A training path intelligent system for a rehabilitation garden

Abstract

The application discloses a kind of training footpath intelligent systems for rehabilitation garden.The system of the application includes detection footpath, rehabilitation garden with multiple training footpaths, main control module;Detection footpath is equipped with multiple pressure detection plates, multiple cameras;Pressure detection plate is electrically connected with the main control module;Camera is electrically connected with main control module;Main control module includes data storage module, calculation module, training footpath deep learning neural network;Data storage module is electrically connected with multiple pressure detection plates, multiple cameras respectively;Calculation module is electrically connected with data storage module;Training footpath deep learning neural network is electrically connected with data storage module, calculation module respectively.The application can collect the lower limb rehabilitation condition of patient in time, guide patient to use suitable rehabilitation footpath type, and utilize sound and light to guide patient to carry out rehabilitation training.

Description

Technical Field

[0001] This invention belongs to the technical field of medical rehabilitation training equipment, specifically relating to an intelligent system for training trails in a rehabilitation garden. Background Technology

[0002] As an outdoor landscape environment, the design of a rehabilitation garden needs to consider the construction of rehabilitation trails. Patients with lower limb motor dysfunction can gradually alleviate their symptoms and even achieve recovery by repeatedly walking on these trails during rehabilitation training. As the training progresses, the pathological condition of the patient's lower limb joints and muscles will continuously change. Patients with lower limb dysfunction need to utilize rehabilitation aids, such as walkers, steppers, and orthotics, to provide support for their lower limbs during rehabilitation training. During the rehabilitation period, the muscles of the lower limbs will gradually recover, but dangerous movements can affect the progress. During rehabilitation treatment, clinicians usually require patients to perform a certain amount of regular training exercises daily and prohibit movements that are detrimental to rehabilitation. The attending physician needs to be aware of the patient's rehabilitation training progress in a timely manner to intervene and provide guidance.

[0003] Currently, the rehabilitation aids and rehabilitation gardens widely used in clinical practice only serve the basic functions of support, fixation, and providing space. When assisting patients in rehabilitation training, they lack the ability to intelligently identify the recovery effect of rehabilitation training, intelligently track the movement during the rehabilitation training process, intelligently warn of dangerous movements during rehabilitation training, and provide timely guidance on the selection of rehabilitation trails. They can only rely on rehabilitation therapists to estimate on-site based on their observation experience, and cannot provide doctors with specific data in a timely manner to determine whether the current rehabilitation training recovery effect meets expectations, making it difficult for doctors to provide timely and effective rehabilitation training guidance. Summary of the Invention

[0004] In order to overcome one or more defects and deficiencies in the existing technology, the present invention provides an intelligent system for training trails in rehabilitation gardens, which guides patients to select appropriate rehabilitation trails for treatment based on their lower limb rehabilitation status when they are undergoing lower limb rehabilitation training.

[0005] To achieve the above objectives, the present invention adopts the following technical solution.

[0006] A smart system for training trails in a rehabilitation garden includes a detection trail, a rehabilitation garden with multiple training trails, and a main control module.

[0007] The rehabilitation garden features multiple training trails with adjustable surface materials, slopes, and step density, designed for lower limb rehabilitation patients to perform lower limb rehabilitation training.

[0008] The inspection walkway is equipped with multiple pressure detection plates and multiple cameras;

[0009] The pressure detection board is electrically connected to the main control module and is used to collect gait pressure data when lower limb rehabilitation patients walk.

[0010] The camera is electrically connected to the main control module and is used to collect gait images of lower limb rehabilitation patients while walking;

[0011] The main control module includes a data storage module, a computing module, and a deep learning neural network for training the trail.

[0012] The data storage module is electrically connected to multiple pressure detection boards and multiple cameras to store gait pressure and gait image data.

[0013] The calculation module is electrically connected to the data storage module and is used to calculate the pressure weights and gait weights of the input channels of the training gait deep learning neural network based on gait pressure and gait images.

