A fine motion control method for prosthetic hand based on muscle fiber micro perception

Abstract

The application discloses a prosthesis hand fine motion control method based on muscle fiber micro perception, and relates to the technical fields of intelligent prosthesis control, human-computer interaction and medical image processing. The method collects forearm muscle dynamic morphology ultrasonic images through an ultrasonic probe fixed on a forearm of a user, extracts gradient features, entropy features and statistical intensity features after median filtering and size normalization processing, and splices to form a multi-dimensional biometric feature fusion vector; the fusion vector is input into a random forest classifier to obtain a finger fine motion intention category; a finite state machine safety machine is further used for performing time sequence verification on continuous multiple frame classification results and a previous time execution state, and when the verification is passed, a prosthesis hand driving instruction is generated, and when the verification is not passed, a previous state is maintained and the last driving instruction is continuously output, so that the prosthesis hand can execute corresponding finger actions.

Description

Technical Field

[0001] This invention discloses a method for fine motion control of a prosthetic hand based on microscopic perception of muscle fibers, which relates to the fields of intelligent prosthetic control and medical image processing technology. Background Technology

[0002] Prosthetic hands and hand rehabilitation assistive devices are important equipment for assisting patients with upper limb dysfunction to restore or compensate for hand motor abilities. In order for the prosthetic hand to perform corresponding movements according to the user's motor intentions, existing prosthetic hand control methods usually need to first collect the user's physiological signals, voice signals, or body movement signals, and then convert these signals into prosthetic hand control commands through a recognition module.

[0003] Currently, common physiological signal control methods include surface electromyography (EMG) signal control and electroencephalography (EEG) signal control. Surface EMG signal control typically involves collecting electrical signals generated by muscle activity using electrodes attached to the skin of the forearm, and then determining the user's hand gesture intention based on the collected EMG signals. This method has been applied in some gesture recognition tasks, but because the electrodes are located on the skin surface, the acquisition results are easily affected by factors such as skin impedance, sweat, electrode positional deviation, and external noise. Furthermore, crosstalk between electrical signals from different forearm muscles can easily occur, making it difficult to accurately distinguish the muscle activity corresponding to a single finger. Therefore, prosthetic hand control methods based on surface EMG signals still have shortcomings in complex gesture recognition, single-finger fine motor recognition, and long-term stable use.

[0004] Brain-computer interface (BCI) control, which involves collecting signals from the cerebral cortex and converting them into control commands for the prosthetic hand, is a non-invasive method. This approach directly reflects central nervous system activity and is significant for prosthetic hand control. However, ECI signals themselves have a low signal-to-noise ratio and are easily affected by various interference factors. In practical applications, they typically only recognize a limited number of movements such as opening and closing, making it difficult to meet the demands of prosthetic hands for multi-category, fine-grained finger movement control.

[0005] Besides physiological signal control, existing prosthetic hands can also be controlled by voice or movement from other body parts. Voice control receives user voice commands via a microphone and controls the prosthetic hand's movements based on the voice recognition results. This method is suitable for users who cannot generate sufficient electromyographic signals, but its recognition results are easily affected by environmental noise, resulting in poor reliability in noisy environments. Body movement control indirectly triggers the prosthetic hand to perform corresponding movements through shoulder, foot, or other body part movements. However, this method requires users to map non-hand movements to hand movements, resulting in a heavy training burden and difficulty in achieving independent, fine control of individual fingers.

[0006] In summary, existing prosthetic hand control methods generally suffer from the following problems: First, some methods indirectly infer hand movement intentions through surface electromyography, electroencephalography, speech, or body movement signals, making it difficult to directly perceive the microscopic morphological changes of the forearm muscles themselves during finger movements; second, existing methods have a weak ability to distinguish the fine movement intentions of individual fingers, making it difficult to meet the needs of complex gesture recognition and multi-category finger movement control; third, if the recognition results are directly used to drive the prosthetic hand, the prosthetic hand is prone to performing unexpected movements due to transient noise or misclassification, affecting control safety and stability.

[0007] In recent years, muscle ultrasound sensing has provided a new technological approach for fine motor control of prosthetic hands. Ultrasound probes can acquire dynamic grayscale images of the internal tissues of the forearm muscles, directly reflecting the contraction, relaxation, and microscopic deformation of the forearm muscles during finger movements. Compared to surface electromyography (EMG) signals, ultrasound images can more directly characterize changes in the physical morphology of muscles and are less affected by skin impedance, sweat, and electrode displacement. However, how to extract effective features reflecting changes in muscle fiber deformation, muscle texture complexity, and grayscale intensity distribution from continuously acquired forearm muscle ultrasound images, how to accurately identify the user's fine finger movement intentions based on these features, and how to perform safety verification before outputting the identification results to the prosthetic hand remain technical problems that need to be solved to achieve stable fine motor control of the prosthetic hand. Summary of the Invention

[0008] This invention provides a method for fine motor control of a prosthetic hand based on microscopic perception of muscle fibers, in order to solve the problems of existing prosthetic hand control methods which have difficulty directly perceiving the microscopic morphological changes of forearm muscles during finger movements, resulting in weak ability to distinguish fine motor intentions of individual fingers, insufficient stability in recognizing complex gestures, and the tendency to produce unexpected movements when the recognition results directly drive the prosthetic hand.

[0009] In a first aspect, the present invention provides a method for fine motor control of a prosthetic hand based on microscopic sensing of muscle fibers, comprising: S1, based on the muscle movement state of the forearm flexor muscle group during the user's finger movement, uses an ultrasound probe fixed to the user's forearm to collect real-time ultrasound images of the dynamic morphology of the forearm muscles. S2, Based on the ultrasound image of the dynamic morphology of the forearm muscles, preprocessing is performed through median filtering and size normalization to obtain the preprocessed ultrasound image; S3. Based on the preprocessed ultrasound image, extract gradient features that characterize the changes in muscle fiber deformation boundaries, entropy features that quantify the changes in muscle texture complexity, and statistical intensity features that reflect the changes in grayscale distribution of muscle micromorphology. S4. Based on the gradient features, the entropy features, and the statistical intensity features, a multidimensional biometric feature fusion vector is constructed by feature splicing and fusion, wherein the multidimensional biometric feature fusion vector includes gradient mean, gradient standard deviation, global entropy, grayscale mean, and grayscale standard deviation. S5, input the multidimensional biometric feature fusion vector into the random forest classifier, and perform classification and recognition through the random forest classifier to obtain the user's fine finger movement intention category; S6. Based on the classification results of consecutive multi-frames and the execution state of the prosthetic hand at the previous moment, a timing verification is performed using a finite state machine. When the classification results of consecutive multi-frames satisfy the consecutive frame consistency condition, and the state transition from the execution state of the prosthetic hand at the previous moment to the recognition intention at the current moment conforms to the physiologically feasible transition table of hand movement, a prosthetic hand driving instruction is generated. When the classification results of consecutive multi-frames do not satisfy the consecutive frame consistency condition, or the state transition does not conform to the physiologically feasible transition table of hand movement, the execution state of the prosthetic hand at the previous moment is maintained, and the previous driving instruction is continued to be output. S7. According to the prosthetic hand driving command, control the prosthetic hand to perform the corresponding finger movements.

