Diabetic foot risk assessment optimization system combining multi-feature signals and deep learning

Abstract

The application discloses a diabetic foot risk assessment optimization system combining a multi-feature signal and deep learning, collects multiple myoelectric signals, foot temperature signals, foot pressure signals and gait signals of a person to be assessed; signals corresponding to each foot during a gait are intercepted as a signal processing segment, each signal processing segment is processed to obtain qualified signal processing segments and splicing, qualified myoelectric signals, qualified foot temperature signals, qualified foot pressure signals and qualified gait signals after splicing are respectively subjected to feature extraction to form feature vectors; an improved gold panning algorithm is constructed, historical samples are used to preliminarily train and verify a model, fitness is obtained, the number of gold panners is set based on the fitness, the model after preliminary training is used to initialize a population of hyperparameters and random variations, the improved gold panning algorithm is used to optimize the hyperparameters to obtain an optimal model constructed by optimal hyperparameters, the feature vectors are input into the optimal model for assessment, and the diabetic foot risk assessment result of the person to be assessed is output.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, in particular to a diabetic foot risk assessment optimization system combining multiple feature signals and deep learning. BACKGROUND

[0002] Diabetic foot is one of the common complications of diabetic patients. If not treated in time, it may lead to foot ulcers, infections, necrosis and even amputation, which will seriously affect the quality of life of diabetic patients. Therefore, it is very important to find the signal of diabetic foot early.

[0003] The present application is designed by the applicant before the application number 2023113420679, the name of a diabetic foot risk level evaluation system based on myoelectric signal invention patent, the invention patent only considers the myoelectric signal as a feature representing the symptoms of diabetic foot, and does not consider other features representing the symptoms of diabetic foot, such as foot temperature, foot pressure, etc. The diabetic foot risk assessment optimization system designed by the present application collects myoelectric signals, foot temperature signals, foot pressure signals and gait signals, which can better represent all the characteristics of diabetic foot patients, so that the prediction result of the patient's diabetic foot risk is accurate.

[0004] Moreover, the neural network model involved in the present invention is directly trained by using the features extracted from the historical samples, and the training accuracy is not high, and the training accuracy needs to be improved. Based on this, the deep learning model constructed by the present application is optimized and trained, and the optimal deep learning model can be obtained, which can realize more accurate prediction of the risk of diabetic foot. SUMMARY

[0005] The present application provides a diabetic foot risk assessment optimization system combining multiple feature signals and deep learning to solve the problems and deficiencies of the prior art.

[0006] The present application solves the above technical problems by the following technical solutions:

[0007] The present application provides a diabetic foot risk assessment optimization system combining multiple feature signals and deep learning, characterized by comprising an intelligent boot and a control host, the intelligent boot is used for collecting myoelectric signals of multiple positions of the foot and lower limbs of the person to be evaluated within a predetermined time, foot temperature signals and foot pressure signals of multiple positions of the foot, and gait signals.

[0008] The control host is used for intercepting each corresponding muscle signal, sole temperature signal, sole pressure signal and gait signal of each foot bottom ground contact period of the to-be-evaluated person, each foot bottom ground contact period corresponding signal is as a signal processing segment, for each signal processing segment, signal preprocessing is carried out first, then the missing signal processing operation is carried out to obtain at least a preset number of qualified signal processing segments and is spliced in sequence, for each qualified muscle signal, qualified sole temperature signal, qualified sole pressure signal and qualified gait signal after splicing, each muscle feature, sole temperature feature, sole pressure feature and gait feature are obtained by feature extraction, and a feature vector is formed; the gold panning algorithm is improved by adding an adaptive weight, a disturbance factor and a gold panning strategy based on fitness, an improved gold panning algorithm is constructed, the deep learning established diabetic foot risk assessment model is preliminarily trained and verified by using historical samples, the fitness f is obtained, the fitness f and the preset fitness f1 are analyzed, when f is greater than or equal to f1, the number of gold panners is set as N1, when f is less than f1, the number of gold panners is set as N2, N1 is less than N2, the population in the improved gold panning algorithm is initialized by using the model hyperparameters after preliminary training and the hyperparameters randomly varied, the hyperparameters are optimized by using the improved gold panning algorithm and are brought into the model training, the optimal model constructed by the optimal hyperparameters is obtained through continuous iteration, the feature vector is input into the optimal model for evaluation, and the diabetic foot risk assessment result of the to-be-evaluated person is output.

