor described in this embodiment of the present invention can be used for the operation of a target attribute recognition method based on three-dimensional force information of a mechanical fingertip, including:

[0103] A robotic arm was used to perform a grasping experiment on a silicone model, collecting force data to construct a dataset. The dataset was preprocessed, and principal component analysis was used for feature dimensionality reduction. A pre-defined support vector machine (SVM) model was used to train the preprocessed dataset, and 10-fold cross-validation was employed to verify accuracy and prevent overfitting. Dynamic inertia weights and adjustment factors were introduced to improve the cuckoo search method, which was then used to optimize the hyperparameters of the SVM model. The SVM model was exported, including necessary information such as support vectors and segmentation hyperplanes. Matlab's GUI design capabilities were used to process and predict the input data to obtain the relevant attributes of the target object. Matlab's force time-series plotting function was used to achieve target attribute recognition.

[0104] In another embodiment of the present invention, a storage medium is also provided, specifically a computer-readable storage medium (memory). This computer-readable storage medium is a memory device in a terminal device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the terminal device and extended storage media supported by the terminal device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, this storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM or non-volatile memory, such as at least one disk storage device.

[0105] One or more instructions stored in a computer-readable storage medium can be loaded and executed by a processor to implement the corresponding steps of the target attribute recognition method based on three-dimensional force information of a mechanical fingertip in the above embodiments; one or more instructions in the computer-readable storage medium are loaded and executed by the processor in the following steps:

[0106] A robotic arm was used to perform a grasping experiment on a silicone model, collecting force data to construct a dataset. The dataset was preprocessed, and principal component analysis was used for feature dimensionality reduction. A pre-defined support vector machine (SVM) model was used to train the preprocessed dataset, and 10-fold cross-validation was employed to verify accuracy and prevent overfitting. Dynamic inertia weights and adjustment factors were introduced to improve the cuckoo search method, which was then used to optimize the hyperparameters of the SVM model. The SVM model was exported, including necessary information such as support vectors and segmentation hyperplanes. Matlab's GUI design capabilities were used to process and predict the input data to obtain the relevant attributes of the target object. Matlab's force time-series plotting function was used to achieve target attribute recognition.

[0107] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0108] Please see Figure 3 and Figure 4 After 300 iterations of hyperparameter optimization, the ICSSVM model improved the classification accuracy of hardness and shape from the initial 87.8% and 88.6% to 96.2% and 96.0%, respectively. It is also worth noting that the ROC curve of the optimized model is closer to the upper left of the coordinate system, and the AUC value also increased from 0.87 to 0.98, which fully demonstrates the good optimization performance of the improved cuckoo search method.

[0109] Please see Figure 5 and Figure 6 The ICSSVM model proposed in this invention has a better effect on the classification and prediction of hardness and shape attributes than other algorithm models, and effectively solves the problems of low accuracy and high difficulty in the classification and recognition of target attributes in existing technologies.

[0110] In summary, this invention presents a target attribute recognition method and system based on three-dimensional force information from a robotic fingertip. Force data is collected through a silicone model grasping experiment. Principal component analysis is used to preprocess the raw data and perform feature dimensionality reduction. A support vector machine algorithm based on an improved cuckoo search optimization is employed for classification. The newly collected data is then classified and visualized through a GUI program. Experimental results show that this invention can effectively classify the hardness and shape attributes of targets grasped by a robotic hand, achieving higher accuracy compared to other algorithms.

[0111] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0112] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0113] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed in this invention can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0114] In the embodiments provided by this invention, it should be understood that the disclosed devices / terminals and methods can be implemented in other ways. For example, the device / terminal embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0115] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0116] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0117] If the integrated module / unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0118] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0119] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0120] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0121] The above content is only for illustrating the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. Any modifications made to the technical solution based on the technical concept proposed in this invention shall fall within the scope of protection of the claims of this invention.

Claims

1. A target attribute recognition method based on three-dimensional force information of a mechanical fingertip, characterized in that, A robotic arm was used to perform grasping experiments on silicone models. Each of the 15 silicone models was grasped 40 times, with each grasp lasting 15 seconds. Force data was collected to construct a dataset. The force data from the Contactile sensor within 15 seconds was saved to a .csv file using ROS system commands. Matlab was used to read the raw data from the .csv file and save all data to a file named data.csv. Additionally, labels were manually created for each data set, using the model's shape and hardness as two sample labels. The created labels were saved to a file named label.csv. Complete the construction of the training dataset by matching the data.csv file and the label.csv file in sequence; The dataset is preprocessed, and principal component analysis is used to reduce the dimensionality of the features. A pre-defined support vector machine model is used to train the preprocessed dataset, and ten-fold cross-validation is used to verify the accuracy and prevent overfitting. The cuckoo search method is improved by introducing dynamic inertia weights and adjustment factors. The improved cuckoo search method is then used to optimize the hyperparameters of the support vector machine model, specifically: Initialize the support vector machine model, pre-setting the number of nests and the maximum number of iterations. Probability of parasitic nests being discovered The optimization dimension and hyperparameters of the support vector machine model; Calculate the individual fitness for each nest and select the optimal nest location at this stage. Add an adjustment factor to enhance the algorithm's global search capability; Update the nest location and compare the fitness of the new and old nests. If the calculation result is better, replace the old solution. Generate random numbers Probability of Bird's Nest being discovered To make a comparison, if > This generates a new solution that replaces the old one; Recalculate the fitness of all nests and repeatedly update the optimal nest location. Until the number of iterations reaches the preset maximum number of iterations. Output The corresponding hyperparameter calculation values ​​are used as the optimal solution of the hyperparameters of the prediction model; Export the support vector machine model, including support vectors and segmentation hyperplane information. Use Matlab's GUI design function to process and predict the input data to obtain the relevant attributes of the target object. Use Matlab's force time series plotting function to achieve target attribute recognition.