[0010] In some embodiments, S1 includes: fixing a linear array ultrasound probe to the distal end of the user's forearm, directly opposite the flexor muscle group between the radius and ulna, and ensuring that the probe direction of the linear array ultrasound probe is perpendicular to the muscle fiber direction; setting the working frequency band, center working frequency, imaging depth, and acquisition frame rate of the linear array ultrasound probe according to the muscle thickness of the forearm flexor muscle group to be acquired, the penetration depth of the ultrasound probe, and the image resolution requirements, such that the working frequency band includes a high-frequency band for acquiring superficial muscle tissue and a low-frequency band for acquiring deep muscle tissue, and that the imaging depth covers the muscle thickness range of the forearm flexor muscle group; and acquiring dynamic morphological ultrasound images of the forearm muscles using the configured linear array ultrasound probe.

[0011] In some embodiments, S2 includes: using 3×3 median filtering to suppress speckle noise in the ultrasound image of the dynamic morphology of the forearm muscles to obtain a filtered ultrasound image; using bilinear interpolation to scale the filtered ultrasound image to a preset uniform pixel resolution so that different frames of ultrasound images can be used for subsequent feature extraction at the same pixel scale to obtain the preprocessed ultrasound image.

[0012] In some implementations, in S3, extracting the gradient features includes: calculating the horizontal gradient and the vertical gradient based on the preprocessed ultrasound image; calculating the gradient magnitude based on the horizontal gradient and the vertical gradient; and calculating the gradient mean and gradient standard deviation of the entire image based on the gradient magnitude to obtain gradient features used to characterize the changes in the deformation boundary of muscle fibers.

[0013] In some implementations, in S3, extracting the entropy feature and the statistical intensity feature includes: calculating the global entropy based on the grayscale histogram of the preprocessed ultrasound image to obtain the entropy feature used to quantify the changes in muscle texture complexity; and calculating the grayscale mean and grayscale standard deviation based on the grayscale intensity values ​​of each pixel in the preprocessed ultrasound image to obtain the statistical intensity feature used to reflect the changes in grayscale distribution of muscle micromorphology.

[0014] In some implementations, S4 and S5 include: concatenating the gradient mean, the gradient standard deviation, the global entropy, the gray-level mean, and the gray-level standard deviation to obtain the multidimensional biometric feature fusion vector; inputting the multidimensional biometric feature fusion vector into a random forest classifier, classifying the multidimensional biometric feature fusion vector through the random forest classifier, and outputting the category of the user's fine finger movement intention.

[0015] In some implementations, S6 includes: constructing the input state of a finite state machine based on the classification results of consecutive multi-frames and the prosthetic hand execution state at the previous moment; determining that the consecutive frame consistency condition is met when the recognition intent in the classification results of consecutive K frames remains consistent, where K is 3 to 5; determining that the state transition feasibility condition is met when the state transition from the prosthetic hand execution state at the previous moment to the recognition intent at the current moment conforms to the physiologically feasible transition table of hand movement; generating the prosthetic hand driving command when both the consecutive frame consistency condition and the state transition feasibility condition are met; and maintaining the prosthetic hand execution state at the previous moment and continuing to output the previous driving command when either the consecutive frame consistency condition or the state transition feasibility condition is not met. In a second aspect, the present invention provides a prosthetic hand fine motion control device based on muscle fiber microscopic perception, comprising: The ultrasound sensing module is used to acquire real-time ultrasound images of the dynamic morphology of the forearm muscles by using an ultrasound probe fixed to the user's forearm, based on the muscle movement state of the forearm flexor muscles during the user's finger movements. The image preprocessing module is used to preprocess the ultrasound image of the dynamic morphology of the forearm muscles by median filtering and size normalization to obtain the preprocessed ultrasound image. The biometric feature extraction module is used to extract gradient features that characterize changes in the deformation boundary of muscle fibers, entropy features that quantify changes in the complexity of muscle texture, and statistical intensity features that reflect changes in the gray-scale distribution of muscle micromorphology, based on the preprocessed ultrasound image. The feature fusion module is used to construct a multidimensional biometric feature fusion vector by splicing and fusing the gradient features, the entropy features, and the statistical intensity features. The multidimensional biometric feature fusion vector includes the gradient mean, gradient standard deviation, global entropy, grayscale mean, and grayscale standard deviation. The random forest classification module is used to classify and identify the user's fine finger movement intention category based on the multidimensional biometric feature fusion vector. A finite state machine safety module is used to perform timing verification based on the classification results of consecutive multi-frames and the execution state of the prosthetic hand at the previous moment. When the classification results of consecutive multi-frames satisfy the consecutive frame consistency condition, and the state transition from the execution state of the prosthetic hand at the previous moment to the recognition intention at the current moment conforms to the physiologically feasible transition table of hand movement, a prosthetic hand driving instruction is generated. When the classification results of consecutive multi-frames do not satisfy the consecutive frame consistency condition, or the state transition does not conform to the physiologically feasible transition table of hand movement, the execution state of the prosthetic hand at the previous moment is maintained, and the previous driving instruction is continued to be output. A prosthetic hand execution module is used to receive the prosthetic hand driving instruction and parse the target gesture category code, target finger number, movement angle, and execution permission flag bit contained in the prosthetic hand driving instruction. When the execution permission flag bit is at a valid level, the corresponding control channel is determined according to the target finger number, and the movement angle is converted into a driving signal for the corresponding finger mechanism to control the corresponding finger mechanism to perform a contraction or extension action.

[0016] Thirdly, the present invention provides an electronic device, including a memory and a processor, wherein the memory is used to store a computer program, and the computer program, when executed by the processor, implements the above-described method for fine motor control of a prosthetic hand based on microscopic perception of muscle fibers.

[0017] Fourthly, the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-described method for fine motor control of a prosthetic hand based on microscopic perception of muscle fibers.

[0018] Compared with the prior art, the present invention has at least the following beneficial effects: This invention uses an ultrasound probe fixed to the user's forearm to collect dynamic morphological ultrasound images of the forearm flexor muscle group in real time. It can directly perceive the morphological changes of the internal tissues of the forearm muscles during finger movement. Compared with prosthetic hand control methods that rely on surface electromyography signals, it can reduce the impact of skin impedance, sweat, electrode position displacement and muscle crosstalk on movement intention recognition.

[0019] This invention extracts gradient features, entropy features, and statistical intensity features from preprocessed ultrasound images. Gradient features characterize changes in muscle fiber deformation boundaries, entropy features quantify changes in muscle texture complexity, and statistical intensity features reflect changes in the grayscale distribution of muscle microstructure. Therefore, it can characterize the differences in muscle microstructure during finger movement from three dimensions: muscle structure boundaries, texture complexity, and grayscale distribution, improving the ability to distinguish the intent of fine finger movements.