[0009] The positive progress effect of the application is that:

[0010] The application considers all the characteristic features of diabetic foot, such as nerve and / or vascular lesions of lower limbs and feet caused by diabetic foot, so that the muscle signal, sole temperature signal, sole pressure signal and gait signal can reflect all the characteristic features of diabetic foot, so that the diabetic foot risk prediction result is more accurate.

[0011] Unlike the prior art, the application does not perform signal preprocessing on the multiple signals collected in the preset time, but intercepts the multiple signals in the foot bottom ground contact period which can better reflect the characteristics of diabetic foot, so that the data processing amount can be further reduced.

[0012] The application improves the existing gold panning algorithm, constructs an improved gold panning algorithm, optimizes and trains the deep learning model constructed by using the improved gold panning algorithm, and can obtain an optimized deep learning model, which can realize more accurate diabetic foot risk prediction.

[0013] Compared with the existing gold panning algorithm, the improved gold panning algorithm has high optimization precision, stronger global optimization ability and good convergence, and is not easy to fall into local optimization.

[0014] The prior art directly initializes the population of the gold panning algorithm, and obtains the optimal hyperparameters of the model through continuous iteration. The existing gold panning algorithm optimization process requires a large amount of data to be processed, and the optimization efficiency is not high. Unlike the prior art, the embodiment first uses historical samples to preliminarily train and verify the model, so as to obtain an adaptability f. When the adaptability f is greater than or equal to f1, it indicates that the adaptability is close to the ideal adaptability f2, that is, the model hyperparameters are close to the optimal hyperparameters, and at this time, a smaller number N1 of gold diggers can be set. Compared with the prior art, the processing mode of initializing the population of the improved gold panning algorithm by the hyperparameters close to the optimal hyperparameters and their mutations obtained through preliminary training and setting a smaller number of gold diggers can greatly reduce the amount of data to be processed in the optimization process of the improved gold panning algorithm, and improve the optimization efficiency. When the adaptability f is less than f1, it indicates that the adaptability is far from the ideal adaptability f2, that is, the model hyperparameters are far from the optimal hyperparameters, and at this time, a larger number N2 of gold diggers needs to be set. Compared with the prior art, the processing mode of initializing the population of the improved gold panning algorithm by the hyperparameters far from the optimal hyperparameters and their mutations obtained through preliminary training and setting a larger number of gold diggers can reduce the amount of data to be processed in the optimization process of the improved gold panning algorithm, and improve the optimization efficiency.

[0015] The prior art initializes the population very randomly in the upper and lower limit range, which is too random. The present application is different, and the model hyperparameters after preliminary training are used as the position of a gold digger in the population. The model hyperparameters after preliminary training are used as the benchmark, and random mutations are performed on this benchmark to obtain the positions of the remaining N-1 gold diggers in the population. This processing mode makes the population initialization position more targeted, realizes randomness on a certain benchmark, and is more conducive to the improved gold panning algorithm to quickly find the optimal position, that is, the optimal hyperparameters of the model. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 The control process diagram of the diabetes foot risk assessment optimization system of the preferred embodiment of the present application. DETAILED DESCRIPTION

[0017] To make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be described clearly and completely below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0018] It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other in the case of no conflict. The terms "middle", "horizontal", "vertical", "longitudinal", "front", "back", "left", "right", "inner", "outer" and the like representing the positional relationship are based on the positional relationship represented by the displayed drawings, and do not represent that the components referred to must be presented in the positional relationship expressed, and do not constitute a limitation on the technical scheme of the present application.

