CN109924981A - A kind of talipes cavus detection system and method - Google Patents
A kind of talipes cavus detection system and method Download PDFInfo
- Publication number
- CN109924981A CN109924981A CN201910328152.7A CN201910328152A CN109924981A CN 109924981 A CN109924981 A CN 109924981A CN 201910328152 A CN201910328152 A CN 201910328152A CN 109924981 A CN109924981 A CN 109924981A
- Authority
- CN
- China
- Prior art keywords
- data
- circuit
- layer
- talipes cavus
- convolutional
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Abstract
The invention discloses a kind of talipes cavus detection system and method, which includes that Intelligent insole is connected with host computer data processor;Intelligent insole with host computer data processor by connecting;Intelligent insole includes flexible printed circuit board, interface circuit, ADC conversion circuit and wireless microcontroller;It is provided with 9 force snesors on flexible printed circuit board passes through interface circuit and connect with ADC conversion circuit, wireless microcontroller is connect with host computer data processor.The present invention is by acquisition inertia and pressure data, while the multi-modal sensing data recorded combines the details that can preferably reflect talipes cavus, can preferably identify talipes cavus;The corresponding gait feature of gait data of acquisition is extracted by 1D convolutional neural networks, it is proved to be the suitable selection of processing pressure and inertial sensor data, in the application of foot type classification, it can reach preferable differentiation performance, the combination of Intelligent insole and 1D neural network can be used for the screening of talipes cavus and be detected.
Description
Technical field
The invention belongs to Design for Medical Device technical fields, and in particular to a kind of talipes cavus detection system and method.
Background technique
Compared with normal foot, talipes cavus patient's arch of foot is higher.Investigation display, the people of about 8%-15% is with height in the world
Bow foot.Talipes cavus will lead to neuromuscular problem, including instability of gait is fixed, foot pain and sprain of ankle joint;To talipes cavus morning
The understanding of phase and appropriate assessment are that Case management is successfully crucial, and up to the present, foot morphological parameters one kind is used for foot class
The classification of type, and visual non-quantitation detection, anthropological measuring, footprint parameter measurement and imaging evaluation technology is mostly used to obtain.So
And these methods need podiatrist to have experience abundant, could obtain by training repeatedly.
Data acquisition and foot function can be kept analysis automated, then according to fortune by designing coherent detection instrument at present
Dynamic learn evaluates foot characteristic with kinetic parameter, and existing instrument includes being distributed for obtaining the pressure of kinetic parameter
Measuring system and 3D pressure plare, and inertial sensor or optically-captured system for obtaining dynamic parameter;However, passing through this
A little systems check and take a long time that many instruments are not particularly suited for periodically evaluating, or due to expense it is high and can not be for
General public uses.It is desirable that portable measurement apparatus is introduced, to support the assessment of foot type.
In gait cycle, each foot type is all associated with specific plantar nervous arch mode;Wherein, flatfoot and outer
Sufficient vertical ground reaction forces peak value and footmark area is turned over to have differences in stance;In addition, the peak value of 5 articulationes metatarsophalangeaes
There is also differences between normal foot and talipes cavus for the minimum speed of Center of Pressure when pressure and terminal position phase;It is maximum under arch of foot
There is also differences between normal foot and talipes cavus for integral and its normalized value when integral, the 5th plantar toe push when power, thumb are exerted oneself
It is different;For sufficient type classification automatic method, at present mainly using artificial neural network, Adaptive Neuro-fuzzy Inference and
K-means algorithm.However capture inertial data and/or single-point force snesor are able to use there is presently no a kind of classifier to have
Effect identification talipes cavus.
Summary of the invention
For above-mentioned deficiency in the prior art, talipes cavus detection system provided by the invention solves existing classifier
In be not available capture inertial data and/or single-point force snesor carrys out effectively the problem of identifying talipes cavus.
In order to achieve the above object of the invention, the technical solution adopted by the present invention are as follows: a kind of talipes cavus detection system, including phase
The Intelligent insole and host computer data processor to connect;
The Intelligent insole includes flexible printed circuit board, interface circuit, ADC conversion circuit and wireless microcontroller;
9 force snesors are provided on the flexible printed circuit board, each force snesor passes through interface circuit
It is connect with ADC conversion circuit, the ADC conversion circuit and interface circuit are connect with wireless microcontroller, the wireless micro-control
Device processed is connect with host computer data processor;
The host computer data processor is the computer detected by 1D neural network to input data.
Further, the flexible printed circuit board is insole shape, and 9 force snesors are respectively arranged at flexible print
On printed circuit board at big toe corresponding with foot bottom, at five caput metatarsales, mesopodium outside, at medial heel and lateral heel
Place;
The flexible printed circuit board include the first flexible polyvinyl chloride plate and the second flexible polyvinyl chloride plate, described first
Flexible polyvinyl chloride plate is set to above the second flexible polyvinyl chloride plate;
The first flexible polyvinyl chloride plate is the soft flexible polyvinyl chloride panel with a thickness of 0.8mm;
The second flexible polyvinyl chloride plate is the rigid flexible polyvinyl chloride panel with a thickness of 0.5mm.