[0020] This invention concatenates gradient mean, gradient standard deviation, global entropy, grayscale mean, and grayscale standard deviation to form a multidimensional biometric feature fusion vector, and uses a random forest classifier for classification and recognition. It can output the user's fine finger movement intention category based on multidimensional complementary features, thereby improving the accuracy and stability of complex gesture recognition.

[0021] This invention performs timing verification based on classification results across multiple consecutive frames and the prosthetic hand's execution state at the previous moment. When both the consecutive frame consistency condition and the state transition feasibility condition are met, a prosthetic hand driving command is generated. If either condition is not met, the prosthetic hand's execution state from the previous moment is maintained, and the previous driving command is output again. This reduces the impact of instantaneous noise, single-frame misclassification, and recognition results that do not meet state transition requirements on the prosthetic hand's movement execution, improving the safety and stability of prosthetic hand control. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0023] Figure 1 This is a flowchart illustrating a method for fine motor control of a prosthetic hand based on microscopic perception of muscle fibers, provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of the installation position and imaging principle of an ultrasonic sensing module provided in an embodiment of the present invention; Figure 3 This is a multi-feature visualization comparison image of a thumb bending gesture provided by an embodiment of the present invention; wherein, (a) is a gesture image, (b) is the original frame of image (a); (c) is the gradient image of image (a); (d) is the entropy heatmap of image (a); and (e) is the feature fusion image of image (a). Figure 4 This is a multi-feature visualization comparison image of the extended index and middle fingers gesture provided in an embodiment of the present invention; wherein (a) is the gesture image, (b) is the original frame of image (a); (c) is the gradient image of image (a); (d) is the entropy heatmap of image (a); and (e) is the feature fusion image of image (a). Figure 5 This is a multi-feature visualization comparison image of a fist gesture provided by an embodiment of the present invention; wherein (a) is a gesture image, (b) is the original frame of image (a); (c) is the gradient image of image (a); (d) is the entropy heatmap of image (a); and (e) is the feature fusion image of image (a). Figure 6 This is a confusion matrix diagram of a random forest classification module on a test set provided in an embodiment of the present invention; Figure 7 This is a schematic diagram of a set of 21 fine gesture actions provided in an embodiment of the present invention; Figure 8 This is a bar chart comparing the recognition accuracy of the method of the present invention with that of CNN and SVM, as provided in an embodiment of the present invention. Figure 9 This is a comparison chart of ablation experiment results provided by an embodiment of the present invention. Detailed Implementation

[0024] 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 embodiments of the present invention, and not all embodiments. 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.

[0025] Example 1 This embodiment provides a method for fine motor control of a prosthetic hand based on microscopic perception of muscle fibers. This method uses ultrasound images of the dynamic morphology of the forearm flexor muscles during finger movements to identify the user's intention for fine finger movements. After successful timing safety verification, it generates prosthetic hand drive commands to control the prosthetic hand to perform corresponding finger actions. The overall process of this method can be found in [link to documentation]. Figure 1 The main functions include forearm muscle dynamic morphology ultrasound image acquisition, image preprocessing, muscle micromorphological feature extraction, multidimensional biometric fusion, random forest classification and recognition, finite state machine safety verification, and prosthetic hand movement execution.

[0026] S1, based on the muscle movement state of the forearm flexor muscles during the user's finger movements, uses an ultrasound probe fixed to the user's forearm to collect real-time dynamic morphological ultrasound images of the forearm muscles.

[0027] Specifically, such as Figure 2 As shown, a linear array ultrasound probe is used to acquire dynamic morphological ultrasound images of the forearm muscles. The linear array ultrasound probe is fixed to the distal end of the user's forearm, directly opposite the flexor muscle group between the radius and ulna, with the probe direction perpendicular to the muscle fiber direction. This fixation method allows the ultrasound probe to capture dynamic morphological changes within the forearm flexor muscle group during the user's finger movements.

[0028] In this embodiment, the parameters of the linear array ultrasound probe are determined based on the muscle thickness of the forearm flexor muscles to be acquired, the penetration depth of the ultrasound probe, and the image resolution requirements. Specifically, the imaging depth is set to cover the muscle thickness range of the forearm flexor muscles; the operating frequency band is set to simultaneously include a high-frequency band for distinguishing superficial muscle tissue and a low-frequency band for penetrating deep muscle tissue; and the acquisition frame rate is set to meet the requirements for acquiring continuous multi-frame classification results and temporal verification.

[0029] In one specific implementation, the linear array ultrasound probe has a probe frequency range of 5MHz to 12MHz, a central operating frequency of 10.5MHz, an imaging depth of 3cm, and an acquisition frame rate of 30Hz. The 12MHz high-frequency band is used to acquire information from the superficial subcutaneous fascia, superficial fascia, and muscle surface, while the 5MHz low-frequency band is used to acquire information from deep muscles. The 10.5MHz central operating frequency, combined with the 3cm imaging depth, ensures that the forearm flexor muscle groups are within the effective penetration and clear resolution range of ultrasound, thereby obtaining dynamic morphological ultrasound images of the forearm muscles that meet the requirements for subsequent feature extraction.

[0030] An ultrasound probe emits high-frequency sound waves into biological tissue and receives reflected echoes from the tissue boundary. The data processing unit reconstructs the echo signals into grayscale ultrasound images, thereby obtaining dynamic morphological ultrasound images of the forearm muscles. These dynamic morphological ultrasound images of the forearm muscles are used to reflect the contraction, relaxation, and microscopic morphological changes of the forearm flexor muscle group during finger movement.

[0031] S2. Based on the ultrasound image of the dynamic morphology of the forearm muscles, preprocessing is performed through median filtering and size normalization to obtain the preprocessed ultrasound image.

[0032] Specifically, the acquired ultrasound images of the dynamic morphology of the forearm muscles are preprocessed to reduce the influence of speckle noise in the ultrasound images and to standardize the image size required for subsequent feature extraction.

[0033] S21, a 3×3 median filter is used to suppress speckle noise in the dynamic morphology ultrasound image of the forearm muscles to obtain the filtered ultrasound image. The calculation method of the median filter is shown in Equation (1).

[0034] (1); In equation (1), This represents the grayscale value of the original ultrasound image at pixel coordinates; This indicates the median-filtered ultrasound image in pixel coordinates. The grayscale value at that location; Represents the pixel coordinates in the horizontal direction, i.e., the column index. Represents the pixel coordinates in the vertical direction, i.e., the row index; and They represent The horizontal and vertical offsets relative to the target pixel within the median filter window, where... ; Indicates the above The median gray value of each pixel within the median filtering window is taken as the median.

[0035] S22, bilinear interpolation is used to scale the filtered ultrasound image to a preset uniform pixel resolution, resulting in a preprocessed ultrasound image. Size normalization enables subsequent feature extraction from different frames of ultrasound images at a uniform pixel scale.

