A neural network-based data classification processing system

Abstract

The application provides a neural network-based data classification processing system, and relates to the technical field of data processing.The neural network-based data classification processing system comprises a data acquisition unit, a temporary storage unit, a configuration unit, an in-memory computing unit and a post-processing unit.The neural network-based data classification processing system provided in the application can improve the calculation efficiency of data classification processing.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and more specifically to a data classification and processing system based on neural networks. Background Technology

[0002] Physiological data, as a special type of input signal, can not only monitor the body's condition but also conveniently control external devices. For example, brain-computer interfaces (BCIs) provide a new control method for people with disabilities or movement disorders, helping them achieve self-care in daily life and even complete rehabilitation. They can also play a role in assisting with daily work and entertainment. With technological advancements, more and more advanced human-computer interfaces have been integrated into society and are widely used in medical rehabilitation and daily life.

[0003] Physiological signals are highly random, possess strong temporal attributes and user specificity, and are extremely weak and easily affected by noise. Compared to classical data analysis methods, the application of machine learning and artificial intelligence provides more reliable and accurate identification results for physiological data. For example, multi-level deep learning models have significant advantages in feature extraction and modeling, and can more effectively solve social and medical problems, such as human-computer interaction and disease monitoring, becoming a mainstream development trend. Research on their clinical applications continues to expand.

[0004] Deep learning models were initially designed to run on traditional hardware like CPUs or GPUs. Their training and inference processes involve high computational complexity and resource requirements, limiting their application primarily to high-performance computers. Furthermore, with the emergence of deeper and more complex neural network algorithms, such as the currently popular Transformer model, the computational demands and parameter count are increasing hundreds of times annually. To leverage the advantages of deep learning, cloud computing is typically chosen. This involves transmitting collected physiological data from sensor nodes to an external computing engine via wired or wireless communication methods such as Bluetooth or Wi-Fi. The external computer or server cluster then performs inference and analysis of the data, and the classification results are sent back to the local device to complete the specific task. Physiological data has a high real-time requirement, and data transmission involves latency. Cloud computing cannot guarantee timeliness, while local classification based on physiological data requires substantial computational resources due to its enormous computational demands. Summary of the Invention

[0005] To address the problems in the prior art, embodiments of the present invention provide a data classification and processing system based on neural networks, which can at least partially solve the problems existing in the prior art.

[0006] On one hand, this invention proposes a data classification and processing system based on neural networks, comprising:

[0007] The system comprises a data acquisition unit, a temporary storage unit, a configuration unit, an in-memory computing unit, and a post-processing unit; wherein:

[0008] The data acquisition unit is used to collect user physiological data and convert the user physiological data into digital signals;

[0009] The temporary storage unit is used to receive and save digital signals and the intermediate output results of the neural network returned by the post-processing unit;

[0010] The configuration unit is used to receive the initial parameters of the neural network input by the external device and the intermediate output results of the neural network returned by the post-processing unit, and generate configuration parameters for in-memory computation.

[0011] The in-memory calculation unit is used to receive configuration parameters and digital signals, and perform in-memory calculations on the configuration parameters and digital signals to obtain intermediate calculation results;

[0012] The post-processing unit is used to receive intermediate calculation results and perform post-processing calculations on the intermediate calculation results to obtain intermediate output results of the neural network and neural network classification results based on a preset task.

[0013] The temporary storage unit includes a preprocessing module, a first storage module, and a first output module; wherein:

[0014] The preprocessing module is used to preprocess digital signals;

[0015] The first storage module is used to save the preprocessed digital signals and intermediate output results of the neural network;

[0016] The first output module is used to send the preprocessed digital signal and the intermediate output result of the neural network to the in-memory computing unit.

[0017] The configuration unit includes a second storage module and a second output module; wherein:

[0018] The second storage module is used to store neural network parameters;

[0019] The second output module is used to send configuration parameters adapted to the output data of the temporary storage unit to the in-memory computing unit.

[0020] The post-processing unit includes a calculation module and a classification module; wherein:

[0021] The calculation module is used to perform post-processing calculations on the intermediate calculation results;

[0022] The classification module is used to receive the post-processing calculation results and classify the user's physiological data based on a preset task to obtain the neural network classification results.