[0014] The training trail deep learning neural network is electrically connected to the data storage module and the computing module, respectively. It is used to set the training trail deep learning neural network according to the pressure weight and gait weight of different input channels, and then input gait images to perform deep learning recognition and output the corresponding type of training trail.

[0015] Preferably, five pressure testing plates are evenly spaced on the testing trail to collect gait pressure generated by the left and right feet of lower limb rehabilitation patients when they walk different lengths on the testing trail.

[0016] On the testing trail, twelve cameras are set up at equal intervals with the center of the testing trail as the center, a radius of one meter, and an angle of 30°. One camera faces the lower limb rehabilitation patient, and one camera faces away from the lower limb rehabilitation patient.

[0017] Furthermore, the calculation module calculates the pressure weight as follows: after the lower limb rehabilitation patient walks the entire testing path, the pressure weight A of the left foot is obtained. The pressure weight B of the right foot is

[0018] Where L represents the average gait pressure of the left foot detected by all pressure detection plates on the detection trail, and R represents the average gait pressure of the right foot detected by all pressure detection plates on the detection trail.

[0019] Furthermore, the calculation module calculates the image weights in the following way:

[0020] After each of the twelve cameras completes one capture, the image data captured by each camera is a channel, resulting in twelve channels of gait images, denoted as L1, L2, L3, L4, L5, R1, R2, R3, R4, R5, W, and S. Among them, L1, L2, L3, L4, and L5 correspond to the gait images of cameras set on the same side of the detection path, R1, R2, R3, R4, and R5 correspond to the gait images of cameras set on the other side of the detection path, W corresponds to the gait image of the camera facing the lower limb rehabilitation patient, and S represents the gait image of the camera facing away from the lower limb rehabilitation patient.

[0021] The image weight C corresponding to the gait images from cameras on each of the L1, L2, L3, L4, and L5 sides is calculated using the following formula:

[0022]

[0023] Among them, X1, X2, X3, X4, and X5 are the independent weights of the gait images corresponding to the five cameras on the same side of the trail, which are obtained by training the trail deep learning neural network through pre-training;

[0024] The image weight D corresponding to the gait images of cameras on one side (R1, R2, R3, R4, R5) is calculated using the following formula:

[0025]

[0026] Among them, Y1, Y2, Y3, Y4, and Y5 are the independent weights of the gait images corresponding to the five cameras on the other side of the trail, which are obtained by training the trail deep learning neural network through pre-training.

[0027] Furthermore, the deep learning neural network for training the trail has input channels: Red channel, Green channel, Blue channel, and Alpha channel.

[0028] The input to the Red channel is the gait image corresponding to W; the input to the Green channel is the gait image corresponding to S; the input to the Blue channel is the gait image corresponding to image weight C, and the weight of the Blue channel is image weight C; the input to the Alpha channel is the gait image corresponding to image weight D, and the weight of the Alpha channel is image weight D.

[0029] The input image for training the gait deep learning neural network is to synthesize the gait images corresponding to the four channels (Red, Green, Blue, and Alpha) into a single four-channel image.

[0030] Furthermore, the training path deep learning neural network also includes a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a third convolutional layer, a third pooling layer, a fourth convolutional layer, a fourth pooling layer, a first fully connected layer, and a second fully connected layer connected in sequence.

[0031] The first convolutional layer is connected to the input channel and has 32 5×5 convolutional kernels; the second convolutional layer has 64 5×5 convolutional kernels; the third convolutional layer has 128 3×3 convolutional kernels; and the fourth convolutional layer has 128 3×3 convolutional kernels.

[0032] The first, second, third, and fourth pooling layers all use 2×2 convolutional kernels;

[0033] Both the first and second fully connected layers are 1152-dimensional.