[0036] In this embodiment, the preset uniform pixel resolution is determined based on the scale consistency and computational cost required for subsequent gradient feature, entropy feature, and statistical intensity feature extraction. As a specific implementation, the filtered ultrasound image is scaled to a resolution of 128×128 pixels. This pixel resolution can reduce the computational cost of subsequent feature extraction and classification while preserving muscle texture, muscle fiber boundaries, and grayscale distribution information.

[0037] S3. Based on the preprocessed ultrasound image, extract gradient features that characterize the changes in the deformation boundary of muscle fibers, entropy features that quantify the changes in the complexity of muscle texture, and statistical intensity features that reflect the changes in the gray-scale distribution of muscle micromorphology.

[0038] In this embodiment, gradient features, entropy features, and statistical intensity features characterize the microscopic morphological changes of the forearm flexor muscle group under different finger movement states from three dimensions: muscle fiber deformation boundary, muscle texture complexity, and gray intensity distribution.

[0039] S31. Calculate the horizontal and vertical gradients based on the preprocessed ultrasound image. The horizontal gradient is calculated as shown in Equation (2), and the vertical gradient is calculated as shown in Equation (3).

[0040] ( (2); (3); S32, calculate the gradient magnitude based on the horizontal and vertical gradients. The gradient magnitude is calculated as shown in equation (4).

[0041] (4); S33. Based on the gradient magnitude, the mean and standard deviation of the gradient across the entire image are statistically analyzed to obtain the gradient characteristics used to characterize the changes in the deformation boundary of muscle fibers. The calculation method of the mean gradient across the entire image is shown in Equation (5), and the calculation method of the standard deviation of the gradient is shown in Equation (6).

[0042] (5); (6); In equations (5) and (6), , These represent the number of pixels in the width and height directions of the image, respectively. This represents the total number of pixels in the preprocessed ultrasound image, where , This represents the gradient amplitude. The gradient characteristics described above can characterize the changes in the boundaries of muscle fibers and fascia tissue structures during finger movement.

[0043] S34. Based on the grayscale histogram of the preprocessed ultrasound image, calculate the global entropy to obtain the entropy feature used to quantify the changes in muscle texture complexity. The calculation method of the global entropy is shown in Equation (7), and the calculation method of the probability of gray level occurrence in the grayscale histogram is shown in Equation (8).

[0044] (7); (8); In equations (7) and (8), Where k is the total number of gray levels, and k represents the gray level index. The range of values ​​is to , grayscale The probability of occurrence in preprocessed ultrasound images grayscale The corresponding number of pixels, through entropy features, can characterize the changes in the complexity of forearm muscle texture under different finger movement states.

[0045] S35. Based on the gray intensity values ​​of each pixel in the preprocessed ultrasound image, calculate the gray mean and gray standard deviation to obtain the statistical intensity characteristics that reflect the gray distribution changes in the microscopic morphology of the muscle. The gray mean is calculated as shown in Equation (9), and the gray standard deviation is calculated as shown in Equation (10).

[0046] (9); (10) In equations (9) and (10), This represents the mean gray level of the preprocessed ultrasound image. This represents the standard deviation of the grayscale values ​​in the preprocessed ultrasound image. Indicates the first [item] in the preprocessed ultrasound image grayscale value of each pixel; Indicates the pixel number. The value range is 1 to ; This represents the total number of pixels in the preprocessed ultrasound image. In equation (1)... With equations (9) and (10) Both are used to represent the grayscale value of an image pixel, where Represented using pixel coordinates, Represented by pixel number.

[0047] Statistical intensity characteristics can be used to characterize the changes in the grayscale distribution and tissue reflectivity of ultrasound images during muscle deformation.

[0048] S4 constructs a multidimensional biological feature fusion vector by splicing and fusing features based on gradient features, entropy features, and statistical intensity features.

[0049] Specifically, the gradient mean, gradient standard deviation, global entropy, grayscale mean, and grayscale standard deviation are concatenated to obtain a multidimensional biometric feature fusion vector. The construction method of the multidimensional biometric feature fusion vector is shown in Equation (11).

[0050] (11); Therefore, the multidimensional biometric fusion vector simultaneously includes information on changes in muscle fiber deformation boundaries, changes in muscle texture complexity, and changes in grayscale distribution of muscle micromorphology, which is used to provide input data for subsequent recognition of the user's fine finger movement intentions.

[0051] S5 inputs the multi-dimensional biometric feature fusion vector into the random forest classifier, and performs classification and recognition through the random forest classifier to obtain the user's fine finger movement intention category.

[0052] Specifically, the random forest classifier uses a multi-dimensional biometric feature fusion vector as input to classify the corresponding microscopic morphological changes in muscles during different finger movements, outputting the user's fine finger movement intention category. This fine finger movement intention category can correspond to... Figure 7 The example shown is a set of fine hand gestures. By inputting a multidimensional biometric fusion vector into a random forest classifier, the system can identify the user's current fine finger movement intentions based on information such as muscle fiber deformation, texture complexity, and grayscale distribution changes contained in ultrasound images of the dynamic morphology of forearm muscles.

[0053] S6. Based on the classification results of multiple consecutive frames and the execution state of the prosthetic hand at the previous moment, perform timing verification through a finite state machine; when both the consistency condition of consecutive frames and the feasibility condition of state transition are met, generate the prosthetic hand driving instruction; when either condition is not met, maintain the execution state of the prosthetic hand at the previous moment and continue to output the previous driving instruction.

[0054] Specifically, the recognition result output by the random forest classifier at the current time step is first used as the input state of the finite state machine for temporal verification. The finite state machine receives classification results from multiple consecutive frames and the prosthetic hand execution state from the previous time step.

[0055] When the recognition intent in the classification results of K consecutive frames remains consistent, the consecutive frame consistency condition is satisfied, where K is 3 to 5.

[0056] The hand movement physiological feasible transition table is used to record whether direct state transitions are allowed between different prosthetic hand execution states. The table uses the previous execution state of the prosthetic hand as the starting state and the finger fine motor intention category currently identified by the random forest classifier as the target state, recording the allowed transition relationships from each starting state to each target state. The finite state machine queries the hand movement physiological feasible transition table based on the previous prosthetic hand execution state and the currently identified intention; when the query result indicates a allowed transition, the state transition feasibility condition is satisfied.

[0057] For example, if the current state of the prosthetic hand is "index and middle fingers extended", and the classification result for multiple consecutive frames is "fist clenched", if the index finger mechanism, middle finger mechanism, and other finger mechanisms of the prosthetic hand can continuously move from the current posture to the flexion position corresponding to the fist clenching, then the state transition from "index and middle fingers extended" to "fist clenched" is recorded as an allowed transition in the physiologically feasible transition table of hand movement. The finite state machine generates the prosthetic hand driving instructions corresponding to the fist clenching action when the consistency condition of consecutive frames is simultaneously met.