[0023] The neural network-based data classification and processing system further includes a feedback unit;

[0024] The feedback unit is used to receive the neural network classification results and feed them back to the user.

[0025] The data acquisition unit includes wearable hardware sensors.

[0026] Specifically, the preprocessing module is used for:

[0027] The digital signal is filtered, aligned, and segmented to obtain data in a preset standard format.

[0028] The in-memory computing unit includes one of SRAM, RRAM, and DRAM.

[0029] Specifically, the calculation module is used for:

[0030] The intermediate calculation results are activated, regularized, pooled, and residuals are calculated.

[0031] Specifically, the feedback unit is used for:

[0032] Based on the classification results, determine the control feedback instructions for external devices and / or issue warning messages.

[0033] The data classification and processing system based on neural networks provided in this invention includes a data acquisition unit, a temporary storage unit, a configuration unit, an in-memory calculation unit, and a post-processing unit. Specifically: the data acquisition unit collects user physiological data and converts it into digital signals; the temporary storage unit receives and stores the digital signals and the intermediate output results of the neural network returned by the post-processing unit; the configuration unit receives initial neural network parameters input from an external device and the intermediate output results of the neural network returned by the post-processing unit, and generates configuration parameters for in-memory calculation; the in-memory calculation unit receives the configuration parameters and the digital signals, performs in-memory calculations on the configuration parameters and the digital signals, and obtains intermediate calculation results; the post-processing unit receives the intermediate calculation results and performs post-processing calculations on the intermediate calculation results to obtain intermediate output results of the neural network and neural network classification results based on a preset task, thereby improving the computational efficiency of data classification and processing. Attached Figure Description

[0034] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings:

[0035] Figure 1 This is a schematic diagram of the structure of a data classification and processing system based on a neural network provided in an embodiment of the present invention.

[0036] Figure 2 This is a schematic diagram of the structure of a data classification and processing system based on a neural network provided in another embodiment of the present invention.

[0037] Figure 3 This is a schematic diagram of the structure of a data classification and processing system based on a neural network provided in another embodiment of the present invention.

[0038] Figure 4 This is a schematic diagram of the structure of a data classification and processing system based on a neural network provided in another embodiment of the present invention.

[0039] Figure 5 This is a schematic diagram of the structure of a data classification and processing system based on a neural network provided in another embodiment of the present invention. Detailed Implementation

[0040] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings. Here, the illustrative embodiments and descriptions of the present invention are used to explain the present invention, but are not intended to limit the present invention. It should be noted that, unless otherwise specified, the embodiments and features in the embodiments of this application can be arbitrarily combined with each other.

[0041] The relevant terms in the embodiments of the present invention are explained as follows:

[0042] Neural network: A mathematical model that uses a structure similar to the synaptic connections of the brain to process information; the model referred to below is the neural network model.

[0043] Inference: This function completes the forward propagation of the neural network and obtains the corresponding prediction results.

[0044] Von Neumann architecture: a computer design architecture that combines program instruction memory and data memory together; in this patent, it refers to the concept of separating storage devices from the processor.

[0045] RAM: Random Access Memory.

[0046] Due to the privacy sensitivity of physiological data and the high real-time requirements of signal analysis tasks, local data processing and analysis is an ideal solution. However, one obstacle preventing the use of deep learning for local physiological data analysis is its high energy consumption. In practical applications, the analysis tasks are closely related to the user's physiological state and require long-term, stable, and continuous operation; therefore, the corresponding hardware and systems should be able to operate independently for extended periods.

[0047] The characteristics of wearable devices severely limit their energy consumption and scalability. Under the traditional von Neumann architecture, deep learning, as a data-intensive and computationally intensive task, faces two major problems: the "power wall" and the "memory wall." Because the memory transfer speed cannot keep up with the processor's performance, the processor's computing power is limited, which restricts high-performance computing and reduces the computation speed of deep learning models. On the other hand, data needs to be constantly moved between storage and computing units, resulting in most of the time, bandwidth, and power being consumed in data movement rather than computation. These additional data movement overheads account for nearly 90% of the total power consumption, wasting a lot of energy.

