Dexterous hand slip sense signal identification method based on KPCA-ICS-BP
The KPCA-ICS-BP method is used to optimize data dimensionality reduction and cuckoo search algorithm, which solves the problems of large calculation volume, low recognition rate and poor parameter optimization of traditional sliding signal recognition methods, and achieves more efficient sliding signal recognition.
Patent Information
- Application Number
- CN202510278709.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-10
- Publication Date
- 2025-07-01
AI Technical Summary
The existing sliding signal recognition method has a large amount of calculation, a lot of time and a low recognition rate under complex operating conditions. In addition, the optimization of traditional BP neural network parameters has the problem of low optimization accuracy, slow speed, and easy to fall into local minimum values.
The KPCA-ICS-BP method is used to optimize the weight, threshold and learning rate of the BP neural network through data dimensionality reduction and improved cuckoo search algorithm to build the optimal BP model.
It significantly improves the recognition accuracy of sliding signal, provides a more efficient recognition method, and overcomes the shortcomings of traditional methods.
Smart Images
Figure CN120234577A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of bionic tactile perception, and in particular relates to a method for identifying dexterous hand slip signals based on KPCA-ICS-BP. Background Art
[0002] The current application scenarios of dexterous hands are becoming increasingly complex, and the requirements for the recognition of slip signals are becoming higher and higher. The existing slip signal recognition methods can no longer effectively cope with complex and changeable working conditions, and there is an urgent need to study new and efficient recognition methods. Traditional slip signal recognition methods mainly include time-frequency feature analysis, support vector machines, k-nearest neighbor classification, etc. These methods are often affected by factors such as whether the sample set is linearly separable, parameter estimation accuracy, and the number of training samples. Not only are they computationally intensive, time-consuming, and have low recognition rates, but they also require manual verification, making it difficult to obtain satisfactory recognition results.
[0003] With the advancement of artificial intelligence technology, many intelligent algorithms have emerged. As a classic artificial neural network model, BP neural network has powerful nonlinear mapping ability and self-learning ability, and has been widely used in the field of pattern recognition. However, BP neural network has problems such as difficulty in determining the network structure, improper selection of learning rate, and easy to fall into local minima. These factors directly affect the prediction accuracy and generalization ability of the network. In recent years, many scholars have conducted in-depth research on the parameter optimization method of BP neural network.
[0004] Studies have shown that the reasonable selection of key parameters such as network weights, thresholds and learning rates of BP neural networks is the key to improving their prediction accuracy. Intelligent optimization algorithms are often used to select key parameters of BP neural networks, including genetic algorithms (GA), grid search methods, particle swarm optimization (PSO), artificial fish swarm algorithms (AFSA), simulated annealing algorithms, fruit fly optimization algorithms (FOA), and grey wolf optimization (GWO). Experiments have shown that the parameter optimization effect of BP neural networks based on the cuckoo algorithm is better than other methods, but there are also some shortcomings, such as low optimization accuracy, slow optimization speed, and easy to fall into local optimal values in the later stage of the algorithm. Therefore, how to further improve the efficiency and accuracy of BP neural network parameter optimization is still the focus and difficulty of current research. Summary of the invention
[0005] The purpose of the present invention is to provide a KPCA-ICS-BP method for identifying dexterous hand slip signals, wherein KPCA (Kernel Principal Component Analysis) is a common data processing method, which is often used for dimensionality reduction of high-dimensional data, can be used to extract the main characteristic components of data, improve the cuckoo algorithm ICS (Improved Cuckoo Search), BP (Back Propagation) neural network.
[0006] A method for identifying hand slip signals based on KPCA-ICS-BP, comprising the following steps:
[0007] Step S1. Using a sensing device to collect the slip signal data generated when the dexterous hand grasps an object;
[0008] Step S2. Standardize the sliding signal data to obtain a data set feature matrix;
[0009] Step S3. Use KPCA to reduce the dimension of the feature matrix of the data set, select the first n principal component array matrices with a cumulative variance contribution rate of 95% to replace the original feature matrix, and perform dimensionality reduction processing;
[0010] Step S4. Set the parameters of the model, the number of bird nests N, the maximum number of iterations tmax, the abandonment probability Pb, the weight w, the threshold th and the upper and lower boundaries of the learning rate lr, and the elimination factor θ;
[0011] Step S5. Setting the fitness function; using the model prediction accuracy as the fitness function, the accuracy calculation method is the ratio of the number of accurate prediction samples to the total number of samples;
[0012] Step S6. Divide the reduced-dimensional data set into a training set and a test set, use the labeled training set to train the KPCA-ICS-BP model and use ICS to optimize the weight w, threshold th and learning rate lr of the BP neural network;
[0013] Step S7. Select the optimal weight W, threshold TH and learning rate LR as the training parameters of the BP model to construct the optimal BP model;
[0014] Step S8. Use the trained KPCA-ICS-BP model to identify the slip signals in the unlabeled test set, that is, use the slip signal test set as input data and input it into the trained KPCA-ICS-BP model for calculation, and count the recognition accuracy and total running time.