[0058] For state transitions that require multiple finger mechanisms to change their direction of movement simultaneously within a short period of time, and where direct transitions are likely to cause significant abrupt changes in posture, these transitions can be recorded as disallowed in the hand movement physiologically feasible transition table. For example, if the prosthetic hand's current execution state is "clenched fist" and the current recognition intent is "thumb flexion," and a direct transition from the clenched fist state to the flexed thumb state requires the index, middle, ring, and little finger mechanisms to simultaneously transition from flexion to extension, while the thumb mechanism remains in or adjusts to the flexed position, then this state transition is recorded as disallowed. In this case, even if the current recognition result already meets the consecutive frame consistency condition, the finite state machine retains the previous prosthetic hand execution state and does not output a new driving instruction corresponding to "thumb flexion." Only after the prosthetic hand enters a hand posture that is allowed as an intermediate state will the next state transition judgment be made based on the classification results of subsequent consecutive frames.

[0059] When both the consecutive frame consistency condition and the state transition feasibility condition are met, the finite state machine generates the corresponding prosthetic hand driving command. When either condition is not met, the finite state machine maintains the prosthetic hand execution state from the previous moment and continues to output the previous driving command. Through the above processing, the impact of transient noise in the ultrasound imaging process, misclassification by the random forest classifier, and recognition results that do not meet the state transition requirements on the execution of prosthetic hand movements can be reduced.

[0060] S7 controls the prosthetic hand to perform corresponding finger movements according to the prosthetic hand drive command.

[0061] Specifically, after the finite state machine safety machine verification is passed, the generated prosthetic hand driving instructions are transmitted to the prosthetic hand execution module. The prosthetic hand driving instructions include the target gesture category code, the target finger number, the movement angle, and the execution permission flag.

[0062] The prosthetic hand execution module includes a microprocessor and multiple finger mechanism drive motors electrically connected to the microprocessor. After receiving the prosthetic hand drive command, the prosthetic hand execution module uses the microprocessor to parse the target gesture category code, target finger number, movement angle, and execution permission flag.

[0063] When the execution enable flag is at a valid level, the microprocessor retrieves the corresponding control channel according to the target finger number and converts the motion angle into a drive signal for the corresponding target finger mechanism drive motor to drive the corresponding finger mechanism to contract or extend, thereby completing the corresponding finger action.

[0064] The method described in this embodiment can directly acquire dynamic morphological ultrasound images of the forearm flexor muscle groups based on an ultrasound probe fixed to the user's forearm. Gradient features, entropy features, and statistical intensity features are extracted from the preprocessed ultrasound images to construct a multidimensional biometric fusion vector. Then, a random forest classifier is used to identify the user's fine finger movement intention category. Finally, a finite state machine is used to perform temporal verification of the classification results of multiple consecutive frames and the execution state of the prosthetic hand at the previous moment, and the prosthetic hand is controlled to perform the corresponding finger movements.

[0065] Example 2 This embodiment illustrates the verification process and results of the fine motor control method for prosthetic hand based on microscopic perception of muscle fibers described in Embodiment 1 in the fine gesture recognition task.

[0066] This embodiment analyzes the performance of the recognition system on a self-constructed dataset of 21 fine hand gestures using ultrasound images. The set of 21 fine hand gestures can be found in [reference needed]. Figure 7The ultrasound images in this dataset were acquired by a linear array ultrasound probe fixed to the distal end of the user's forearm, facing the flexor muscle group between the radius and ulna, and were used to characterize the dynamic morphological changes of the internal tissues of the forearm muscles under different finger movement states.

[0067] Each frame of dynamic morphological ultrasound image of forearm muscles in the dataset is processed according to steps S2 to S4 in Example 1 to obtain a multidimensional biometric fusion vector composed of gradient mean, gradient standard deviation, global entropy, gray mean and gray standard deviation. The category of fine hand gesture performed by the user when the ultrasound image frame was acquired is used as the corresponding category label, thereby forming labeled samples for training and testing of the random forest classifier.

[0068] The random forest classifier consists of multiple decision trees. During training each decision tree, samples with replacement are taken from the training data to form a training subset for that decision tree. During node splitting in the decision tree, a subset of features is randomly selected from the features contained in the multidimensional biometric fusion vector as candidate features. The features used for node splitting and the splitting threshold are determined based on the Gini impurity corresponding to the candidate features. After each decision tree is trained, the multidimensional biometric fusion vector to be identified is input into each decision tree, and each decision tree outputs its corresponding gesture category. The random forest classifier selects the gesture category with the most votes as the user's fine motor intention category.

[0069] This embodiment is for Figure 7 For each of the 21 fine gesture movements shown, 200 valid samples were collected. Each valid sample corresponds to a frame of ultrasound image showing the dynamic morphology of the forearm muscles with the movement category labeled. Each valid sample was processed according to steps S2 to S4 to obtain a multi-dimensional biometric fusion vector composed of gradient mean, gradient standard deviation, global entropy, gray-level mean, and gray-level standard deviation. The fine gesture movement category corresponding to the sample was used as the category label. Of the 200 valid samples for each fine gesture movement, 160 samples were used to train a random forest classifier, and 40 samples were used to test the classification results of the random forest classifier, thus forming a training sample set and a test sample set containing the 21 fine gesture movement categories.

[0070] This embodiment uses a random forest classifier comprising 100 decision trees. During training of each decision tree, a training subset corresponding to that decision tree is formed by sampling with replacement from the training sample set. When splitting at each node, two features are randomly selected as candidate features from five input features: gradient mean, gradient standard deviation, global entropy, gray-level mean, and gray-level standard deviation. The Gini impurity of each candidate feature is calculated at different splitting thresholds. The feature that maximizes the reduction in Gini impurity after splitting, along with its corresponding splitting threshold, is selected to complete the node split. After training of each decision tree, the multi-dimensional biometric feature fusion vector corresponding to the sample to be identified is input into each of the 100 decision trees. The gesture category receiving the most votes is determined as the fine motor intention category of the finger corresponding to that sample.

[0071] To visually demonstrate the ability of gradient features, entropy features, statistical intensity features, and multi-feature fusion schemes to represent the characteristics of hand gesture muscle deformation, this embodiment selects three typical hand gestures—thumb bending, index and middle finger extension, and fist clenching—for visualization verification. In the visualization verification, to preserve image spatial information, the preprocessed ultrasound image is divided into non-overlapping blocks, and the local standard deviation of each block is calculated to generate a statistical heatmap, which is used to characterize the local grayscale fluctuations and tissue reflectivity changes caused by muscle deformation. Figure 3 A multi-feature visualization comparison of the thumb bending gesture. Figure 4 A multi-feature visualization comparison of the index and middle finger extended hand gestures. Figure 5 This is a multi-feature visualization comparison of the fist-clenching gesture. The non-overlapping blocks and statistical heatmaps described above are used to visually demonstrate the differences in muscle morphology changes corresponding to different gestures. They are not used as the actual classification input for the random forest classifier, which is the multi-dimensional biofeedback vector obtained in step S4.