[0048] Computing in-memory (CIM) is a technology that embeds computing units into memory, aiming to address the "Von Neumann bottleneck" in current hardware architectures: performance degradation and power consumption loss caused by data transfer between the processor and computer memory. By implementing computational operations in memory, the boundary between memory and processing units is eliminated, eliminating the need to reload data from memory, providing faster computing speeds and lower energy consumption. Furthermore, CIM features high energy efficiency, high parallelism, and a compact footprint, enabling lighter and faster hardware implementations, making it more suitable for wearable devices used in edge computing.

[0049] Figure 1 This is a schematic diagram of the structure of a data classification and processing system based on a neural network according to an embodiment of the present invention, as shown below. Figure 1 As shown, the data classification processing system based on neural networks provided in this embodiment of the invention includes:

[0050] It includes a data acquisition unit 1, a temporary storage unit 2, a configuration unit 3, an in-memory computing unit 4, and a post-processing unit 5; wherein:

[0051] The data acquisition unit 1 is used to collect user physiological data and convert it into digital signals; the conversion can be achieved through analog-to-digital conversion. The data acquisition unit 1 includes wearable hardware sensors, such as EEG helmets and fitness trackers.

[0052] The temporary storage unit 2 is used to receive and save digital signals and the intermediate output results of the neural network returned by the post-processing unit;

[0053] The configuration unit 3 is used to receive the initial parameters of the neural network input by the external device and the intermediate output results of the neural network returned by the post-processing unit, and generate configuration parameters for in-memory computation. The configuration parameters specifically include the parameters of each layer of the neural network, with the first layer parameter denoted as W1, the second layer parameter denoted as W2, and so on.

[0054] The in-memory calculation unit 4 is used to receive configuration parameters and digital signals, and perform in-memory calculations on the configuration parameters and digital signals to obtain intermediate calculation results; the digital signals are calculated with the first layer parameters to obtain intermediate calculation result R1, and then the calculations are performed sequentially based on the first layer neural network result Y1 and the second layer parameters to obtain intermediate calculation result R2, and so on.

[0055] The in-memory computing unit 4 includes one of SRAM, RRAM, and DRAM. The hardware used to implement the unit's functions can employ various storage media, including but not limited to SRAM, RRAM, and DRAM, as well as flash and MRAM, etc.

[0056] The post-processing unit 5 is used to receive intermediate calculation results and perform post-processing calculations on the intermediate calculation results to obtain intermediate output results of the neural network and neural network classification results based on a preset task. Referring to the above example, post-processing calculations are performed on intermediate calculation results R1 to obtain the first layer neural network result Y1; post-processing calculations are performed on intermediate calculation results R2 to obtain the second layer neural network result Y2, and so on.

[0057] like Figure 2 As shown, the temporary storage unit 2 includes a preprocessing module 21, a first storage module 22, and a first output module 23; wherein:

[0058] The preprocessing module 21 is used to preprocess the digital signal; specifically, the preprocessing module 21 is used for:

[0059] The digital signal is filtered, aligned, and segmented to obtain data in a preset standard format. The preset standard format data can be set according to actual needs.

[0060] The first storage module 22 is used to store the preprocessed digital signals and intermediate output results of the neural network; see the above description.

[0061] The first output module 23 is used to send the preprocessed digital signal and the intermediate output result of the neural network to the in-memory computing unit 4.

[0062] like Figure 3As shown, the configuration unit 3 includes a second storage module 31 and a second output module 32; wherein:

[0063] The second storage module 31 is used to store neural network parameters; the neural network parameters include the initial parameters of the neural network and the parameters of each layer of the neural network at each time.

[0064] The second output module 32 is used to send configuration parameters adapted to the output data of the temporary storage unit to the in-memory computing unit 4. Since the in-memory computing unit 4 needs to perform calculations based on the data in the configuration unit 3 and the temporary storage unit 2, the data of the two need to be adapted to facilitate the calculations of the in-memory computing unit 4.

[0065] like Figure 4 As shown, the post-processing unit 5 includes a calculation module 51 and a classification module 52; wherein:

[0066] The calculation module 51 is used to perform post-processing calculations on the intermediate calculation results; specifically, the calculation module 51 is used for:

[0067] The intermediate calculation results are then subjected to activation, regularization, pooling, and residual calculations. These calculations are all standard neural network calculations.