[0034] Furthermore, the pre-training of the deep learning neural network for the trail specifically includes:

[0035] Training samples, test samples, and independent validation sets were randomly selected from the gait image dataset of lower limb rehabilitation patients in a ratio of 70%:20%:10%, and the corresponding training gait type was used as the label for the gait images of lower limb rehabilitation patients.

[0036] The training samples are input into the training trail deep learning neural network for training, and then the training trail deep learning neural network generates the corresponding parameters.

[0037] Input the test samples into the training deep learning neural network of the trail and perform input-output tests, and adjust the learning rate and number of iterations of the training deep learning neural network of the trail.

[0038] An independent validation set is input into the training gait deep learning neural network to evaluate the accuracy of the training gait deep learning neural network in identifying the training gait type when inputting gait images of lower limb rehabilitation patients. Then, the corresponding parameters are adjusted until the training gait deep learning neural network converges.

[0039] Preferably, it also includes a wearable lower limb module;

[0040] The lower limb wearable module communicates wirelessly with the data storage module of the main control module to collect the location coordinates of lower limb rehabilitation patients.

[0041] Preferably, the rehabilitation garden is equipped with an adjustment module;

[0042] The adjustment module is electrically connected to the main control module and is used to adjust the road surface material and / or slope and / or step density of the training trail according to the type of training trail identified by the deep learning neural network of the training trail.

[0043] Preferably, the rehabilitation garden is equipped with an audio-visual guidance module;

[0044] The sound and light guidance module is electrically connected to the main control module and is used to send sound and light signals to guide lower limb rehabilitation patients to select the corresponding type of training path based on the training path type identified by the deep learning neural network of the training path.

[0045] Compared with the prior art, the technical solution of the present invention has the following beneficial effects:

[0046] This invention trains a deep learning neural network for the trail to identify gait images of lower limb rehabilitation patients collected on the detection trail, and obtains the most suitable training trail for the corresponding lower limb rehabilitation patient in the current state, thereby guiding the lower limb rehabilitation patient into the corresponding training trail for rehabilitation training; it has the ability to intelligently identify the recovery effect of rehabilitation training, and can provide timely guidance on the selection of rehabilitation trails, enabling rehabilitation physicians to provide effective rehabilitation training guidance remotely in a timely manner.

[0047] The main control module's calculation module further calculates the relevant parameter weights of the deep learning neural network for the training path corresponding to the gait image by setting the pressure weights of the left and right feet of the lower limb rehabilitation patient. This can avoid interference in the deep learning process and specifically analyze the pathological condition of the lower limbs of patients with lower limb dysfunction. This helps the training path deep learning neural network accurately identify the current rehabilitation status of the lower limb rehabilitation patient and further select the most suitable training path type for the patient to walk.

[0048] Compared to the experience-based guidance of rehabilitation physicians, the deep learning neural network training path has the advantages of standardization and continuous learning and correction. The adjustable module design of the training path can provide more diverse training methods, reflecting the strong correlation between the selection of the training path and the current gait of patients with lower limb dysfunction. This invention uses a detection, identification, diagnosis, and training approach to help patients with lower limb dysfunction undergo rehabilitation training in a reasonable and rapid manner, avoiding overtraining, incorrect training, and ineffective training. Attached Figure Description

[0049] Figure 1 This is a schematic diagram of the framework structure of one embodiment of the present invention;

[0050] Figure 2 This is a schematic diagram of the framework for training the deep learning neural network for trails according to the present invention;

[0051] Figure 3 A schematic diagram of the framework for inspecting the trails and rehabilitation gardens. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0053] Example

[0054] like Figure 1 , Figure 2 , Figure 3 As shown, the intelligent system for training trails in the rehabilitation garden in this embodiment includes a detection trail, a rehabilitation garden with N training trails (N≥3), a main control module, and a lower limb wearable module.

[0055] The rehabilitation garden features multiple training trails with adjustable surface materials, slopes, and step density, designed for lower limb rehabilitation patients to perform lower limb rehabilitation training.