2. The target attribute recognition method based on three-dimensional force information of a mechanical fingertip according to claim 1, characterized in that, The silicone model is specifically as follows: Food-grade silicone solution for manual molding is used. Silicone A liquid and silicone B liquid are mixed in proportion to prepare 5 different hardness and 3 different shapes of silicone solution. Then, the solution is poured into the mold and stirred evenly. After standing and solidifying, it is demolded to obtain the final product.

3. The target attribute recognition method based on three-dimensional force information of a mechanical fingertip according to claim 2, characterized in that, The five different hardnesses are: -5 below Shore 0, 0 Shore 0, 5 Shore 5, 10 Shore 10, and 15 Shore 15; the three different shapes are:

1. Column, 2. Sphere, and 3. Cube.

4. The target attribute recognition method based on three-dimensional force information of a mechanical fingertip according to claim 1, characterized in that, The specific steps for feature dimensionality reduction using principal component analysis are as follows: The dataset retains 97% of the maximum explained variance and hides the variance of the least important components. After dimensionality reduction, 10 feature components are retained. The dataset after dimensionality reduction is then split according to a ratio of 75% training set and 25% prediction set.

5. The target attribute recognition method based on three-dimensional force information of a mechanical fingertip according to claim 1, characterized in that, The kernel function of the support vector machine model is a Gaussian function with a kernel scale of 21 and a bounding box constraint level of 1. A one-to-one multi-class classification method is used, and the training data is standardized. A ten-fold cross-validation method is used to divide the dataset into ten folds and estimate the accuracy of each fold.

6. The target attribute recognition method based on three-dimensional force information of a mechanical fingertip according to claim 1, characterized in that, Adjustment factor for: in, τ min and τ max Indicates adjustment factor τ The maximum and minimum values, t and T These represent the current number of iterations and the maximum number of iterations, respectively. σ It is a constant.

7. The target attribute recognition method based on three-dimensional force information of a mechanical fingertip according to claim 1, characterized in that, The Bird's Nest has been updated as follows: in, This represents the position of the a-th bird's nest after t iterations, where a = 1, 2, ..., n. Indicates the dot multiplication symbol. β Indicates the compensation control amount. Indicates the search path.

8. A target attribute recognition system based on three-dimensional force information of a mechanical fingertip, characterized in that, include: The data acquisition module uses a robotic arm to perform grasping experiments on silicone models. Each of the 15 types of silicone models was grasped 40 times, with each grasp lasting 15 seconds. Force data was collected to construct a dataset. The force data from the Contactile sensor within 15 seconds was saved to a .csv file using ROS system commands. Matlab was used to read the raw data from the .csv file and save all data to a file named data.csv. Additionally, labels were manually created for each data set, using the model's shape and hardness as two sample labels. The created labels were saved to a corresponding file named label.csv. Complete the construction of the training dataset by matching the data.csv file and the label.csv file in sequence; The preprocessing module preprocesses the dataset constructed by the acquisition module, using principal component analysis to reduce the dimensionality of features. The training module uses a pre-defined support vector machine model to train the dataset preprocessed by the preprocessing module, and employs ten-fold cross-validation to verify accuracy and prevent overfitting. The optimization module introduces dynamic inertia weights and adjustment factors to improve the cuckoo search method. The improved cuckoo search method is then used to optimize the hyperparameters of the support vector machine model obtained from the training module. Specifically: Initialize the support vector machine model, pre-setting the number of nests and the maximum number of iterations. Probability of parasitic nests being discovered The optimization dimension and hyperparameters of the support vector machine model; Calculate the individual fitness for each nest and select the optimal nest location at this stage. Add an adjustment factor to enhance the algorithm's global search capability; Update the nest location and compare the fitness of the new and old nests. If the calculation result is better, replace the old solution. Generate random numbers Probability of Bird's Nest being discovered To make a comparison, if > This generates a new solution that replaces the old one; Recalculate the fitness of all nests and repeatedly update the optimal nest location. Until the number of iterations reaches the preset maximum number of iterations. Output The corresponding hyperparameter calculation values ​​are used as the optimal solution of the hyperparameters of the prediction model; The recognition module exports the support vector machine model obtained from the optimization module, including support vectors and segmentation hyperplane information. It uses Matlab's GUI design function to process and predict the input data to obtain the relevant attributes of the target object. It uses Matlab's force time series plotting function to realize target attribute recognition.

Images