[0068] The classification module 52 is used to receive the post-processing calculation results and classify the user's physiological data based on a preset task to obtain a neural network classification result. The preset task can be understood as the task to be performed based on the neural network, such as classifying and recognizing the user's emotions in the physiological data. The neural network classification result can include happiness, sadness, etc.

[0069] like Figure 5 As shown, the neural network-based data classification processing system further includes a feedback unit 6; the feedback unit 6 is used to receive the neural network classification result and feed it back to the user. Specifically, the feedback unit 6 is used for:

[0070] Based on the classification results, determine the control feedback commands for external devices and / or issue warning messages. This includes, but is not limited to, manipulating devices and sending messages. Referring to the above description, control the corresponding external devices according to different user emotions, and perform emotion-regulating operations on the user through the external devices.

[0071] It can also send warning messages related to happiness and sadness, such as "Please do not be overly sad".

[0072] The aforementioned physiological data classification system can be divided into three parts: ① collecting physiological data and providing classification results; ② configuring and deploying a deep learning neural network model; and ③ running the neural network model using in-memory computing technology to complete data inference calculations. ① and ③ are the continuously running parts during system use, while ② is the part that participates in the work at any time before the system is activated and during operation.

[0073] The above-mentioned data acquisition unit 1 to feedback unit 6, these six components, can be designed as a tightly integrated circuit. The following division is based on the different functions performed in the system:

[0074] Data acquisition unit 1: may include wearable hardware sensors such as EEG helmets and fitness trackers to collect user physiological data and convert it into digital signals that can be processed by the system.

[0075] Temporary storage unit 2 includes a preprocessing module 21, a first storage module 22, and a first output module 23. The preprocessing module 21 performs filtering, alignment, and segmentation operations on the collected data, processing the physiological data into a standard data format for use as input to the neural network.

[0076] The first storage module 22 stores the preprocessed input data and temporary data in the neural network model calculation, such as the activation values, gradients, and post-processing results of intermediate layers.

[0077] The first output module 23 sends the data sequentially to the in-memory computing unit 4 according to the neural network deployment method until the entire network calculation is completed and the data classification result is obtained.

[0078] In-memory computing unit 4: A specially designed in-memory computing hardware, specifically implemented using, but not limited to, SRAM, RRAM, DRAM, etc. In-memory computing unit 4 stores a portion of the network weight data received from configuration unit 3 for subsequent multiplication and accumulation calculations. After receiving data input from temporary storage unit 2 and completing the calculation of network weights, the result is sent to post-processing unit 5.

[0079] Configuration unit 3 includes a second storage module 31 and a second output module 32. It receives initial parameters of the neural network model from an external device and configures the computational function modules in the in-memory computation unit 4. The computational function modules contain parameters of the entire neural network model, such as the weight data of each layer. The second output module 32 adapts to the data transmitted from the temporary storage unit 2 and configures the corresponding weight matrix required for computation in the in-memory computation unit 4.

[0080] Post-processing unit 5 includes a calculation module 51 and a classification module 52. The calculation module 51 receives the data output from the in-memory calculation unit 4, performs common neural network calculations such as activation, regularization, pooling, and residuals, and chooses to write these results back to the temporary storage unit 2 to await the next part of the operation or send them to the classification module 52 according to the overall network inference progress. The classification module 52 classifies the neural network calculation results for specific tasks, and can use methods including but not limited to maximum value comparison or other classification methods.

[0081] Feedback Unit 6: Sends the classification results back to the user.

[0082] The following is an explanation at the system execution level:

[0083] The transmission of physiological data information in the system can be divided into three parts: interaction between the user and the device, parameter configuration and updating of the neural network model, and data calculation and inference.

[0084] During use, the system continuously acquires one or more types of physiological signals from the user through the acquisition unit, such as EEG and heart rate. These signals are preprocessed and converted into a standard data format, serving as input for the neural network inference calculations within the system. After classification and recognition by the model, the feedback unit performs specific functions, including controlling external hardware devices and issuing corresponding medical warnings, to achieve the user's preset goals.