[0056] The testing walkway is equipped with five pressure detection plates and twelve cameras. The pressure detection plates are electrically connected to the main control module and are used to collect gait pressure data from lower limb rehabilitation patients while walking. The cameras are also electrically connected to the main control module and are used to capture gait images of the patients while walking. The pressure detection plates are evenly spaced along the length of the walkway to collect gait pressure data generated by the left and right feet of the patients as they walk different distances. The cameras are evenly spaced along the walkway, with a radius of one meter and an angle of 30° around the center. One camera faces the patient directly, and another faces away from the patient.

[0057] The main control module includes a data storage module, a computing module, and a deep learning neural network for training the trail.

[0058] The data storage module is electrically connected to five pressure detection plates and twelve cameras to store gait pressure and gait image data. The calculation module is electrically connected to the data storage module and calculates the pressure and gait weights of the input channels for training the deep learning neural network for the gait path based on gait pressure and gait images. The training deep learning neural network for the gait path is electrically connected to both the data storage module and the calculation module. After setting the training deep learning neural network according to the pressure and gait weights of different input channels, it inputs gait images for deep learning to identify the corresponding type of training gait path, providing rehabilitation training for lower limb rehabilitation patients.

[0059] The calculation module calculates the pressure weight as follows: The duration of the patient's walk on the path is set to T seconds, and the sampling frequency of the pressure gait detection board is k times per second. After the pressure gait detection board completes T×k sampling, the lower limb rehabilitation patient walks the entire detection path, and the pressure weight A of the left foot is obtained as follows: The pressure weight B of the right foot is: Where L represents the average gait pressure of the left foot detected by all pressure detection plates on the detection trail, and R represents the average gait pressure of the right foot detected by all pressure detection plates on the detection trail.

[0060] The calculation module calculates the image weights as follows: After each of the twelve cameras completes one capture, the image data captured by each camera is a channel, resulting in twelve gait images, denoted as L1, L2, L3, L4, L5, R1, R2, R3, R4, R5, W, and S. L1, L2, L3, L4, and L5 correspond to the gait images of cameras positioned on the same side of the detection path; R1, R2, R3, R4, and R5 correspond to the gait images of cameras positioned on the other side of the detection path; W corresponds to the gait image of the camera facing the lower limb rehabilitation patient; and S represents the gait image of the camera facing away from the lower limb rehabilitation patient. The formula for calculating the image weight C corresponding to the gait images of cameras on one side of L1, L2, L3, L4, and L5 is as follows:

[0061]

[0062] Among them, X1, X2, X3, X4, and X5 are the independent weights of the gait images corresponding to the five cameras on the same side of the trail, which are obtained by training the trail deep learning neural network through pre-training;

[0063] The image weight D corresponding to the gait images of cameras on one side (R1, R2, R3, R4, R5) is calculated using the following formula:

[0064]

[0065] Among them, Y1, Y2, Y3, Y4, and Y5 are the independent weights of the gait images corresponding to the five cameras on the other side of the trail, which are obtained by training the trail deep learning neural network through pre-training.

[0066] The training deep learning neural network for gait tracking has four input channels: Red, Green, Blue, and Alpha. The Red channel input is the gait image corresponding to weight W; the Green channel input is the gait image corresponding to weight S; the Blue channel input is the gait image corresponding to weight C, with the weight of the Blue channel being the same as image weight C; and the Alpha channel input is the gait image corresponding to weight D, with the weight of the Alpha channel being the same as image weight D. The input image for training the deep learning neural network for gait tracking is obtained by combining the gait images corresponding to the four channels (Red, Green, Blue, and Alpha) into a single four-channel image.

[0067] The training trail deep learning neural network consists of a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a third convolutional layer, a third pooling layer, a fourth convolutional layer, a fourth pooling layer, a first fully connected layer, and a second fully connected layer, connected in sequence.