[0015] The specific process of the above step S1 is:
[0016]
[0017] where x * is the feature matrix of the dataset after normalization, x is the feature matrix of the dataset before normalization, mean(x) is the mean of the feature matrix of the dataset, std(x) is the standard deviation of the feature matrix of the dataset, and the result of standardization is between (0,1).
[0018] The specific process of the above step S4 is:
[0019] Using mean square error as the fitness function, the smaller the mean square error, the higher the model accuracy:
[0020] .
[0021] The KPCA-ICS-BP recognition method proposed in this paper optimizes the BP network weight w, threshold th and learning rate lr through data dimension reduction and improved cuckoo search algorithm, which overcomes the defect that the traditional method is difficult to obtain the optimal model parameters and improves the slip signal recognition effect. By comparing with the classic slip signal recognition method, the present invention greatly improves the accuracy of slip signal recognition and provides a new research idea for the slip signal recognition of dexterous hands. BRIEF DESCRIPTION OF THE DRAWINGS
[0022] Figure 1 This is the flow chart of the KPCA-ICS-BP model’s recognition of slip signals.
[0023] Figure 2 It is the fitness graph of KPCA-ICS-BP and CS-BP.
[0024] Figure 3 This is a comparison chart of the prediction types of the training set.
[0025] Figure 4 This is a comparison chart of the prediction types of the test set. DETAILED DESCRIPTION
[0026] A method for identifying hand slip signals based on KPCA-ICS-BP, comprising the following steps:
[0027] Step S1. Using a sensing device to collect the slip signal data generated when the dexterous hand grasps an object;
[0028] Step S2. Standardize the sliding signal data to obtain a data set feature matrix;
[0029] Step S3. Use KPCA to reduce the dimension of the feature matrix of the data set, select the first n principal component array matrices with a cumulative variance contribution rate of 95% to replace the original feature matrix, and perform dimensionality reduction processing;
[0030] Step S4. Set the parameters of the model, the number of bird nests N, the maximum number of iterations tmax, the abandonment probability Pb, the weight w, the threshold th and the upper and lower boundaries of the learning rate lr, and the elimination factor θ;
[0031] Step S5. Setting the fitness function; using the model prediction accuracy as the fitness function, the accuracy calculation method is the ratio of the number of accurate prediction samples to the total number of samples;
[0032] Step S6. Divide the reduced-dimensional data set into a training set and a test set, use the labeled training set to train the KPCA-ICS-BP model and use ICS to optimize the weight w, threshold th and learning rate lr of the BP neural network;
[0033] Step S7. Select the optimal weight W, threshold TH and learning rate LR as the training parameters of the BP model to construct the optimal BP model;
[0034] Step S8. Use the trained KPCA-ICS-BP model to identify the slip signals in the unlabeled test set, that is, use the slip signal test set as input data and input it into the trained KPCA-ICS-BP model for calculation, and count the recognition accuracy and total running time.
[0035] The specific process of the above step S1 is:
[0036]
[0037] where x * is the feature matrix of the dataset after normalization, x is the feature matrix of the dataset before normalization, mean(x) is the mean of the feature matrix of the dataset, std(x) is the standard deviation of the feature matrix of the dataset, and the result of standardization is between (0,1).
[0038] The specific process of the above step S4 is:
[0039] Using mean square error as the fitness function, the smaller the mean square error, the higher the model accuracy:
[0040] .
[0041] Establish the KPCA-ICS-BP prediction model. The specific steps are as follows:
[0042] (1) Data preprocessing. First, the slip signal data is standardized to between (0, 1) to facilitate the subsequent model calculation:
[0043]
[0044] (2) Then KPCA is used to reduce the dimension of the feature matrix of the data set, and the first n principal component array matrices with a cumulative variance contribution rate of 95% are selected to replace the original matrix to achieve the purpose of dimensionality reduction. This operation can reduce the model construction time;
[0045] (3) Set the parameters of the model, the number of bird nests N, the maximum number of iterations tmax, the abandonment probability Pb, the weight w, the threshold th and the upper and lower bounds of the learning rate lr, and the elimination factor θ;
[0046] (4) Set the fitness function. Use the model prediction accuracy as the fitness function;
[0047] (5) The reduced-dimensional dataset is divided into a training set and a test set. The KPCA-ICS-BP model is trained with the labeled training set and ICS is used to optimize the weight w, threshold th, and learning rate lr of the BP neural network.
[0048] (6) Select the optimal parameter W best , TH best LR best As the training parameters of the BP model, the optimal BP model is constructed;
[0049] (7) Use the trained KPCA-ICS-BP model to identify the slip signals in the unlabeled test set, and calculate the recognition accuracy and total running time. The KPCA-ICS-BP model is used to identify the slip signals. Figure 1 shown.