[0019] As shown in Figure 1 The present embodiment provides a diabetic foot risk assessment optimization system combining multi-feature signals and deep learning, which comprises an intelligent boot and a control host.

[0020] The inner bottom front of the shoe body of the intelligent boot is fixed with pressure sensors at positions corresponding to the two sides of the toes of the person to be evaluated, the inner bottom rear of the shoe body is fixed with pressure sensors at positions corresponding to the two sides of the heels of the person to be evaluated, and a flexible pressure sensor is fixed at a middle position of the inner bottom of the shoe body corresponding to the center of the foot of the person to be evaluated. The five flexible pressure sensors are used to detect the plantar pressure signals of the corresponding positions of the feet of the person to be evaluated.

[0021] A flexible temperature sensor is fixed at a middle position of the inner bottom front, middle and rear of the shoe body, respectively. The three flexible temperature sensors are used to detect the plantar temperature signals of the corresponding positions of the feet of the person to be evaluated.

[0022] A posture sensor is arranged on the shoe body, and the posture sensor is used to detect the gait signals of the person to be evaluated.

[0023] A flexible electromyography sensor is fixed at a position of the inner bottom middle of the shoe body corresponding to the arch of the person to be evaluated, a flexible electromyography sensor is fixed at a position of the inner bottom rear of the shoe body corresponding to the heel of the person to be evaluated, a flexible electromyography sensor is fixed at a position of the inner top of the shoe body corresponding to the dorsal of the person to be evaluated, a flexible electromyography sensor is fixed at a position of the inner side of the boot body corresponding to the peroneus longus muscle of the person to be evaluated, a flexible electromyography sensor is fixed at a position of the inner side of the boot body corresponding to the peroneus brevis muscle of the person to be evaluated, a flexible electromyography sensor is fixed at a position of the inner side of the boot body corresponding to the posterior tibial muscle of the person to be evaluated, a flexible electromyography sensor is fixed at a position of the inner side of the boot body corresponding to the lateral posterior of the person to be evaluated, and a flexible electromyography sensor is fixed at a position of the inner side of the boot body corresponding to the lateral lower leg of the person to be evaluated. The eight flexible electromyography sensors are used to detect the electromyography signals of the corresponding positions of the feet and lower limbs of the person to be evaluated.

[0024] The shoe body is further provided with a single-chip microcomputer and a wireless communication module. The single-chip microcomputer is used for collecting myoelectric signals of multiple positions of the foot and lower limbs of the person to be evaluated within a preset time, foot temperature signals and foot pressure signals of multiple positions of the foot sole, and gait signals, and transmitting each myoelectric signal, each foot pressure signal, each foot temperature signal and the gait signal to the control host computer through the wireless communication module.

[0025] The structure of the intelligent long boots of the embodiment can refer to the invention patent with the application number 2023113420679 and the name of a diabetes foot risk level evaluation system based on myoelectric signals applied by the inventor before.

[0026] The control host computer is used for intercepting each myoelectric signal, each foot temperature signal, each foot pressure signal and gait signal corresponding to each foot sole contact period (the foot sole contact period is a period during which the pressure value of any one of the five flexible pressure sensors is not zero) of the person to be evaluated. Each signal corresponding to each foot sole contact period is a signal processing segment. For each signal processing segment, signal preprocessing (such as noise reduction filtering) is performed first, and then an abnormal signal processing operation (such as using qualified signals to correct abnormal signals to obtain normal qualified signals) is performed, so as to obtain at least a preset number of qualified signal processing segments and splice them in order. For each qualified myoelectric signal, each qualified foot temperature signal, each qualified foot pressure signal and the qualified gait signal after splicing, feature extraction is performed to obtain each myoelectric feature, each foot temperature feature, each foot pressure feature and the gait feature. Each myoelectric feature, each foot temperature feature, each foot pressure feature and the gait feature constitute a feature vector.