Further, wireless microcontroller includes model nRF52832 main control chip and its peripheral circuit;
The peripheral circuit of the main control chip includes pierce circuit, power circuit, trans-impedance amplifier circuit, program downloading
Circuit, environmental sensor circuit and inertial sensor circuit;
The power circuit also with trans-impedance amplifier circuit, program download circuit, environmental sensor circuit and inertia sensing
Device circuit connection;
The trans-impedance amplifier circuit is also connect with interface circuit and ADC conversion circuit.
Further, the model A301 of the force snesor;
Main control chip model ADS7041 in the ADC conversion circuit;
The main control chip in main control chip and trans-impedance amplifier circuit in the interface circuit is model MCP6001
Operational amplifier;
The main control chip model BME280 of the environmental sensor circuit;
The main control chip model BMI160 of the inertial sensor circuit.
A kind of talipes cavus detection method, comprising the following steps:
S1, gait data when acquiring user's walking by Intelligent insole, and are uploaded to host computer data processor;
S2, the gait data received is normalized by host computer data processor;
S3, the gait data after normalized is input in trained 1D convolutional neural networks, obtains the gait
The testing result of the corresponding talipes cavus of data.
Further, the method that gait data is normalized in the step S2 specifically:
The length of data sequences all in gait data is obtained into the equal number of time span by way of filling zero
According to sequence, the normalization of gait data is realized.
Further, the 1D neural network in the step S2 include sequentially connected first convolutional layer, the second convolutional layer,
First maximum pond layer, third convolutional layer, Volume Four lamination, the second maximum pond layer, flat layer, Dropout layers, first connect entirely
Connect layer, the second full articulamentum and output layer;
The number of the convolution kernel of first convolutional layer and the second convolutional layer is 64, and the length of convolution kernel is 7;
The number of the third convolutional layer and Volume Four lamination is 128, and the length of convolution kernel is 7;
Pool_size parameter in described first maximum pond layer and the second maximum pond layer is respectively 7 and 5;Described
The number of neuron is 128 in one full articulamentum and the second full articulamentum.
Further, to input data in first convolutional layer, the second convolutional layer, third convolutional layer and Volume Four lamination
Successively carry out convolutional calculation and batch normalized;
Wherein, the formula of convolutional calculation are as follows:
In formula, CjFor the ith feature figure after convolution;
Xi is the initial data or ith feature figure in i-th of channel;
It * is convolution operator;
wiFor i-th of convolution kernel;
bjFor bias term;
M is the port number of initial data or the data of characteristic pattern;
When carrying out criticizing normalized to the data after convolutional calculation, being normalized to mean value to each small batch data is 0
And the data that variance is 1;
Wherein, the activation primitive in the first convolutional layer, the second convolutional layer, third convolutional layer and Volume Four lamination is ReLU
Activation primitive;
ReLU activation primitive are as follows:
In formula, f (xij) it is activation primitive;
cijFor i-th of channel or j-th of value of ith feature figure.
Further, the described first maximum pond layer and the second maximum Chi Huacengzhongchiization formula are as follows:
In formula,For the value of ith feature figure Chi Huahou;
xi(j·m:(j+1)·m)It is characterized the input value of figure;
L is the data length of Chi Huahou characteristic pattern;
Length (x) is the length of primitive character figure;
For the arithmetic operation symbol that rounds up;
J is the number of the number of iterations of secondary maximizing each to input feature vector figure, and j=0,1 ..., l;
M is the length of template;
Activation primitive in the output layer is softmax nonlinear activation function;
Softmax nonlinear activation function are as follows:
In formula, f (zj) be Current neural member j output probability;
zjFor the output of the preceding layer related Neurons of j-th of output neuron and the product of respective weights.
Further, optimizer when being trained to 1D convolutional neural networks is ADADELTA optimizer.
The invention has the benefit that talipes cavus detection system provided by the invention and method, pass through customized prototype
Data, wherein contain inertia and pressure detecting, while the multi-modal sensing data recorded combines and can preferably reflect
The details of talipes cavus can preferably identify talipes cavus;The gait data for extracting acquisition by 1D convolutional neural networks is corresponding
Gait feature is proved to be the suitable selection of processing pressure and inertial sensor data, in the application of foot type classification, can reach
Preferable to differentiate performance, the combination of Intelligent insole and 1D neural network can be used for the screening of talipes cavus and be detected.
Detailed description of the invention
Fig. 1 is talipes cavus detection system structure chart provided by the invention.
Fig. 2 is the peripheral circuit diagram of main control chip in wireless microcontroller provided by the invention.
Fig. 3 is power circuit principle figure provided by the invention.
Fig. 4 is the interface circuit schematic diagram in Intelligent insole provided by the invention.
Fig. 5 is the reference voltage input circuit schematic diagram in interface circuit provided by the invention.
ADC conversion circuit schematic diagram in Fig. 6 Intelligent insole provided by the invention.
Trans-impedance amplifier circuit diagram in Fig. 7 Intelligent insole provided by the invention.
Inertial sensor circuit diagram in Fig. 8 Intelligent insole provided by the invention.