[0068] In the deep learning neural network for training trails, the first convolutional layer is connected to the input channels and is used to receive the four-channel image to be identified. It has 32 5×5 convolutional kernels. The second convolutional layer has 64 5×5 convolutional kernels. The third convolutional layer has 128 3×3 convolutional kernels. The fourth convolutional layer has 128 3×3 convolutional kernels. The first, second, third, and fourth pooling layers all have 2×2 convolutional kernels. The first and second fully connected layers are both 1152-dimensional. The second fully connected layer is used to output the identified training trail type.

[0069] The pre-training of the deep learning neural network for the trail includes the following steps:

[0070] S1. Randomly select training samples, test samples, and independent validation sets from the gait image dataset of lower limb rehabilitation patients in a ratio of 70%:20%:10%, and label the gait images of lower limb rehabilitation patients with the corresponding training gait type.

[0071] S2. Input the training samples into the training trail deep learning neural network for training, and wait for the training trail deep learning neural network to generate the corresponding parameters.

[0072] S3. Input the test samples into the training deep learning neural network of the trail and perform input-output testing. Adjust the learning rate and number of iterations of the training deep learning neural network.

[0073] S4. Input the independent validation set into the training path deep learning neural network, evaluate the accuracy of the training path deep learning neural network in identifying the training path type when inputting gait images of lower limb rehabilitation patients, and then adjust the corresponding parameters until the training path deep learning neural network converges.

[0074] There are two wearable lower limb modules, one worn on the left foot and the other on the right foot of the lower limb rehabilitation patient. The lower limb wearable modules communicate wirelessly with the data storage module of the main control module to collect the location coordinates of the lower limb rehabilitation patient. This allows the calculation module to distinguish whether the detected gait pressure belongs to the left or right foot, and also enables the main control module to identify the current spatial position of the lower limb rehabilitation patient.

[0075] The rehabilitation garden is equipped with an adjustment module and an audio-visual guidance module. The adjustment module is electrically connected to the main control module and is used to adjust parameters such as the surface material, slope, and step density of the training trail according to the type of training trail identified by the deep learning neural network of the training trail. Specifically, the adjustment of the training trail is achieved by changing the state of the corresponding mechanical structure. The audio-visual guidance module is also electrically connected to the main control module and is used to emit audio-visual signals to guide lower limb rehabilitation patients to select the corresponding type of training trail according to the type of training trail identified by the deep learning neural network of the training trail. The audio-visual signals of the audio-visual guidance module come from a voice transmitter and LED indicator lights, respectively. In this preferred embodiment, the training trail is composed of multiple training trail slabs with different surface materials and / or slopes and / or step densities stacked from top to bottom. The adjustment module will, according to the material and / or slope and / or step density of the corresponding training trail given by the deep learning neural network of the training trail, move and sink the training trail slabs of the incompatible type above the training trail, and move the training trail slabs of the compatible type to the surface of the training trail. In a further preferred embodiment, the audio-visual guidance module will also provide audio-visual signal prompts for the patient to adjust parameters such as walking speed, stride length, balance, and training time during the walking process, guiding the patient to carry out lower limb rehabilitation training, avoiding situations where the patient walks too fast, has uneven stride length, loses balance, or has too long a training time due to lack of guidance, thereby effectively avoiding overtraining or ineffective training.

[0076] When using the intelligent training trail system for rehabilitation gardens in this embodiment, lower limb rehabilitation patients first walk across the detection trail, and the camera captures the gait image to be identified. Then, the gait image to be identified is transmitted to the pre-trained deep learning neural network for training trails in the main control module. The deep learning neural network for training trails identifies the corresponding type of training trail. Then, the adjustment module of the training trail adjusts the surface material, slope, or density of the corresponding training trail. Finally, the sound and light guidance module emits sound and light signals to guide the lower limb rehabilitation patients into the corresponding training trail for rehabilitation training.