[0027] Unlike the prior art, the embodiment does not perform signal preprocessing on the multiple signals collected within the preset time, but intercepts the multiple signals during the foot sole contact period for signal preprocessing. The advantage of this processing method is that the signals collected during the foot sole contact period can better reflect the characteristics of diabetic foot, and the data processing amount can be further reduced.

[0028] The control host computer is used for adding adaptive weights, disturbance factors and fitness-based gold panning strategy improvements to the gold panning algorithm, and constructing an improved gold panning algorithm.

[0029] Specifically,

[0030] Migration strategy:

[0031]

[0032] C1 = 2 * r2

[0033]

[0034] In the above formula, represents the new position of the gold panner i, denotes the current position of a prospector i, denotes the current best gold mine position, denotes the migration vector, A1, C1 denote the variable coefficients, 11 denotes the convergence factor, r1 and r2 are random numbers between 0 and 1, t denotes the current iteration number, T denotes the maximum iteration number, W(t) denotes the adaptive weight, W min denotes the minimum weight, W max denotes the maximum weight;

[0035] mining strategy:

[0036]

[0037] A2 = 12(2r3 - 1)

[0038]

[0039] In the above formula, denotes the current position of a randomly selected prospector g, A2 denotes the variable coefficient, denotes the mining vector, 12 denotes the convergence factor, r3 is a random number between 0 and 1;

[0040] cooperation strategy:

[0041]

[0042] In the above formula, denotes the cooperation vector, r4 denotes a random number between 0 and 1, denotes the current position of a randomly selected prospector g1, denotes the current position of a randomly selected prospector g2;

[0043] combination strategy of migration and mining strategy:

[0044]

[0045] combination strategy of mining and cooperation strategy:

[0046]

[0047] In the combination strategy,

[0048] In the above formula, r5, r6 and r7 are all random numbers between 0 and 1, and ε is a perturbation factor.

[0049] The traditional population algorithm has the advantages of easy implementation, high optimization efficiency, etc., but also has the disadvantages of being easy to fall into local optimum and poor convergence. In view of the defects of the traditional population algorithm, the improved gold panning algorithm is used to optimize the hyperparameters of the model in the embodiment, which has high optimization efficiency, is not easy to fall into local optimum and has good convergence.

[0050] In the embodiment, the self-adaptive weight W(t) is added in the migration strategy, the exploitation strategy and the cooperation strategy, and the self-adaptive weight gradually decreases with the increase of the iteration number, which is beneficial to convergence.

[0051] The control host is used to preliminarily train and verify the diabetic foot risk assessment model established by deep learning using historical samples, obtain the fitness f, analyze the fitness f and the preset fitness f1, set the number of gold panners as N1 when f≥f1, set the number of gold panners as N2 when f

[0052] The prior art directly initializes the population of the gold panning algorithm, and obtains the optimal hyperparameters of the model through continuous iteration. The data amount required to be processed in the optimization process of the existing gold panning algorithm is large, and the optimization efficiency is not high. The embodiment is different from the prior art, and the model is preliminarily trained and verified using historical samples, so that a fitness f can be obtained. When the fitness f≥f1, it indicates that the fitness is close to the ideal fitness f2, and at this time, a small number N1 of gold panners can be set. Through the population initialization of the improved gold panning algorithm using the hyperparameters and the mutation of the preliminary training, and the processing mode of setting a small number of gold panners, the data amount required to be processed in the optimization process of the improved gold panning algorithm can be greatly reduced, and the optimization efficiency can be improved. When the fitness f

[0053] Specifically,

[0054] 1) using historical samples to preliminarily train and verify the diabetic foot risk assessment model established based on deep learning, obtain fitness f, analyze the fitness f and the preset fitness f1, when f≥f1, set the number of gold diggers as N1, when f

[0055] Specifically includes:

[0056] The hyperparameters of the diabetic foot risk assessment model established by deep learning are set and initialized, and the hyperparameters such as the number of neurons and the learning rate can be set according to actual needs.

[0057] The historical samples are divided into historical training samples and historical verification samples.