Environment sensor circuit schematic diagram in Fig. 9 Intelligent insole provided by the invention.
Figure 10 is pierce circuit schematic diagram provided by the invention.
Figure 11 is talipes cavus detection method flow chart provided by the invention.
Figure 12 is 1D convolutional neural networks structure chart provided by the invention.
Figure 13 is the equivalent circuit diagram of interface circuit in embodiment provided by the invention.
Figure 14 is the training of 1D convolutional neural networks and verifying loss effect experiment topic figure in embodiment provided by the invention.
Figure 15 is confusion matrix schematic diagram in embodiment provided by the invention.
Specific embodiment
A specific embodiment of the invention is described below, in order to facilitate understanding by those skilled in the art this hair
It is bright, it should be apparent that the present invention is not limited to the ranges of specific embodiment, for those skilled in the art,
As long as various change is in the spirit and scope of the present invention that the attached claims limit and determine, these variations are aobvious and easy
See, all are using the innovation and creation of present inventive concept in the column of protection.
As shown in Figure 1, a kind of talipes cavus detection system, including Intelligent insole interconnected and host computer data processing
Device;
Intelligent insole includes flexible printed circuit board, interface circuit, ADC conversion circuit and wireless microcontroller;
9 force snesors are provided on flexible printed circuit board, each force snesor passes through interface circuit and ADC is converted
Circuit connection, ADC conversion circuit and interface circuit are connect with wireless microcontroller, at wireless microcontroller and host computer data
Manage device connection;Host computer data processor is the computer detected by 1D neural network to input data.
Above-mentioned flexible printed circuit board be insole shape, 9 force snesors be respectively arranged on flexible printed circuit board with
At the corresponding big toe of foot bottom, at five caput metatarsales, mesopodium outside, at medial heel and at lateral heel;In Fig. 1
T1-T9 is the setting position of 9 force snesors.
Flexible printed circuit board includes the first flexible polyvinyl chloride plate and the second flexible polyvinyl chloride plate, the first flexible polychlorostyrene
Vinyl plate is set to above the second flexible polyvinyl chloride plate;
First flexible polyvinyl chloride plate is the soft flexible polyvinyl chloride panel with a thickness of 0.8mm;
Second flexible polyvinyl chloride plate is the rigid flexible polyvinyl chloride panel with a thickness of 0.5mm.
The model A301 of force snesor;The diameter in the active sensing region of the sensor is 9.53mm, with existing FSR
(force sensing resistance) is compared, and flexible circular sensor has more superior performance, therefore is typically used in insole prototype, according to
In the past to the research of maximum plantar pressure limit, the sensor version of 445N range, a disadvantage of a wide range of sensor are selected
It is that its sensitivity in narrower range is lower, therefore subsequent interface circuit design determines the overall sensitivity of system.
Above-mentioned wireless microcontroller includes model nRF52832 main control chip and its peripheral circuit;The wireless microcontroller
It ensures from sensor collection and transmits these data;The periphery of the main control chip in the wireless microcontroller is illustrated in Fig. 2
Circuit.
The peripheral circuit of main control chip includes pierce circuit, power circuit, trans-impedance amplifier circuit, program downloading electricity
Road, environmental sensor circuit and inertial sensor circuit;Power circuit also with trans-impedance amplifier circuit, program download circuit, ring
Border sensor circuit and inertial sensor circuit connection;Trans-impedance amplifier circuit is also connect with interface circuit and ADC conversion circuit.
The wireless microcontroller is mounted on the front of shoes, to reflect the motion profile of foot well, within the system
The chip of model nRF52832 as a System on Chip/SoC, hardware configuration include a 2.4G radio receiver and
The radio transmitter of one 2.4G is compatible with 1 MbpsLow-power consumption mode.In software aspects, nRF52832 chip
It include the executable software of an entitled SoftDevice S132, it provides the function of 5 protocol stack of bluetooth.Intelligent insole
Application program executable file and software executable file are stored in the program internal memory of nRF52832.No using section
In the case where energy strategy, exist in the minimum situation of transimission power, during active data transmission, the power consumption of equipment is no more than
2.4 milliamperes, data are by bluetooth low energy (BLE) technical transmission into the host equipment for receiving data.
As shown in figure 3, power circuit includes power supply sub-circuit (1) interconnected and voltage conversion circuit (2);Wherein supply
Electronic circuit passes through the battery of model CR2450 mainly as whole system power supply, and voltage conversion circuit mainly passes through model
The voltage regulator of REF3025 provides institute for artificial circuit part (interface circuit, ADC conversion circuit and trans-impedance amplifier circuit)
The reference voltage needed.
As shown in figure 4, illustrate the interface circuit schematic diagram of Intelligent insole in the present invention, the wherein model of main control chip
MCP6001, the interface circuit ensure the accuracy of 9 force snesors and the direct data transmission of wireless microcontroller;This connects
Mouth circuit also passes through reference voltage input circuit as shown in Figure 5 and connect with power circuit, and required base is provided for interface circuit
Quasi- voltage.