[0072] like Figures 3 to 5As shown, each type of gesture includes the original ultrasound image, gradient feature map, texture entropy feature map, and multi-feature fusion map. The original ultrasound image is used to display the basic tissue structure and soft tissue distribution of the forearm muscles during hand movements, providing a reference for muscle deformation analysis. The gradient feature map is used to reflect the local gray-level changes in the muscle region in the ultrasound image and to highlight the boundaries of muscle fiber and fascia tissue structures; the gradient edge distribution and deformation features corresponding to different gestures are different, which can be used to distinguish the muscle activation structures in different finger movement states. The texture entropy feature map is used to reflect the spatial complexity of muscle texture and the intensity of local muscle activation deformation. Among them, areas with higher texture complexity correspond to areas with stronger muscle activation, and areas with more uniform texture correspond to areas with weaker deformation. The multi-feature fusion map is used to simultaneously present local gray-level changes, texture complexity, and muscle fiber structure boundaries and deformation trajectories, so that different hand gestures form distinguishable spatial features in the dimensions of gradient structure distribution, entropy texture distribution, and gray-level statistical changes.

[0073] Depend on Figures 3 to 5 It can be seen that the muscle regions corresponding to different gestures exhibit different characteristics in terms of structural boundaries, texture complexity, and grayscale distribution. This demonstrates that the gradient features, entropy features, and statistical intensity features extracted in Example 1 can characterize the microscopic morphological changes of the forearm flexor muscle group during finger movement from different dimensions, and the three types of features are complementary.

[0074] Furthermore, this embodiment employs a random forest classifier to classify the multidimensional biometric feature fusion vector and obtains the classification results on the test set. The confusion matrix of the random forest classifier on the test set is as follows: Figure 6 As shown. From Figure 6 It can be seen that most samples were correctly classified into the corresponding gesture categories, indicating that the method described in Example 1 can effectively distinguish different fine gestures.

[0075] To verify the recognition performance of the method of this invention compared with other classification methods, this embodiment compares the recognition accuracy of the method described in Embodiment 1 with that of convolutional neural networks and support vector machines. The comparison results are as follows: Figure 8 As shown. By Figure 8 It can be seen that the multidimensional biometric fusion method based on the microscopic perception of muscle fibers has a good recognition effect in the fine gesture recognition task.

[0076] To further analyze the roles of gradient features, entropy features, and statistical intensity features in gesture recognition, this embodiment conducted an ablation experiment. The experimental results are as follows: Figure 9 As shown. By Figure 9It can be seen that the classification accuracy of the single entropy feature is 50.00%, indicating that it is difficult to stably distinguish various muscle deformation patterns by relying solely on texture complexity; the classification accuracy of the single gradient feature is 80.95%, indicating that gradient features can effectively extract structural boundaries and local spatial change information in ultrasound images; and the classification accuracy of the single statistical intensity feature is 93.81%, indicating that the global grayscale distribution contains relatively stable gesture classification information.

[0077] With the fusion of multiple features, the recognition performance is further improved. Specifically, the accuracy of the combination of statistical intensity features and entropy features reaches 98.00%, indicating that global grayscale distribution and local texture complexity can complement each other; the accuracy of the combination of gradient features and statistical intensity features is 97.62%, indicating that structural features and grayscale statistical features can jointly characterize muscle activity; the accuracy of adding entropy features to the gradient feature combination is 94.76%, indicating some information overlap in some feature combinations; and the highest accuracy of 98.10% is achieved after fusing all features. This demonstrates that the gradient feature, entropy feature, and statistical intensity feature fusion method used in Example 1 can integrate different dimensions of muscle micromorphological information, improving the accuracy of recognizing the user's fine finger movement intentions.

[0078] The above verification shows that the method described in Example 1 can extract multi-dimensional biometric fusion vectors based on ultrasound images of the dynamic morphology of forearm muscles, and use a random forest classifier to identify the user's fine finger movement intentions, thereby providing a classification result basis for subsequent finite state machine safety verification and prosthetic hand movement execution.

[0079] Example 3 This embodiment provides a fine motor control device for a prosthetic hand based on microscopic sensing of muscle fibers. This device can be used to implement the fine motor control method for a prosthetic hand based on microscopic sensing of muscle fibers described in Embodiment 1.

[0080] The device includes an ultrasonic sensing module, an image preprocessing module, a biometric extraction module, a feature fusion module, a random forest classification module, a finite state machine safety module, and a prosthetic hand execution module.

[0081] The ultrasound sensing module is used to acquire real-time dynamic morphological ultrasound images of the forearm muscles based on the muscle movement state of the forearm flexor muscles during the user's finger movements, using an ultrasound probe fixed to the user's forearm. Specifically, the ultrasound sensing module includes a linear array ultrasound probe and a data processing unit. The linear array ultrasound probe is fixed to the distal end of the user's forearm, directly opposite the flexor muscles between the radius and ulna, with the probe direction perpendicular to the muscle fiber direction. The linear array ultrasound probe acquires echo signals from the forearm muscle tissue during finger movements, and the data processing unit reconstructs the echo signals into grayscale ultrasound images, obtaining dynamic morphological ultrasound images of the forearm muscles.

[0082] Specifically, the ultrasound sensing module sets the operating frequency band, center operating frequency, imaging depth, and acquisition frame rate of the linear array ultrasound probe based on the muscle thickness of the forearm flexor muscles to be acquired, the penetration depth of the ultrasound probe, and the image resolution requirements. This ensures that the operating frequency band includes a high-frequency band for acquiring superficial muscle tissue and a low-frequency band for acquiring deep muscle tissue, and that the imaging depth covers the range of muscle thickness of the forearm flexor muscles. As a specific embodiment, the linear array ultrasound probe has a probe frequency range of 5MHz to 12MHz, a center operating frequency of 10.5MHz, an imaging depth of 3cm, and an acquisition frame rate of 30Hz.

[0083] The image preprocessing module is used to preprocess the forearm muscle dynamic morphology ultrasound image through median filtering and size normalization to obtain a preprocessed ultrasound image. Specifically, the image preprocessing module uses 3×3 median filtering to suppress speckle noise in the forearm muscle dynamic morphology ultrasound image, and uses bilinear interpolation to scale the filtered ultrasound image to a preset uniform pixel resolution, so that different frames of ultrasound images can be extracted for subsequent features at the same pixel scale. As a specific implementation, the preset uniform pixel resolution is 128×128 pixel resolution.

[0084] The biometric extraction module is used to extract gradient features characterizing changes in muscle fiber deformation boundaries, entropy features quantifying changes in muscle texture complexity, and statistical intensity features reflecting changes in grayscale distribution of muscle microstructure based on the preprocessed ultrasound image. Specifically, the biometric extraction module calculates the horizontal and vertical gradients based on the preprocessed ultrasound image, calculates the gradient magnitude based on the horizontal and vertical gradients, and then calculates the gradient mean and standard deviation of the entire image based on the gradient magnitude to obtain gradient features. The biometric extraction module also calculates the global entropy based on the grayscale histogram of the preprocessed ultrasound image to obtain entropy features; and calculates the grayscale mean and standard deviation of each pixel in the entire preprocessed ultrasound image to obtain statistical intensity features.