[0058] The initialized hyperparameters are substituted into the established diabetic foot risk assessment model, and the historical training samples and the historical verification samples are used to preliminarily train and verify the diabetic foot risk assessment model, respectively, to obtain the fitness f.

[0059] Among them, the composition of the historical sample is:

[0060] The intelligent long boots are used to collect the myoelectric signals of multiple positions of the foot and lower limbs, the plantar temperature signals and plantar pressure signals of multiple positions of the foot, and the gait signals within a preset time of each historical detector, and label the diabetic foot risk detection results of each historical detector, and the diabetic foot risk detection results include different degrees of diabetic foot risk degree.

[0061] The control host is used to intercept each myoelectric signal, each plantar temperature signal, each plantar pressure signal and gait signal corresponding to each foot during the period of each historical detector, each signal corresponding to each foot during the period is taken as a signal processing segment, for each signal processing segment of each historical detector, signal preprocessing is performed first, then the missing signal processing operation is performed to obtain at least a preset number of qualified signal processing segments and sequentially spliced, for each qualified myoelectric signal, each qualified plantar temperature signal, each qualified plantar pressure signal and qualified gait signal of each historical detector after splicing, feature extraction is performed to obtain each myoelectric feature, each plantar temperature feature, each plantar pressure feature and gait feature, to form the feature vector of the historical detector, thereby obtaining the feature vector of each historical detector, and the feature vector of each historical detector and the corresponding diabetic foot risk detection result constitute the historical sample.

[0062] 2) The model hyperparameters after preliminary training are denoted as U, and U is used to initialize the position of one gold digger in the population S, and random mutation operation is performed on U within the upper and lower limits of the hyperparameters to initialize the remaining N-1 gold diggers in the population S N-1= (U, rand(N-1, D)), each prospector in population S represents a set of hyperparameters.

[0063] Wherein, N is the population size, N=N1 or N2, D is the spatial dimension, the spatial dimension D corresponds to the hyperparameters one by one, T is the maximum number of iterations.

[0064] In this embodiment, unlike the prior art, the population is not initialized in the upper and lower limit range at will, but the hyperparameters of the model after preliminary training are used as the position of a prospector in the population, and the hyperparameters of the model after preliminary training are used as the reference, and random mutation is performed on the basis of the reference to obtain the positions of the remaining N-1 prospectors in the population. This processing method makes the population initialization more targeted and more conducive to quickly finding the optimal hyperparameters.

[0065] 3) Substitute the hyperparameters corresponding to each prospector in 2) into the diabetic foot risk assessment model after preliminary training, and train using historical samples to obtain the fitness f of each prospector. The maximum value of the fitness f of each prospector is the best fitness of the historical population, and the current position of the prospector corresponding to the best fitness of the historical population is the best gold mine position.

[0066] 4) Enter iteration, when 0.5 < τ < 1, use migration strategy or exploitation strategy for position update, and when the best fitness of the historical population of the last two consecutive iterations is not updated, use the combination of migration and exploitation strategies for position update.

[0067] When τ≤0.5, use exploitation strategy or cooperation strategy for position update, and when the best fitness of the historical population of the last two consecutive iterations is not updated, use the combination of exploitation and cooperation strategies for position update.

[0068] Wherein, the iteration correlation coefficient

[0069] Update the position of each prospector Substitute the corresponding hyperparameters into the current diabetic foot risk assessment model, and train using historical samples to obtain the current fitness f of each prospector. Compare the current fitness f of each prospector with the historical individual best fitness of the corresponding prospector. When the current fitness f of the prospector is greater than the corresponding historical individual best fitness, update the historical individual best fitness of the prospector = the current fitness f of the prospector, and update the position of the prospector Otherwise, do not update; the maximum value of the historical individual optimal fitness of each gold digger is the current population optimal fitness, if the current population optimal fitness is greater than the historical population optimal fitness, update the historical population optimal fitness = the current population optimal fitness, the current position of the gold digger corresponding to the current population optimal fitness is the best gold mine position, otherwise, do not update.