[0085] The neural network model can be trained outside the system using external datasets, including but not limited to host computers and central servers. For a deep learning neural network model with sufficient inference capability and recognition accuracy, it can be deployed to the system's configuration unit (CIM) via wired or wireless means, storing the parameters of the entire neural network model. During system operation, each part of the model is calculated sequentially within the CIM unit. At any point during operation, a new neural network model can be redeployed through the configuration unit. Optionally, with appropriate hardware design, backpropagation and training of the neural network can be completed simultaneously with inference, enabling the system to spontaneously update the model during operation.

[0086] In system operation, neural network parameters and input data are stored in the temporary storage unit and configuration unit, respectively. They are stored and transmitted sequentially using methods including, but not limited to, layer-by-layer and data-stream-by-data method. Large-scale multiplication and accumulation calculations are primarily performed in the CIM unit, followed by activation, normalization, and pooling operations—common neural network operations—in the post-processing unit. The calculation process involves one or more loops throughout the entire computation, calculating one or more layers of the overall neural network each time, until all layers have completed their inference. The final result is the neural network's classification of the physiological data, which is transmitted to the feedback unit to complete subsequent specific tasks, such as tasks related to a specific motor imagery. The specific execution steps are as follows:

[0087] s1. Using the existing physiological information dataset of the corresponding task, train a general neural network model W with sufficient reasoning ability and recognition accuracy. The model can be divided into three layers {W1, W2, W3}.

[0088] s2. Deploy W to the user's local device via the host PC and store it in the storage module of the configuration unit.

[0089] s3. The user uses a local device to collect electroencephalographic signal data and convert it into corresponding signals.

[0090] s4. After filtering, alignment and other preprocessing operations, the signal is divided into data segments D of fixed time length. The original data D is stored in the storage module of the temporary storage unit.

[0091] s5. Select the first layer parameter W1 of the model in the configuration unit and send it to the CIM unit.

[0092] s6. The temporary storage unit sends D to the CIM unit, and completes the calculation with W1 to obtain the intermediate result R1.

[0093] s7.R1 is sent to the post-processing unit to complete the calculation and obtain the first layer neural network result Y1, which is then sent back to the temporary storage unit.

[0094] In step s8, select the second-level parameter W2 of the model in the configuration unit and send it to the CIM unit.

[0095] s9. The temporary storage unit sends Y1 to the CIM unit, and together with W2, completes the calculation to obtain the intermediate result R2.

[0096] s10.R2 is sent to the post-processing unit to complete the calculation and obtain the second layer neural network result Y2, which is then sent back to the temporary storage unit.

[0097] s11. Select the third layer parameter W3 of the model in the configuration unit and send it to the CIM unit.

[0098] s12. The temporary storage unit sends Y2 to the CIM unit, and together with W3, completes the calculation to obtain the intermediate result R3.

[0099] s13.R3 is sent to the post-processing unit to complete the calculation and obtain the overall reasoning result Y3, which is then sent to the classification module.

[0100] s14. Calculate the final result T based on the classification target of motion imagery and output the feedback unit.

[0101] s15. The feedback unit controls the corresponding mechanical prosthesis to complete the corresponding action based on the final result T, thereby achieving the movement task imagined by the user.

[0102] Physiological data possesses strong privacy and personal characteristics. Furthermore, tasks related to physiological data analysis are time-constrained, and methods using cloud-based and external devices for deep learning neural network computation have limitations and risks. Traditional hardware-based local computation is unsuitable for continuous daily use due to energy consumption and hardware scalability issues. This solution proposes a physiological data classification system utilizing in-memory computing technology. This system achieves low-power, high-efficiency local neural network computation, is hardware-friendly, facilitates lightweight wearable devices, and can operate continuously for extended periods, providing efficient and accurate physiological data classification results.

[0103] Hardware devices utilizing in-memory computing technology also offer strong scalability. Optionally, sparsity in neural networks can be achieved by designing corresponding hardware components within the in-memory computing unit, further reducing the energy requirements and time consumption for inference; or encrypted computing can be implemented using the computational characteristics of the in-memory computing unit to ensure the privacy and security of physiological data from within the device.

[0104] The physiological data classification system proposed in this solution, which utilizes in-memory computing technology, has advantages such as high speed, low power consumption, high reliability, hardware optimization, and strong scalability. It can be applied to various types of physiological data analysis and various physiological health-related tasks. It is suitable for application scenarios that emphasize data privacy, task timeliness, and hardware convenience, and has great development potential.