[0085] The feature fusion module is used to construct a multi-dimensional biometric feature fusion vector by concatenating and fusing gradient features, entropy features, and statistical intensity features. This multi-dimensional biometric feature fusion vector includes the gradient mean, gradient standard deviation, global entropy, gray-level mean, and gray-level standard deviation. Through the feature fusion module, information on muscle fiber deformation boundary changes, muscle texture complexity changes, and muscle micro-morphological gray-level distribution changes can be combined into a single feature vector.

[0086] The random forest classification module is used to classify and identify the user's fine finger movement intention category based on the multidimensional biometric feature fusion vector. Specifically, the random forest classification module receives the multidimensional biometric feature fusion vector output by the feature fusion module, and outputs the corresponding fine finger movement intention category based on the multidimensional biometric feature fusion vector.

[0087] The finite state machine safety module is used to perform timing verification based on the classification results of consecutive frames and the previous prosthetic hand execution state. Specifically, when the recognition intent in the classification results of consecutive K frames remains consistent, the consecutive frame consistency condition is satisfied, where K is 3 to 5; when the state transition from the previous prosthetic hand execution state to the current recognition intent conforms to the physiologically feasible transition table of hand movement, the state transition feasibility condition is satisfied. When both the consecutive frame consistency condition and the state transition feasibility condition are met, the finite state machine safety module generates prosthetic hand driving instructions; when either condition is not met, the finite state machine safety module maintains the previous prosthetic hand execution state and continues to output the previous driving instruction. The prosthetic hand execution module is used to control the prosthetic hand to perform corresponding finger movements according to the prosthetic hand driving instructions. Specifically, the prosthetic hand execution module includes a microprocessor and multiple finger mechanism drive motors electrically connected to the microprocessor. After receiving the prosthetic hand drive instruction output by the finite state machine safety module, the prosthetic hand execution module is then analyzed by the microprocessor, which parses the target gesture category code, target finger number, motion angle, and execution permission flag contained in the prosthetic hand drive instruction.

[0088] When the execution enable flag is at a valid level, the microprocessor retrieves the corresponding control channel according to the target finger number and converts the motion angle into a drive signal for the corresponding target finger mechanism drive motor to drive the corresponding finger mechanism to contract or extend, thereby completing the corresponding finger action.

[0089] When the finite state machine safety module fails the verification, the prosthetic hand execution module continues to receive the previous driving instruction and maintains the prosthetic hand execution state at the previous moment.

[0090] In this embodiment, the ultrasound sensing module outputs a dynamic ultrasound image of the forearm muscles to the image preprocessing module; the image preprocessing module outputs the preprocessed ultrasound image to the biometric extraction module; the biometric extraction module outputs gradient features, entropy features, and statistical intensity features to the feature fusion module; the feature fusion module outputs a multi-dimensional biometric fusion vector to the random forest classification module; the random forest classification module outputs the classification result to the finite state machine safety module; the finite state machine safety module outputs a prosthetic hand driving command to the prosthetic hand execution module when the timing verification passes, and continues to output the previous driving command when the timing verification fails, so that the prosthetic hand execution module maintains the prosthetic hand execution state at the previous moment. Thus, a complete control chain is formed from the acquisition of the dynamic ultrasound image of the forearm muscles to the execution of the prosthetic hand movements.

[0091] Example 4 This embodiment provides an electronic device and a computer-readable storage medium.

[0092] The electronic device includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program. When the computer program is executed by the processor, it implements the fine motor control method for the prosthetic hand based on microscopic perception of muscle fibers as described in Embodiment 1.

[0093] Specifically, when the processor executes the computer program, it can acquire real-time dynamic morphological ultrasound images of the forearm muscles using an ultrasound probe fixed to the user's forearm, based on the muscle movement state of the forearm flexor muscles during the user's finger movements. Based on the dynamic morphological ultrasound images of the forearm muscles, it performs preprocessing through median filtering and size normalization to obtain preprocessed ultrasound images. Based on the preprocessed ultrasound images, it extracts gradient features characterizing changes in muscle fiber deformation boundaries, entropy features quantifying changes in muscle texture complexity, and statistical intensity features reflecting changes in the grayscale distribution of muscle micromorphology. Based on the gradient features, entropy features, and statistical intensity features, it constructs a multidimensional biometric fusion vector through feature splicing and fusion. The multidimensional biometric fusion vector is input into a random forest classification module for classification and recognition to obtain the user's fine motor intention category. Based on the classification results of multiple consecutive frames and the prosthetic hand execution state at the previous moment, it performs temporal verification using a finite state machine. When both the consecutive frame consistency condition and the state transition feasibility condition are met, a prosthetic hand driving instruction is generated; when either condition is not met, the prosthetic hand execution state at the previous moment is maintained, and the previous driving instruction is output.

[0094] The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the fine motor control method for prosthetic hand based on muscle fiber microscopic perception as described in Embodiment 1.

[0095] The electronic device and computer-readable storage medium provided in this embodiment enable the acquisition of dynamic morphological ultrasound images of forearm muscles, image preprocessing, extraction of muscle micromorphological features, fusion of multidimensional biometric features, random forest classification and recognition, finite state machine safety verification, and control of prosthetic hand movements in the method described in Embodiment 1.

[0096] In the several embodiments provided in this invention, it should be understood that the disclosed systems and methods can also be implemented in other ways. The system and method embodiments described above are merely illustrative.

[0097] It should be noted that, in this document, the terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. The terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0098] While the embodiments disclosed in this invention are as described above, the content is merely for the purpose of facilitating understanding of the invention and is not intended to limit the invention. Any person skilled in the art to which this invention pertains may make any modifications and changes in form and detail of the implementation without departing from the spirit and scope disclosed herein; however, the scope of patent protection of this invention shall still be determined by the scope defined in the appended claims.

Claims

1. A method for fine motion control of a prosthetic hand based on muscle fiber microperception, characterized by, include: S1, based on the muscle movement state of the forearm flexor muscle group during the user's finger movement, uses an ultrasound probe fixed to the user's forearm to collect real-time ultrasound images of the dynamic morphology of the forearm muscles. S2, Based on the ultrasound image of the dynamic morphology of the forearm muscles, preprocessing is performed through median filtering and size normalization to obtain the preprocessed ultrasound image; S3. Based on the preprocessed ultrasound image, extract gradient features that characterize the changes in muscle fiber deformation boundaries, entropy features that quantify the changes in muscle texture complexity, and statistical intensity features that reflect the changes in grayscale distribution of muscle micromorphology. S4. Based on the gradient features, the entropy features, and the statistical intensity features, a multidimensional biometric feature fusion vector is constructed by feature splicing and fusion, wherein the multidimensional biometric feature fusion vector includes gradient mean, gradient standard deviation, global entropy, grayscale mean, and grayscale standard deviation. S5, input the multidimensional biometric feature fusion vector into the random forest classifier, and perform classification and recognition through the random forest classifier to obtain the user's fine finger movement intention category; S6. Based on the classification results of consecutive multi-frames and the execution state of the prosthetic hand at the previous moment, a timing verification is performed using a finite state machine. When the classification results of consecutive multi-frames satisfy the consecutive frame consistency condition, and the state transition from the execution state of the prosthetic hand at the previous moment to the recognition intention at the current moment conforms to the physiologically feasible transition table of hand movement, a prosthetic hand driving instruction is generated. When the classification results of consecutive multi-frames do not satisfy the consecutive frame consistency condition, or the state transition does not conform to the physiologically feasible transition table of hand movement, the execution state of the prosthetic hand at the previous moment is maintained, and the previous driving instruction is continued to be output. S7. According to the prosthetic hand driving command, control the prosthetic hand to perform the corresponding finger movements.