[0070] 5) Determine whether the historical population optimal fitness reaches the preset fitness f2 or the iteration number reaches the maximum iteration number T, if yes, obtain the optimal hyperparameter based on the position of the gold digger corresponding to the historical population optimal fitness, the optimal hyperparameter constructs an optimal model, input the feature vector into the optimal model for evaluation, and output the diabetic foot risk assessment result of the to-be-evaluated person, f2 > f1, if no, enter 4) for the next iteration.

[0071] In the embodiment, unlike the existing gold digger algorithm, when the random value r takes a value of 0 < r < 1 / 3, a migration strategy is selected for position updating; when the random value r takes a value of 1 / 3≤r<2 / 3, a mining strategy is selected for position updating; and when the random value r takes a value of 2 / 3≤r<1, a cooperation strategy is selected for position updating. Instead, an iteration-related coefficient τ related to the iteration number is set, τ decreases with the increase of the iteration number. In the early iteration, the migration strategy or the mining strategy is used for position updating, and when the historical population optimal fitness of the last two consecutive iterations is not updated, the combination strategy of the migration and mining strategies is used for position updating, and a disturbance factor is added to the adaptive weight to prevent the improved gold digger algorithm from falling into local optimum. In the later iteration, the mining strategy or the cooperation strategy is used for position updating, and when the historical population optimal fitness of the last two consecutive iterations is not updated, the combination strategy of the mining and cooperation strategies is used for position updating, and a disturbance factor is added to the adaptive weight to prevent the improved gold digger algorithm from falling into local optimum.

[0072] Although the specific embodiments of the present application are described above, those skilled in the art should understand that these are only illustrative, the protection scope of the present application is defined by the appended claims. Those skilled in the art can make various changes or modifications to these embodiments without departing from the principles and essence of the present application, and these changes and modifications all fall within the protection scope of the present application.

Claims

1. A system for optimizing risk assessment of diabetic foot using a combination of multi-feature signals and deep learning, characterized in that, Intelligent long boots and a control host are included, the intelligent long boots are used to collect myoelectric signals of multiple positions of feet and lower limbs, plantar temperature signals and plantar pressure signals of multiple positions of foot bottom, and gait signals of a predetermined time of an evaluated person; The control host is used to intercept each myoelectric signal, plantar temperature signal, plantar pressure signal and gait signal corresponding to each foot bottom ground period of the evaluated person, and each signal corresponding to each foot bottom ground period is taken as a signal processing segment, for each signal processing segment, signal preprocessing is performed first, then an abnormal signal processing operation is performed to obtain at least a predetermined number of qualified signal processing segments and splice them in order, and for each qualified myoelectric signal, qualified plantar temperature signal, qualified plantar pressure signal and qualified gait signal after splicing, feature extraction is performed to obtain each myoelectric feature, plantar temperature feature, plantar pressure feature and gait feature, thereby forming a feature vector; The control host is used to add adaptive weight, disturbance factor and fitness-based gold panning strategy improvement to the gold panning algorithm to construct an improved gold panning algorithm: Migration and exploitation strategy combination strategy: Exploitation and cooperation strategy combination strategy: In connection with the strategy, In the above formula, represents the new position of the prospector i, represents the migration strategy, represents the exploitation strategy, represents the cooperation strategy, r5, r6 and r7 are random numbers between 0 and 1, and ε is a disturbance factor; The control host is used to preliminarily train and verify a diabetic foot risk assessment model established by deep learning using historical samples to obtain fitness f, analyze the fitness f and a predetermined fitness f1, set the number of gold panners as N1 when f≥f1, set the number of gold panners as N2 when f<f1, N1<N2, use the model hyperparameters after preliminary training and the hyperparameters randomly varied from the hyperparameters to initialize the population in the improved gold panning algorithm, optimize the hyperparameters by using the improved gold panning algorithm and input the hyperparameters into the model training, obtain an optimal model constructed by optimal hyperparameters through continuous iteration, input the feature vector into the optimal model for evaluation, and output the diabetic foot risk assessment result of the evaluated person.