As shown in fig. 6, illustrating the ADC conversion circuit in the Intelligent insole in the present invention;Wherein, the model of main control chip
For ADS7041.
As shown in fig. 7, the circuit diagram of the trans-impedance amplifier circuit in the Intelligent insole in the present invention is illustrated, wherein
Main control chip model MCP6001.In the trans-impedance amplifier circuit, firstly, its output voltage and acting on sensor
On pressure present it is linearly related, this to avoid processing pressure sensor conductance rate and apply power on a sensor it
Between non-linear relation when complexity.Secondly, super simple amplifier ensures that the voltage on sensor is constant, to avoid
As sensor conductance dependent on the voltage applied and caused by complexity.
As shown in figure 8, illustrating the inertial sensor circuit diagram in the Intelligent insole in the present invention, the inertia sensing
The model BMI160 of device;The sensor is made of 3-axis acceleration and gyroscope, can acquire exercise data, and pass to rely on oneself
The exercise data of sensor.The foot movement of the dynamics and kinematics information that obtain simultaneously, makes complete gait analysis.
As shown in figure 9, illustrating the environmental sensor circuit diagram in the Intelligent insole in the present invention, the environmentally sensitive
The model BME280 of device;The environmental sensor measures the wet of environment during force snesor is calibrated and during normal operation
Degree and temperature;These data are for correcting sensor readings.Influence to avoid temperature to reading.
As shown in Figure 10, the pierce circuit in the Intelligent insole in the present invention, including higher-order of oscillation sub-circuit are illustrated
(1) and low-frequency oscillation sub-circuit (2);In order to guarantee the stable operation of equipment, wireless microcontroller selects two external oscillators
Instead of internal oscillator, the stability of oscillator and stringent timing are most important to the execution of Bluetooth protocol, while also guaranteeing
Stable sample rate.Wherein, higher-order of oscillation sub-circuit (1) includes crystal oscillator XT4, and the frequency of oscillation of crystal oscillator XT4 is 3200MHz;
Within the system, the timing sampled to force sensor signals is constituted using the clock signal of higher-order of oscillation sub-circuit.Low-frequency oscillation
Sub-circuit (2) includes crystal oscillator XT3, and the frequency of oscillation of crystal oscillator XT3 is 32768Hz;Within the system, low-frequency oscillation sub-circuit generates
Clock signal be used to form crucial sequential of fault required by Bluetooth protocol.
As shown in figure 11, the present invention also provides a kind of talipes cavus detection methods, comprising the following steps:
S1, gait data when acquiring user's walking by Intelligent insole, and are uploaded to host computer data processor;
S2, the gait data received is normalized by host computer data processor;
Due to the variation of user's gait, data length may be different in different tests for the data of same subject;It is different
The data sequence length of the not homogeneous test of subject may also be different;Therefore it needs data sequences all in gait data
Length obtains the equal data sequence of time span by way of filling zero, realizes the normalization of gait data.
S3, by the gait data after normalized, be input in trained 1D convolutional neural networks, obtain the gait
The testing result of the corresponding talipes cavus of data.
As shown in figure 12, the 1D neural network in the present invention includes sequentially connected first convolutional layer, the second convolutional layer, the
One maximum pond layer, third convolutional layer, Volume Four lamination, the second maximum pond layer, flat layer, Dropout layers, the first full connection
Layer, the second full articulamentum and output layer;Convolutional layer therein and pond layer allow the Automatic Feature Extraction from rudimentary to advanced, institute
The fill pattern for having convolution algorithm is same, so that having output and input identical dimension;
The number of the convolution kernel of above-mentioned first convolutional layer and the second convolutional layer is 64, and the length of convolution kernel is 7;Third
The number of convolutional layer and Volume Four lamination is 128, and the length of convolution kernel is 7;
Pool_size parameter in first maximum pond layer and the second maximum pond layer is respectively 7 and 5;With larger size
Filter compare, the parameter that the filter of smaller size needs to learn is smaller, can also learn more complicated characteristic;
The number of neuron is 128 in first full articulamentum and the second full articulamentum.
In 1D convolutional neural networks in the present invention, the two dimensional character figure that flattening layer exports maximum pond layer is converted
For a n dimensional vector n, practical batch processing normalization shortens the training time with drop-out, improves network performance;In neural network
In training process, carry out the network optimization using ADADELTA optimizer, it is one of ADAGRAD extension, with study into
Row, learning rate monotone decreasing.
In the first convolutional layer, the second convolutional layer, third convolutional layer and Volume Four lamination in the present invention to input data according to
Secondary progress convolutional calculation and batch normalized;
Wherein, the formula of convolutional calculation are as follows:
In formula, CjFor the ith feature figure after convolution;
xiFor the initial data or ith feature figure in i-th of channel;
It * is convolution operator;
wiFor i-th of convolution kernel;
bjFor bias term;
M is the port number of initial data or the data of characteristic pattern;
When carrying out criticizing normalized to the data after convolutional calculation, being normalized to mean value to each small batch data is 0
And the data that variance is 1;
Wherein, the mean value of the minimum lot data of upper one layer of output data is μβAre as follows:
In formula, xiFor upper one layer of output data;
M is the quantity of batching data;
The variance of the minimum lot data of upper one layer of output dataAre as follows:
The formula of normalized are as follows:
In formula,For the data after normalization;
ε be prevent denominator be 0 one added level off to 0 value;
The y that the value of previous step output is reconstructediAre as follows:
In formula, γ and β are the parameter for reconstructing study;
Wherein, the activation primitive in the first convolutional layer, the second convolutional layer, third convolutional layer and Volume Four lamination is ReLU
Activation primitive;
ReLU activation primitive are as follows:
In formula, f (xij) it is activation primitive;
cijFor i-th of channel or j-th of value of ith feature figure.