[0105] The data classification and processing system based on neural networks provided in this invention includes a data acquisition unit, a temporary storage unit, a configuration unit, an in-memory calculation unit, and a post-processing unit. Specifically: the data acquisition unit collects user physiological data and converts it into digital signals; the temporary storage unit receives and stores the digital signals and the intermediate output results of the neural network returned by the post-processing unit; the configuration unit receives initial neural network parameters input from an external device and the intermediate output results of the neural network returned by the post-processing unit, and generates configuration parameters for in-memory calculation; the in-memory calculation unit receives the configuration parameters and the digital signals, performs in-memory calculations on the configuration parameters and the digital signals, and obtains intermediate calculation results; the post-processing unit receives the intermediate calculation results and performs post-processing calculations on the intermediate calculation results to obtain intermediate output results of the neural network and neural network classification results based on a preset task, thereby improving the computational efficiency of data classification and processing.

[0106] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0107] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0108] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0109] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0110] In the description of this specification, the references to terms such as "an embodiment," "a specific embodiment," "some embodiments," "for example," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0111] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A neural network-based data classification processing system, characterized by, It includes a data acquisition unit, a temporary storage unit, a configuration unit, an in-memory computing unit, and a post-processing unit; wherein: The data acquisition unit is used to collect user physiological data and convert the user physiological data into digital signals; The temporary storage unit is used to receive and save digital signals and the intermediate output results of the neural network returned by the post-processing unit; The configuration unit is used to receive the initial parameters of the neural network input by the external device and the intermediate output results of the neural network returned by the post-processing unit, and generate configuration parameters for in-memory computation. The in-memory calculation unit is used to receive configuration parameters and digital signals, and perform in-memory calculations on the configuration parameters and digital signals to obtain intermediate calculation results; The post-processing unit is used to receive intermediate calculation results, perform post-processing calculations on the intermediate calculation results to obtain intermediate output results of the neural network, and send the intermediate output results of the neural network back to the temporary storage unit and the configuration unit; the configuration unit is used to generate configuration parameters that are compatible with the current output data of the temporary storage unit according to the returned intermediate output results, so as to drive the in-memory computing unit to perform in-memory computing in a loop by layer or by data flow until the neural network classification result based on the preset task is obtained.

2. The data classification and processing system based on neural networks according to claim 1, characterized in that, The temporary storage unit includes a preprocessing module, a first storage module, and a first output module; wherein: The preprocessing module is used to preprocess digital signals; The first storage module is used to save the preprocessed digital signals and intermediate output results of the neural network; The first output module is used to send the preprocessed digital signal and the intermediate output result of the neural network to the in-memory computing unit.

3. The data classification and processing system based on neural networks according to claim 2, characterized in that, The configuration unit includes a second storage module and a second output module; wherein: The second storage module is used to store neural network parameters; The second output module is used to send configuration parameters adapted to the output data of the temporary storage unit to the in-memory computing unit.

4. The data classification and processing system based on neural networks according to claim 1, characterized in that, The post-processing unit includes a calculation module and a classification module; wherein: The calculation module is used to perform post-processing calculations on the intermediate calculation results; The classification module is used to receive the post-processing calculation results and classify the user's physiological data based on a preset task to obtain the neural network classification results.

5. The data classification and processing system based on neural networks according to claim 1, characterized in that, The neural network-based data classification and processing system also includes a feedback unit; The feedback unit is used to receive the neural network classification results and feed them back to the user.

6. The data classification and processing system based on neural networks according to claim 1, characterized in that, The data acquisition unit includes wearable hardware sensors.

7. The data classification and processing system based on neural networks according to claim 2, characterized in that, The preprocessing module is specifically used for: The digital signal is filtered, aligned, and segmented to obtain data in a preset standard format.

8. The data classification and processing system based on neural networks according to claim 1, characterized in that, The in-memory computing unit includes one of SRAM, RRAM, and DRAM.

9. The data classification and processing system based on neural networks according to claim 4, characterized in that, The calculation module is specifically used for: The intermediate calculation results are activated, regularized, pooled, and residuals are calculated.

10. The data classification and processing system based on neural networks according to claim 5, characterized in that, The feedback unit is specifically used for: Based on the classification results, determine the control feedback instructions for external devices and / or issue warning messages.

Images