2. The multi-feature signal and deep learning combined diabetic foot risk assessment optimization system of claim 1, wherein, The control host is used to preliminarily train and verify a diabetic foot risk assessment model established by deep learning using historical samples to obtain fitness f, specifically including: The upper and lower limits of the hyperparameters of the diabetic foot risk assessment model established by deep learning are set and initialized respectively; The historical samples are divided into historical training samples and historical verification samples; The initialized hyperparameters are substituted into the established diabetic foot risk assessment model, and the historical training samples and the historical verification samples are used to preliminarily train and verify the diabetic foot risk assessment model respectively to obtain fitness f.

3. The multi-feature signal and deep learning combined diabetic foot risk assessment optimization system of claim 1, wherein, The composition of the historical samples is specifically: The intelligent long boots are used to collect myoelectric signals of multiple positions of feet and lower limbs, plantar temperature signals and plantar pressure signals of multiple positions of foot bottom, and gait signals of a predetermined time of each historical detector, and the diabetic foot risk detection results of each historical detector are labeled, and the diabetic foot risk detection results include different degrees of diabetic foot risk degrees. The control host is used to intercept each foot bottom ground period corresponding signal of each historical detector, each foot bottom ground period corresponding signal is as a signal processing segment, for each historical detector each signal processing segment, signal preprocessing is carried out first, then the missing signal processing operation is carried out to obtain at least a preset number of qualified signal processing segments and is spliced in order, for each historical detector, the qualified myoelectric signal, the qualified foot bottom temperature signal, the qualified foot bottom pressure signal and the qualified gait signal are respectively extracted to obtain the myoelectric feature, the foot bottom temperature feature, the foot bottom pressure feature and the gait feature, and the feature vector of the historical detector is formed, thereby obtaining the feature vector of each historical detector, and the feature vector of each historical detector and the corresponding diabetic foot risk detection result form a historical sample.

4. The multi-feature signal and deep learning combined diabetic foot risk assessment optimization system of claim 1, wherein the control host is used to preliminarily train and verify the diabetic foot risk assessment model established by deep learning using the historical sample, obtain an adaptability f, analyze the adaptability f and a preset adaptability f1, set the number of gold diggers as N1 when f≥f1, set the number of gold diggers as N2 when f 1) preliminarily train and verify the diabetic foot risk assessment model established based on deep learning using the historical sample, obtain an adaptability f, analyze the adaptability f and a preset adaptability f1, set the number of gold diggers as N1 when f≥f1, and set the number of gold diggers as N2 when f 2) After the initial training, the model hyperparameters are denoted as U, and the positions of one gold digger in the population S are initialized with U. The remaining N-1 gold diggers in the population S are initialized by randomly mutating the hyperparameters within the upper and lower limits of the hyperparameters. N-1 = (U, rand(N-1, D)) positions, and each gold digger in the population S represents a set of hyperparameters. wherein N is the population number, N=N1 or N2, D is the spatial dimension, the spatial dimension D is one-to-one corresponding to the hyperparameter, and T is the maximum number of iterations; 3) the hyperparameters corresponding to each gold digger in 2) are substituted into the preliminarily trained diabetic foot risk assessment model, and the historical sample is trained to obtain the adaptability f corresponding to each gold digger, the maximum value of the adaptability f of each gold digger is the best adaptability of the population, and the current position of the gold digger corresponding to the best adaptability of the population is the best gold mine position; 4) when 0.5<τ<1, the migration strategy or the mining strategy is used for position updating, and when the historical best adaptability of the population of the last two consecutive iterations is not updated, the migration and mining strategy combination strategy is used for position updating; when τ≤0.5, the mining strategy or the cooperation strategy is used for position updating, and when the historical best adaptability of the population of the last two consecutive iterations is not updated, the mining and cooperation strategy combination strategy is used for position updating, and τ is an iteration correlation coefficient; The current diabetes foot risk assessment model corresponding to the updated position of each prospector is substituted into the current diabetes foot risk assessment model, and the historical samples are trained respectively to obtain the current fitness f of each prospector. The current fitness f of each prospector is compared with the historical individual optimal fitness of the corresponding prospector. When the current fitness f of the prospector is greater than the corresponding historical individual optimal fitness, the historical individual optimal fitness of the prospector is updated to the current fitness f of the prospector, and the position of the prospector is updated. Otherwise, it is not updated. The maximum value in the historical individual optimal fitness of each prospector is taken as the current population optimal fitness. When the current population optimal fitness is greater than the historical population optimal fitness, the historical population optimal fitness is updated to the current population optimal fitness, and the current position of the prospector corresponding to the current population optimal fitness is taken as the best gold mine position. Otherwise, it is not updated. 5) Determine whether the historical population optimal fitness reaches the preset fitness f2 or the iteration number reaches the maximum iteration number T. When yes, the optimal hyperparameters are obtained based on the prospector corresponding to the historical population optimal fitness, the optimal model is constructed based on the optimal hyperparameters, the feature vector is input into the optimal model for evaluation, and the diabetes foot risk assessment result of the to-be-evaluated person is output. f2>f1, and when no, go to 4) for the next iteration.