Above-mentioned first maximum pond layer and the second maximum Chi Huacengzhongchiization formula are as follows:
In formula,For the value of ith feature figure Chi Huahou;
xi(j·m:(j+1)·m)It is characterized the input value of figure;
L is the data length of Chi Huahou characteristic pattern;
Length (x) is the length of primitive character figure;
For the arithmetic operation symbol that rounds up;
J is the number of the number of iterations of secondary maximizing each to input feature vector figure, and j=0,1 ..., l;
Activation primitive in output layer is softmax nonlinear activation function;
Softmax nonlinear activation function are as follows:
In formula, f (zj) be Current neural member j output probability;
zjFor the output of the preceding layer related Neurons of j-th of output neuron and the product of respective weights.
In one embodiment of the invention, the work of corresponding hardware circuit when force snesor acquisition data-signal is provided
Make principle, the simplification functional diagram of a sensor is as shown in figure 13 in interface circuit (Fig. 4);
In many modern times AD, there is the function of a selection supply voltage as the reference voltage, we are directly obtained using it
Obtain electric conductivity value:
In this way, measured value is just independent of analog power (Vref) voltage value, actually electricity of the sensor under fixed force
Stream-voltage characteristic be it is nonlinear, in order to avoid this nonlinear influence, analog power is derived from accurate 2.5V adjuster,
It needs successively to activate force snesor in the actual process, be realized by selecting the input and output pin of wireless microcontroller, when
Preceding working sensor is connected with ground, remaining sensor is then in unactivated state, and the pin of wireless microcontroller is arranged with height
Resistance state is as input;In addition, inactive sensor is connected to the defeated of interface circuit by high-impedance resistors at anti-interference aspect
In entering, under high impedance status, the resistance of the resistance value ratio pin of these resistance is much smaller, therefore by inactive sensors both ends
Voltage is set as input voltage, avoids any influence to activation sensor, when sensor is active, series electrical
Resistance will not influence to measure, capacitor and and series electrical in Fig. 5 because electric current is ignored in this way between input voltage VI and Grod
Resistance is in parallel, provides low impedance path for external disturbance, by series resistance VI by buffer U7, is avoided that active sensor string
Join the influence of resistance.
Anti-aliasing solution of the RC filter as ADC;Its cutoff frequency is significantly larger than single pass requirement.It is selecting
When selecting it, it is important to ensure that its time constant is sufficiently low relative to selected sample frequency, to avoid previous sensor
Reading influence work at present sensor reading.The design assume compared with the strong signal from sensor, interference signal
The amplitude of component is very low.
Since the current-voltage characteristic of force snesor is nonlinear, the voltage at sensor both ends and additional power and mould
Quasi- supply voltage is in non-linear relation;This make consider institute related model derivation become complexity.Transimpedance amplifier exists
The voltage kept constant on sensor, so as to avoid relevant complexity.
In current design, the quantity of component can be reduced using ADC built in microcontroller.But it is selecting
NRF52832 and Vdd as the reference voltage in the case where, the gain of input stage must be chosen to be between 0.6 or 1.In former feelings
Under condition, transformed error is unacceptable big, and in the latter case, input range is limited;Wherein, Vdd is
Voltage on nRF52832 pin 13/36/48.
Intelligent insole in the present invention is related to inertial sensor data reading, force snesor control in data acquisition
System and reading and the transmission of Bluetooth Low Energy data, in data acquisition, the sample frequency of all the sensors is
100Hz, all operations are all final drives, and reading data is all using buffering.Mainly utilize inertial sensor internal damping
Area, for force snesor, there are two ping-pong buffer in the memory of wireless microcontroller, its reading and sensor
Data exchange is high ten times of frequency of the frequency that Interruption generates than single channel by Timer Controlling.In each interruption thing
In part, subsequent sensor is read, next to be activated.Therefore, the voltage " tail of previous sensor is had enough time so that
It fades out and corresponds to the output voltage stabilization of new sensor in portion ";For wireless transmission, data are divided into the packet-of four seed types
It is respectively used to left and right inertial sensor and left and right force sensor data.Data packet is by preamble counter and verifies sum
It header and is distinguished comprising field, force sensor data packet includes also the information for closing innersole dimensions, while also having been transmitted powered-down
The information of pond electricity.In order to receive data on personal computer (PC), we devise the data reception module of a customization,
In computer receiving end, data packet is sent to a virtual serial port and receives.The data reception module is one customized
Separate hardware module, it is connected to the USB port of computer, realizes the wireless communication with Intelligent insole.