[0077] Compared with the prior art, the beneficial effects of this embodiment are as follows:

[0078] This embodiment trains a deep learning neural network for the trail to identify gait images of lower limb rehabilitation patients collected on the detection trail, and obtains the most suitable training trail for the corresponding lower limb rehabilitation patient in the current state, thereby guiding the lower limb rehabilitation patient into the corresponding training trail for rehabilitation training; it has the ability to intelligently identify the recovery effect of rehabilitation training, and can provide timely guidance on the selection of rehabilitation trails, realizing the function of effective rehabilitation training guidance that rehabilitation physicians can provide remotely in a timely manner.

[0079] The main control module's calculation module further calculates the relevant parameter weights of the deep learning neural network for the training path corresponding to the gait image by setting the pressure weights of the left and right feet of the lower limb rehabilitation patient. This can avoid interference in the deep learning process and specifically depict the pathological condition of the lower limbs of patients with lower limb dysfunction. This helps the training path deep learning neural network accurately identify the current rehabilitation status of the lower limb rehabilitation patient and further select the most suitable training path type for the patient to walk.

[0080] Compared to the experience-based guidance of rehabilitation physicians, the deep learning neural network training path has the advantages of standardization and continuous learning and correction. The adjustable module design of the training path can provide more diverse training methods, reflecting the strong correlation between the selection of the training path and the current gait of patients with lower limb dysfunction. This embodiment uses a detection, identification, diagnosis, and training approach to help patients with lower limb dysfunction undergo rehabilitation training in a reasonable and rapid manner, avoiding overtraining, incorrect training, and ineffective training.

[0081] The above embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments. Any changes, modifications, substitutions, combinations, or simplifications made without departing from the spirit and principle of the present invention shall be considered equivalent substitutions and shall be included within the protection scope of the present invention.

Claims

1. A training path intelligent system for a rehabilitation garden, characterized in that, The application relates to a rehabilitation garden with a detection path and a plurality of training paths. The plurality of training paths in the rehabilitation garden can be respectively adjusted in road surface material, slope and ladder density, and are used for lower limb rehabilitation training of lower limb rehabilitation patients. The detection path is provided with a plurality of pressure detection plates and a plurality of cameras. The pressure detection plates are electrically connected with the master control module and are used for collecting gait pressure of the lower limb rehabilitation patients when walking. The cameras are electrically connected with the master control module and are used for collecting gait images of the lower limb rehabilitation patients when walking. The master control module comprises a data storage module, a calculation module and a training path deep learning neural network. The data storage module is electrically connected with the plurality of pressure detection plates and the plurality of cameras and is used for storing data of the gait pressure and the gait images. Five pressure detection plates are arranged at equal intervals on the detection path and are used for collecting gait pressure of the left foot and the right foot of the lower limb rehabilitation patients when walking through different lengths on the detection path. Twelve cameras are arranged at equal intervals on the detection path, with the center of the detection path as the center, one meter as the radius and 30 DEG as the included angle, wherein one camera faces the lower limb rehabilitation patient and one camera faces away from the lower limb rehabilitation patient. The calculation module is electrically connected with the data storage module and is used for calculating pressure weight and gait weight of an input channel of the training path deep learning neural network according to the gait pressure and the gait images. The training path deep learning neural network is electrically connected with the data storage module and the calculation module, inputs the gait images for deep learning identification after the pressure weight and the gait weight of different input channels are set, and outputs corresponding types of training paths. The calculation module calculates the pressure weight in the following manner: after the lower limb rehabilitation patient walks through the entire detection path, the pressure weight A of the left foot is obtained as The pressure weight B of the right foot is obtained as L represents average gait pressure of the left foot detected by all the pressure detection plates on the detection path, and R represents average gait pressure of the right foot detected by all the pressure detection plates on the detection path. The calculation module calculates the gait weight in the following mode: After each of the twelve cameras completes one shooting, the image data of each camera is one channel, and a total of twelve channels of gait images are obtained and are respectively denoted as L1, L2, L3, L4, L5, R1, R2, R3, R4, R5, W and S. The gait images of the cameras on one side correspond to the gait weight C, and the calculation formula is as follows: X1, X2, X3, X4 and X5 are independent weights of the gait images of the five cameras on the same side and are obtained by pre-training of the training path deep learning neural network. The gait images of the cameras on one side correspond to the gait weight D, and the calculation formula is as follows: Y1, Y2, Y3, Y4, Y5 are independent weights of the gait images corresponding to the five cameras on the other side of the training path, obtained by pre-training through the training path deep learning neural network; The training path deep learning neural network is provided with input channels Red channel, Green channel, Blue channel and Alpha channel; The Red channel input is the gait image corresponding to W; the Green channel input is the gait image corresponding to S; the Blue channel input is the gait image corresponding to the gait weight C, and the weight of the Blue channel is the gait weight C; the Alpha channel input is the gait image corresponding to the gait weight D, and the weight of the Alpha channel is the gait weight D; The input image of the training path deep learning neural network is a four-channel image composed of the gait images corresponding to the Red channel, Green channel, Blue channel and Alpha channel respectively.