2. The method for fine motion control of prosthetic hand based on muscle fiber microperception according to claim 1, characterized in that, S1 includes: S11, Fix the linear array ultrasound probe to the distal end of the user's forearm, facing the flexor muscle group between the radius and ulna, and make the probe direction of the linear array ultrasound probe perpendicular to the direction of the muscle fibers. S12, based on the muscle thickness of the forearm flexor muscle group to be acquired, the penetration depth of the ultrasound probe, and the image resolution requirements, set the working frequency band, center working frequency, imaging depth, and acquisition frame rate of the linear array ultrasound probe, so that the working frequency band includes a high-frequency band for acquiring superficial muscle tissue and a low-frequency band for acquiring deep muscle tissue, and the imaging depth covers the muscle thickness range of the forearm flexor muscle group. S13, the dynamic morphological ultrasound images of the forearm muscles are acquired using the configured linear array ultrasound probe.

3. The method for fine motion control of prosthetic hand based on muscle fiber micro-perception according to claim 1, characterized in that S2 include: S21, use 3×3 median filtering to suppress speckle noise in the dynamic morphological ultrasound image of the forearm muscles to obtain the filtered ultrasound image. S22, the filtered ultrasound image is scaled to a preset uniform pixel resolution using bilinear interpolation, so that different frames of ultrasound images are subjected to subsequent feature extraction at the same pixel scale, thus obtaining the preprocessed ultrasound image.

4. The method for fine motor control of a prosthetic hand based on microscopic perception of muscle fibers according to claim 1, characterized in that, In S3, extracting the gradient features includes: S31, Calculate the horizontal gradient and vertical gradient based on the preprocessed ultrasound image; S32, Calculate the gradient magnitude based on the horizontal gradient and the vertical gradient; S33. Based on the gradient magnitude, the gradient mean and gradient standard deviation of the entire image are statistically analyzed to obtain gradient features used to characterize the changes in the deformation boundary of muscle fibers.

5. The method for fine motor control of a prosthetic hand based on microscopic perception of muscle fibers according to claim 4, characterized in that, In S3, extracting the entropy feature and the statistical intensity feature includes: S34. Based on the grayscale histogram of the preprocessed ultrasound image, calculate the global entropy to obtain the entropy feature used to quantify the changes in muscle texture complexity. S35. Based on the gray intensity values ​​of each pixel in the preprocessed ultrasound image, calculate the gray mean and gray standard deviation to obtain statistical intensity features that reflect the gray distribution changes in the microscopic morphology of the muscle.

6. The method for fine motor control of a prosthetic hand based on microscopic perception of muscle fibers according to claim 5, characterized in that, S4 and S5 include: S41, the gradient mean, the gradient standard deviation, the global entropy, the gray-scale mean, and the gray-scale standard deviation are concatenated to obtain the multidimensional biometric feature fusion vector; S51, the multidimensional biometric feature fusion vector is input into a random forest classifier, and the random forest classifier classifies the multidimensional biometric feature fusion vector to output the category of the user's fine finger movement intention.

7. The method for fine motor control of a prosthetic hand based on microscopic perception of muscle fibers according to claim 6, characterized in that, S6 include: S61, Based on the classification results of multiple consecutive frames and the execution state of the prosthetic hand in the previous moment, construct the input state of the finite state machine; S62, when the recognition intent in the classification results of K consecutive frames remains consistent, it is determined that the consecutive frame consistency condition is met, where K is 3 to 5; S63, when the state transition from the previous prosthetic hand execution state to the current intention recognition state conforms to the hand movement physiological feasible transition table, it is determined that the state transition feasibility condition is met. S64, when both the continuous frame consistency condition and the state transition feasibility condition are met, the prosthetic hand driving command is generated; S65, when either the continuous frame consistency condition or the state transition feasibility condition is not met, maintain the previous prosthetic hand execution state and continue to output the previous drive command.

8. A fine motor control device for a prosthetic hand based on microscopic sensing of muscle fibers, characterized in that, include: The ultrasound sensing module is used to acquire real-time ultrasound images of the dynamic morphology of the forearm muscles by using an ultrasound probe fixed to the user's forearm, based on the muscle movement state of the forearm flexor muscles during the user's finger movements. The image preprocessing module is used to preprocess the ultrasound image of the dynamic morphology of the forearm muscles by median filtering and size normalization to obtain the preprocessed ultrasound image. The biometric feature extraction module is used to extract gradient features that characterize changes in the deformation boundary of muscle fibers, entropy features that quantify changes in the complexity of muscle texture, and statistical intensity features that reflect changes in the gray-scale distribution of muscle micromorphology, based on the preprocessed ultrasound image. The feature fusion module is used to construct a multidimensional biometric feature fusion vector by splicing and fusing the gradient features, the entropy features, and the statistical intensity features. The multidimensional biometric feature fusion vector includes the gradient mean, gradient standard deviation, global entropy, grayscale mean, and grayscale standard deviation. The random forest classification module is used to classify and identify the user's fine finger movement intention category based on the multidimensional biometric feature fusion vector. A finite state machine safety module is used to perform timing verification based on the classification results of consecutive multi-frames and the execution state of the prosthetic hand at the previous moment. When the classification results of consecutive multi-frames satisfy the consecutive frame consistency condition, and the state transition from the execution state of the prosthetic hand at the previous moment to the recognition intention at the current moment conforms to the physiologically feasible transition table of hand movement, a prosthetic hand driving instruction is generated. When the classification results of consecutive multi-frames do not satisfy the consecutive frame consistency condition, or the state transition does not conform to the physiologically feasible transition table of hand movement, the execution state of the prosthetic hand at the previous moment is maintained, and the previous driving instruction is continued to be output. The prosthetic hand execution module is used to receive the prosthetic hand driving command and parse the target gesture category code, target finger number, motion angle and execution permission flag bit contained in the prosthetic hand driving command; when the execution permission flag bit is at an active level, the module determines the corresponding control channel according to the target finger number and converts the motion angle into a driving signal for the corresponding finger mechanism to control the corresponding finger mechanism to perform a contraction or extension action.

9. An electronic device, characterized in that, It includes a memory and a processor, the memory being used to store a computer program, which, when executed by the processor, implements the fine motor control method for a prosthetic hand based on muscle fiber microsensing as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the fine motor control method for prosthetic hand based on muscle fiber micro-sensing as described in any one of claims 1 to 7.

Images