In one embodiment of the invention, the test process that talipes cavus detection is carried out using present system is provided:
64 volunteers are shared in test and have participated in test, wherein normal foot 44, talipes cavus patient 20, subject
Feature is as shown in table 1, and during the test, subject wears the instrument shoes of suitable dimension, is gone ahead 7 meters with natural speed,
Period does not allow sudden change gait, and each subject carries out 20 tests, by testing each time under camera record, further
It analyzes since the test data sheet of data is lost and lost in wireless transmission, 1130 records are acquired in total;
Table 1: research object feature
Traditional foot type classification method is all based on the plantar pressure feature of manual extraction, these features can not be completely anti-
The feature of foot type is reflected, because not finding the feature of most distinctive, classifier may be unable to reach optimal performance, and deep
The method of degree study has the characteristics that stronger learning ability.
During the test, we respectively survey 1D neural network with the data of inertial sensor and force snesor
Examination, and put them on together, in order to be trained to network, for training, remaining sample is used to test 80% sample,
We train 1D neural network with 100 training samples, by 1D neural network obtain accuracy rate it is as shown in table 2, training and
As shown in figure 14, confusion matrix is as shown in figure 15 for verifying loss, and (a) is the confusion matrix of inertial data in Figure 15;It (b) is pressure
The confusion matrix of force data;It (c) is the confusion matrix of all data.
The performance of table 2:1DCNN
In order to be compared with conventional method, we test the random forest grader with default parameters setting, because
It is the better method classified to gait analysis correlated characteristic for it;Classifier uses the input feature vector being calculated, these
It is that maximum value, minimum value, average value, range, zero-crossing rate, root-mean-square value, variance, standard that each sequence obtains be anti-friction, the degree of bias
And kurtosis, table 3 give the performance of the random forest grader, the classifier is in inertial sensor data and force sensor data
On have preferable application effect, it is used due to using using the classifier precision of force sensor data when using single mode data
The classifier of property sensing data;
Table 3: the performance of random forest classification
Available to draw a conclusion by above-mentioned experimentation: 1D neural network shows in terms of Intelligent insole data processing
Good performance is gone out, the accuracy of pressure data test data set and all data has respectively reached 100%, tests at three
In variable, the performance of pressure data and total evidence is best, this result can prove that plantar nervous arch mode is arch of foot exception
Reliability index;Inertial data is also proved to be suitable for talipes cavus detection, up to the present, can be used for evaluating there are no parameter
Talipes cavus and splayfooted difference when heel contact, our inertial data classification results show that 1D neural network can be found
Parameter with distinguishing ability, although having only used an inertial sensor in the system, the identification interpretation of two kinds of foots reaches
To 96.46%, the data for increasing sensor and the position for improving sensor can be further improved precision, when we are by inertia
When data and pressure data merge, obtained precision is equal with the precision only obtained from pressure data, and multi-modal feature can be used for
The reliable recognition of foot type, and the random forest method of manual extraction feature is used, the performance in three data inputs is all
Not as good as 1D neural network.
The invention has the benefit that talipes cavus detection system provided by the invention and method, pass through customized prototype
Data, wherein contain inertia and pressure detecting, while the multi-modal sensing data recorded combines and can preferably reflect
The details of talipes cavus can preferably identify talipes cavus;The gait data for extracting acquisition by 1D convolutional neural networks is corresponding
Gait feature is proved to be the suitable selection of processing pressure and inertial sensor data, in the application of foot type classification, can reach
Preferable to differentiate performance, the combination of Intelligent insole and 1D neural network can be used for the screening of talipes cavus and be detected.
Claims (10)
1. a kind of talipes cavus detection system, which is characterized in that including Intelligent insole interconnected and host computer data processor;
The Intelligent insole includes flexible printed circuit board, interface circuit, ADC conversion circuit and wireless microcontroller;
9 force snesors are provided on the flexible printed circuit board, each force snesor passes through interface circuit and ADC
Conversion circuit connection, the ADC conversion circuit and interface circuit connect with wireless microcontroller, the wireless microcontroller and
The connection of host computer data processor;
The host computer data processor is the computer detected by 1D neural network to input data.
2. talipes cavus detection system according to claim 1, which is characterized in that the flexible printed circuit board is insole shaped
Shape, 9 force snesors are respectively arranged on flexible printed circuit board at big toe corresponding with foot bottom, five caput metatarsales
Place, mesopodium outside, at medial heel and at lateral heel;
The flexible printed circuit board includes the first flexible polyvinyl chloride plate and the second flexible polyvinyl chloride plate, and described first is flexible
Polyvinyl chloride panel is set to above the second flexible polyvinyl chloride plate;
The first flexible polyvinyl chloride plate is the soft flexible polyvinyl chloride panel with a thickness of 0.8mm;
The second flexible polyvinyl chloride plate is the rigid flexible polyvinyl chloride panel with a thickness of 0.5mm.