[0050] The parameter settings of CS-BP and KPCA-ICS-BP are as follows: total number of bird nests is 20, maximum number of iterations tmax=100, discovery probability Pb=0.25, network weight w ranges from [0.001,1000], threshold th ranges from [0.001,1000], learning rate lr ranges from [0.0001,1], and elimination factor θ=3 of KPCA-ICS-BP model. The program was written using MATLAB mathematical software, and the improved cuckoo algorithm was used to optimize the network weight w, threshold th, and learning rate lr. The optimal values were 0.1562, 0.00371, and 0.03, respectively. One-to-one classification method was used for BP classification. The optimal fitness values of each iteration of the two algorithms were compared, as shown in the figure. Figure 2 As shown, from Figure 2 It can be seen that with the increase of the number of iterations, the fitness of the improved cuckoo search algorithm tends to be stable around the 8th generation, that is, it reaches the convergence value, and the fitness is significantly better than the traditional cuckoo search algorithm. The CS-BP algorithm reaches convergence at about the 10th generation, indicating that the ICS-BP algorithm is higher than the traditional CS-BP algorithm in convergence speed and accuracy.
[0051] 500 sample data are randomly selected from the overall data set. Among them, 400 sample data are designated for model training, and the remaining 100 sample data are used as test samples for subsequent model performance testing. The test results are as follows Figure 3 Comparison chart of training set prediction types and Figure 4 Comparison chart of prediction types of test sets. It can be seen from the figure that when the samples are trained by KPCA-ICS-BP, the training accuracy is as high as 99.5%, but when the trained model is used to test the test set, the accuracy is 95%. Therefore, the number of training sets needs to be further expanded to improve the recognition accuracy of the model.
[0052] The number of training set samples is continuously increased, and the statistical recognition accuracy of KPCA-ICS-BP, CS-BP and BP is shown in Table 1 below:
[0053] Table 1 KPCA-ICS-BP, CS-BP and BP recognition accuracy (%)
[0054]
[0055] When the number of training samples is 500, 1000, 2000 and 3000 respectively, the recognition accuracy of the test set of KPCA-ICS-BP, CS-BP and BP in total recognition increases, and the accuracy of BP, CS-BP and KPCA-ICS-BP increases in order, indicating that the improved cuckoo algorithm can significantly improve the recognition accuracy of the BP model by optimizing the network weight w, threshold th and learning rate lr.
[0056] The KPCA-ICS-BP classification method proposed in the present invention optimizes the BP network weight w, threshold th and learning rate lr through data dimension reduction and improved cuckoo search algorithm, thereby overcoming the defect that the traditional method is difficult to obtain the most reasonable parameters, and improving the recognition effect of dexterous hand slip signals. By comparing the method with the classic slip signal recognition method, the recognition method of the present invention greatly improves the accuracy of dexterous hand slip signal recognition, and provides a new research idea for dexterous hand slip signal recognition.
Claims
1. A method for identifying hand slip signals based on KPCA-ICS-BP, characterized in that The steps include: Step S1. Using a sensing device to collect the slip signal data generated when the dexterous hand grasps an object; Step S2. Standardize the sliding signal data to obtain a data set feature matrix; Step S3. Use KPCA to reduce the dimension of the feature matrix of the data set, select the first n principal component array matrices with a cumulative variance contribution rate of 95% to replace the original feature matrix, and perform dimensionality reduction processing; Step S4. Set the parameters of the model, the number of bird nests N, the maximum number of iterations tmax, the abandonment probability Pb, the weight w, the threshold th and the upper and lower boundaries of the learning rate lr, and the elimination factor θ; Step S5. Setting the fitness function; using the model prediction accuracy as the fitness function, the accuracy calculation method is the ratio of the number of accurate prediction samples to the total number of samples; Step S6. Divide the reduced-dimensional data set into a training set and a test set, use the labeled training set to train the KPCA-ICS-BP model and use ICS to optimize the weight w, threshold th and learning rate lr of the BP neural network; Step S7. Select the optimal weight W, threshold TH and learning rate LR as the training parameters of the BP model to construct the optimal BP model; Step S8. Use the trained KPCA-ICS-BP model to identify the slip signals in the unlabeled test set, that is, use the slip signal test set as input data and input it into the trained KPCA-ICS-BP model for calculation, and count the recognition accuracy and total running time.
2. The method for identifying hand slip signals based on KPCA-ICS-BP according to claim 1, characterized in that: The specific process of step S1 is as follows: where x * is the feature matrix of the dataset after normalization, x is the feature matrix of the dataset before normalization, mean(x) is the mean of the feature matrix of the dataset, std(x) is the standard deviation of the feature matrix of the dataset, and the result of standardization is between (0,1).
3. The method for identifying hand slip signals based on KPCA-ICS-SVM according to claim 1 is characterized in that: The specific process of step S4 is as follows: Using mean square error as the fitness function, the smaller the mean square error, the higher the model accuracy: 。