2. The intelligent system for the training path of the rehabilitation garden according to claim 1, characterized in that, The training path deep learning neural network is further provided with a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a third convolutional layer, a third pooling layer, a fourth convolutional layer, a fourth pooling layer, a first full connection layer and a second full connection layer connected in sequence; The first convolutional layer is connected with the input channel and is provided with 32 5*5 convolutional kernels; the second convolutional layer is provided with 64 5*5 convolutional kernels; the third convolutional layer is provided with 128 3*3 convolutional kernels; and the fourth convolutional layer is provided with 128 3*3 convolutional kernels; The first pooling layer, the second pooling layer, the third pooling layer and the fourth pooling layer are all 2*2 convolutional kernels; The first full connection layer and the second full connection layer are both 1152-dimensional. 3.The intelligent system for the training path of the rehabilitation garden according to claim 2, characterized in that, The pre-training of the training path deep learning neural network specifically includes: According to the ratio of 70%:20%:10%, randomly select training samples, test samples and independent verification sets from the gait image data set of lower limb rehabilitation patients, and take the corresponding training path type as the label of the gait image of the lower limb rehabilitation patient; Input the training samples into the training path deep learning neural network for training, and wait for the training path deep learning neural network to generate corresponding parameters; Input the test samples into the training path deep learning neural network for input and output test, and adjust the learning rate and iteration step number of the training path deep learning neural network; Input the independent verification set into the training path deep learning neural network, evaluate the accuracy of the training path type recognized by the training path deep learning neural network when inputting the gait image of the lower limb rehabilitation patient walking, and then adjust the corresponding parameters until the training path deep learning neural network converges.

4. The intelligent system for the training path of the rehabilitation garden according to claim 1, characterized in that, It also includes a lower limb wearable module; The lower limb wearable module is in wireless communication with the data storage module of the main control module, and is used for collecting the position coordinates of the lower limb rehabilitation patient.

5. The intelligent system for the training path of the rehabilitation garden according to claim 1, characterized in that, The rehabilitation garden is provided with an adjusting module; The adjusting module is electrically connected with the main control module, and is used for adjusting the road surface material and / or slope and / or stair density of the training path according to the training path type recognized by the training path deep learning neural network.

6. The intelligent system for the training path of the rehabilitation garden according to claim 1, characterized in that, The rehabilitation garden is provided with an audible and light guiding module; The sound and light guiding module is electrically connected with the master control module, and is used for emitting a sound and light signal for guiding the lower limb rehabilitation patient to select a corresponding type of training walkway according to a training walkway type recognized by the training walkway depth learning neural network.

Images