3. talipes cavus detection system according to claim 1, which is characterized in that wireless microcontroller includes model
NRF52832 main control chip and its peripheral circuit;
The peripheral circuit of the main control chip includes pierce circuit, power circuit, trans-impedance amplifier circuit, program downloading electricity
Road, environmental sensor circuit and inertial sensor circuit;
The power circuit is also electric with trans-impedance amplifier circuit, program download circuit, environmental sensor circuit and inertial sensor
Road connection;
The trans-impedance amplifier circuit is also connect with interface circuit and ADC conversion circuit.
4. talipes cavus detection system according to claim 3, which is characterized in that the model A301 of the force snesor;
Main control chip model ADS7041 in the ADC conversion circuit;
The main control chip in main control chip and trans-impedance amplifier circuit in the interface circuit is the fortune of model MCP6001
Calculate amplifier;
The main control chip model BME280 of the environmental sensor circuit;
The main control chip model BMI160 of the inertial sensor circuit.
5. a kind of talipes cavus detection method, which comprises the following steps:
S1, gait data when acquiring user's walking by Intelligent insole, and are uploaded to host computer data processor;
S2, the gait data received is normalized by host computer data processor;
S3, the gait data after normalized is input in trained 1D convolutional neural networks, obtains the gait data
The testing result of corresponding talipes cavus.
6. talipes cavus detection method according to claim 5, which is characterized in that the step S2 returns gait data
One changes the method for processing specifically:
The length of data sequences all in gait data is obtained into the equal data sequence of time span by way of filling zero
Column, realize the normalization of gait data.
7. talipes cavus detection method according to claim 5, which is characterized in that the 1D neural network packet in the step S2
Include sequentially connected first convolutional layer, the second convolutional layer, the first maximum pond layer, third convolutional layer, Volume Four lamination, second most
Great Chiization layer, flat layer, Dropout layers, the first full articulamentum, the second full articulamentum and output layer;
The number of the convolution kernel of first convolutional layer and the second convolutional layer is 64, and the length of convolution kernel is 7;
The number of the third convolutional layer and Volume Four lamination is 128, and the length of convolution kernel is 7;
Pool_size parameter in described first maximum pond layer and the second maximum pond layer is respectively 7 and 5;Described first is complete
The number of neuron is 128 in articulamentum and the second full articulamentum.
8. talipes cavus detection method according to claim 7, which is characterized in that first convolutional layer, the second convolutional layer,
Convolutional calculation and batch normalized are successively carried out in third convolutional layer and Volume Four lamination to input data;
Wherein, the formula of convolutional calculation are as follows:
In formula, CjFor the ith feature figure after convolution;
xiFor the initial data or ith feature figure in i-th of channel;
It * is convolution operator;
wiFor i-th of convolution kernel;
bjFor bias term;
M is the port number of initial data or the data of characteristic pattern;
When carrying out criticizing normalized to the data after convolutional calculation, being normalized to mean value to each small batch data is 0 and side
The data that difference is 1;
Wherein, the activation primitive in the first convolutional layer, the second convolutional layer, third convolutional layer and Volume Four lamination is ReLU activation
Function;
ReLU activation primitive are as follows:
In formula, f (xij) it is activation primitive;
cijFor i-th of channel or j-th of value of ith feature figure.
9. talipes cavus detection method according to claim 5, which is characterized in that the described first maximum pond layer and second is most
Pond formula in great Chiization layer are as follows:
In formula,For the value of ith feature figure Chi Huahou;
xi(j·m:(j+1)·m)It is characterized the input value of figure;
L is the data length of Chi Huahou characteristic pattern;
Length (x) is the length of primitive character figure;
For the arithmetic operation symbol that rounds up;
J is the number of the number of iterations of secondary maximizing each to input feature vector figure, and j=0,1 ..., l;
M is the length of template;
Activation primitive in the output layer is softmax nonlinear activation function;
Softmax nonlinear activation function are as follows:
In formula, f (zj) be Current neural member j output probability;
zjFor the output of the preceding layer related Neurons of j-th of output neuron and the product of respective weights.