5. The multi-feature signal and deep learning combined diabetic foot risk assessment optimization system of claim 4, wherein, Iterative correlation coefficient 6. The multi-feature signal and deep learning combined diabetic foot risk assessment optimization system of claim 1, wherein, The control host is used to add adaptive weight, disturbance factor and prospector strategy improvement based on fitness to the prospector algorithm, and construct an improved prospector algorithm. Specifically: Migration strategy: C1＝2*r2 In the above formula, denotes the new position of the prospector i, denotes the current position of the prospector i, denotes the current best gold mine position, denotes the migration vector, A1, C1 denote the variable coefficients, 11 denotes the convergence factor, r1 and r2 are random numbers between 0 and 1, t denotes the current iteration number, T denotes the maximum iteration number, W(t) denotes the adaptive weight, W min denotes the minimum weight, W max denotes the maximum weight; Mining strategy: A2＝l2(2r3-1) In the above formula, A2 represents a variable coefficient, represents a mining vector, l2 represents a convergence factor, and r3 is a random number between 0 and 1; Collaboration strategy: In the above formula, represents a cooperation vector, r4 represents a random number between 0 and 1, represents the current position of the randomly selected gold digger g1, represents the current position of the randomly selected gold digger g2.

7. The multi-feature signal and deep learning combined diabetic foot risk assessment optimization system of claim 1, wherein, The inner bottom of the shoe body of the intelligent long boots is fixed with pressure sensors at positions corresponding to the two sides of the toes of the to-be-evaluated person, the inner bottom of the shoe body is fixed with pressure sensors at positions corresponding to the two sides of the heels of the to-be-evaluated person, and a flexible pressure sensor is fixed at a position corresponding to the center of the to-be-evaluated person. The five flexible pressure sensors are used to detect the plantar pressure signals of the corresponding positions of the to-be-evaluated person. The inner bottom of the shoe body is fixed with flexible temperature sensors at positions corresponding to the center of the front part, the center of the middle part and the center of the rear part, respectively. The three flexible temperature sensors are used to detect the plantar temperature signals of the corresponding positions of the to-be-evaluated person. A posture sensor is arranged on the shoe body, and the posture sensor is used to detect the gait signals of the to-be-evaluated person. A single-chip microcomputer and a wireless communication module are further arranged on the shoe body. The single-chip microcomputer is used to transmit each electromyographic signal, each plantar pressure signal, each plantar temperature signal and gait signal to the control host through the wireless communication module.

8. The multi-feature signal and deep learning combined diabetic foot risk assessment optimization system of claim 7, wherein, The plantar part is in a period during which the pressure value of any one of the five flexible pressure sensors is not zero.

Images