10. according to the talipes cavus detection method in claim 5, which is characterized in that in the step S3, to 1D convolutional Neural net
Optimizer when network is trained is ADADELTA optimizer.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910328152.7A CN109924981A (en) | 2019-04-23 | 2019-04-23 | A kind of talipes cavus detection system and method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910328152.7A CN109924981A (en) | 2019-04-23 | 2019-04-23 | A kind of talipes cavus detection system and method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109924981A true CN109924981A (en) | 2019-06-25 |
Family
ID=66990737
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910328152.7A Pending CN109924981A (en) | 2019-04-23 | 2019-04-23 | A kind of talipes cavus detection system and method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109924981A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110597153A (en) * | 2019-10-22 | 2019-12-20 | 江苏金亿达能科技有限公司 | Novel six-axis attitude detection and attitude identification control system |
CN113057623A (en) * | 2021-03-30 | 2021-07-02 | 上海理工大学 | Flat foot disease screening insole and flat foot disease screening gait analysis system |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104082905A (en) * | 2014-06-18 | 2014-10-08 | 南京纳联信息科技有限公司 | Multifunctional intelligent insole and gait similarity detection method |
CN105852866A (en) * | 2016-04-01 | 2016-08-17 | 浙江大学 | Wearable sensing system and measuring method for knee-joint adduction torque measurement |
CN205885432U (en) * | 2016-06-22 | 2017-01-18 | 河北师范大学 | Plantar pressure testing arrangement based on shoe -pad |
CN107229325A (en) * | 2016-03-25 | 2017-10-03 | 蔡宇辉 | Data processing method, the device of intelligent foot wearable device and its system and system |
CN107247949A (en) * | 2017-08-02 | 2017-10-13 | 北京智慧眼科技股份有限公司 | Face identification method, device and electronic equipment based on deep learning |
CN107260176A (en) * | 2017-06-07 | 2017-10-20 | 深圳市奇诺动力科技有限公司 | Plantar pressure measuring device and method |
CN107890166A (en) * | 2017-11-03 | 2018-04-10 | 温州鞋都网络科技有限公司 | A kind of Intelligent insole based on plantar pressure gesture recognition |
CN108814566A (en) * | 2018-04-04 | 2018-11-16 | 太原理工大学 | Intelligent portable human body condition sensor |
-
2019
- 2019-04-23 CN CN201910328152.7A patent/CN109924981A/en active Pending
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104082905A (en) * | 2014-06-18 | 2014-10-08 | 南京纳联信息科技有限公司 | Multifunctional intelligent insole and gait similarity detection method |
CN107229325A (en) * | 2016-03-25 | 2017-10-03 | 蔡宇辉 | Data processing method, the device of intelligent foot wearable device and its system and system |
CN105852866A (en) * | 2016-04-01 | 2016-08-17 | 浙江大学 | Wearable sensing system and measuring method for knee-joint adduction torque measurement |
CN205885432U (en) * | 2016-06-22 | 2017-01-18 | 河北师范大学 | Plantar pressure testing arrangement based on shoe -pad |
CN107260176A (en) * | 2017-06-07 | 2017-10-20 | 深圳市奇诺动力科技有限公司 | Plantar pressure measuring device and method |
CN107247949A (en) * | 2017-08-02 | 2017-10-13 | 北京智慧眼科技股份有限公司 | Face identification method, device and electronic equipment based on deep learning |
CN107890166A (en) * | 2017-11-03 | 2018-04-10 | 温州鞋都网络科技有限公司 | A kind of Intelligent insole based on plantar pressure gesture recognition |
CN108814566A (en) * | 2018-04-04 | 2018-11-16 | 太原理工大学 | Intelligent portable human body condition sensor |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110597153A (en) * | 2019-10-22 | 2019-12-20 | 江苏金亿达能科技有限公司 | Novel six-axis attitude detection and attitude identification control system |
CN113057623A (en) * | 2021-03-30 | 2021-07-02 | 上海理工大学 | Flat foot disease screening insole and flat foot disease screening gait analysis system |
CN113057623B (en) * | 2021-03-30 | 2022-09-06 | 上海理工大学 | Gait analysis system for screening flat foot |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN103383292B (en) | Wearable real-time monitoring and feedback alarm device and method for scissors gaits | |
CN101394788B (en) | Gait analysis | |
US20110054359A1 (en) | Footwear-based body weight monitor and postural allocation, physical activity classification, and energy expenditure calculator | |
CN105125216B (en) | A kind of gait detecting system based on plantar pressure | |
CN107157485A (en) | A kind of intellectual monitoring shoe-pad and its intelligent monitor system | |
CN109171734A (en) | Human body behavioural analysis cloud management system based on Fusion | |
CN109567812B (en) | Gait analysis system based on intelligent insole | |
CN108814617A (en) | Freezing of gait recognition methods and device and gait detector | |
CN109924981A (en) | A kind of talipes cavus detection system and method | |
CN104101699A (en) | Improved analyte meter and method of operation | |
Cho | Design and implementation of a lightweight smart insole for gait analysis | |
CN108042106A (en) | A kind of artificial intelligence method for correcting error for improving human body physical sign Non-invasive detection equipment accuracy of detection | |
CN105054486A (en) | Health monitoring shoes | |
CN113331829A (en) | Sole information monitoring method and intelligent insole device | |
CN108567431A (en) | A kind of intelligent sensing boots for measuring body gait and leg speed | |
CN110638462B (en) | Wearable plantar physiological information acquisition system with gait analysis function | |
CN115868935B (en) | Wearable-based portable balance evaluation device | |
CN112263254A (en) | Human body energy consumption prediction system based on surface electromyogram signal sensor and prediction method thereof | |
Jiang et al. | An easy-to-use wearable step counting device for slow walking using ankle force myography | |
Shayan et al. | ShrewdShoe, a smart pressure sensitive wearable platform | |
CN206147905U (en) | System based on multi -sensor fusion fall detection | |
CN109998221B (en) | Intelligent insole for gait analysis | |
CN210673317U (en) | Plantar pressure sensor for correcting frequency influence | |
CN110200634B (en) | Plantar pressure sensor for correcting frequency influence and corresponding correction method | |
CN209789858U (en) | Plantar pressure measuring device |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190625 |