Distributed cognitive decision-making method and device based on SNN model, equipment and medium

Abstract

The invention relates to the technical field of machine control, and discloses a distributed cognitive decision-making method based on an SNN model, and the method comprises the steps: constructing the SNN model in advance, and configuring the initial parameters of the SNN model; acquiring environmental data, and encoding the environmental data to generate a pulse stream; the pulse flow is input into an SNN model, the SNN model carries out calculation based on an event-driven mechanism, and signal processing features are output; determining a local cognitive decision according to the signal processing characteristics; transmitting the local cognitive decision to an edge node for feature matching and verification through a cooperative communication module, and feeding back a verification signal; and generating a final global decision through a global fusion module according to the verification signal fed back by each edge node. According to the invention, the changed environmental data is processed through the pre-constructed SNN model, the pulse stream is generated, the data matching and verification are carried out according to the pulse stream, the final global decision is generated according to the feedback verification signal, the redundant computing power is reduced, and the system power consumption is reduced.

Description

Technical Field

[0001] This invention relates to the field of machine control technology, and in particular to a distributed cognitive decision-making method, apparatus, device, and storage medium based on an SNN model. Background Technology

[0002] With the continuous evolution of Internet of Things (IoT), edge computing, and smart terminal technologies, the industry's demand for low-power, high-real-time, and distributed intelligent decision-making systems is becoming increasingly urgent. This demand is widely applied across mobile terminals, wearable devices, drones, and distributed edge nodes. However, current mainstream large-scale deep learning models, such as those based on Artificial Neural Networks (ANNs) or Convolutional Neural Networks (CNNs), while exhibiting outstanding perception and decision-making capabilities, have several shortcomings in meeting these specific needs.

[0003] From a computational perspective, traditional neural networks employ synchronous computation mechanisms, relying heavily on multiply-accumulate operations and requiring global synchronous updates. This characteristic makes them extremely unsuitable for power-sensitive scenarios. Not only does this place an excessive burden on the computing power and energy consumption of local devices, but in multi-node collaborative environments, the massive amount of intermediate data transmission leads to severe communication redundancy, further exacerbating power consumption pressures. Furthermore, current mainstream deep learning systems (such as PyTorch and TensorFlow frameworks running on GPUs / TPUs) generally adopt a centralized training and inference model. This model completely fails to meet the needs of distributed independent cognition and local autonomous decision-making in resource-constrained environments, resulting in a lack of lightweight, autonomous intelligent behavior modeling solutions for multi-node collaboration.

[0004] More importantly, traditional neural networks have failed to draw on the core processing characteristics of the human brain: on the one hand, they ignore the brain's temporal processing capabilities and cannot effectively simulate the brain's efficient information processing mechanisms such as sparse coding and event-driven processing, making their energy efficiency far inferior to that of the human brain and making it difficult to balance low power consumption and high performance; on the other hand, existing models cannot effectively model brain-inspired mechanisms such as temporal dynamics, asynchronous processing, and event response, resulting in a low overall computing power and energy efficiency ratio of the system, making it difficult to meet practical application standards in scenarios with extremely high requirements for real-time perception and rapid response, such as intelligent security, edge monitoring, and emergency response systems. Summary of the Invention

[0005] The main objective of this invention is to provide a distributed cognitive decision-making method, device, equipment, and storage medium based on an SNN model, aiming to solve the problems of high power consumption and difficult distributed deployment of traditional neural networks in the prior art.

[0006] To achieve the above objectives, this invention provides a distributed cognitive decision-making method based on an SNN model, comprising: Pre-build the SNN model and configure its initial parameters; Acquire environmental data, encode the environmental data, and generate a pulse stream; The pulse stream is input into the SNN model, which calculates based on an event-driven mechanism and outputs signal processing features. Determine local cognitive decisions based on signal processing characteristics; The local cognitive decision is transmitted to the edge node for feature matching and verification through the collaborative communication module, and a verification signal is fed back. Based on the verification signals fed back by each edge node, the final global decision is generated through the global fusion module.

[0007] Furthermore, to achieve the above objectives, the present invention provides a distributed cognitive decision-making device based on an SNN model, comprising: The model building module is used to pre-build the SNN model and configure the initial parameters of the SNN model; The data encoding module is used to acquire environmental data, encode the environmental data, and generate a pulse stream; The pulse stream calculation module is used to input the pulse stream into the SNN model, which calculates based on an event-driven mechanism and outputs signal processing features. The cognitive decision-making module is used to determine local cognitive decisions based on signal processing characteristics; The data verification module is used to transmit the local cognitive decision to the edge node for feature matching and verification through the collaborative communication module, and to provide a verification signal. The final decision module is used to generate the final global decision based on the verification signals fed back by each edge node through the global fusion module.

[0008] Furthermore, to achieve the above objectives, the present invention also provides a computer device, the computer device including a memory, a processor, and a distributed cognitive decision-making program based on an SNN model stored in the memory and executable on the processor, wherein when the distributed cognitive decision-making program based on an SNN model is executed by the processor, it implements the steps of the distributed cognitive decision-making method based on an SNN model as described above.

[0009] Furthermore, to achieve the above objectives, the present invention also provides a computer-readable storage medium storing a distributed cognitive decision-making program based on an SNN model, wherein when the distributed cognitive decision-making program based on an SNN model is executed by a processor, it implements the steps of the distributed cognitive decision-making method based on an SNN model as described above.

[0010] Beneficial Effects: This invention relates to the field of machine control technology and discloses a distributed cognitive decision-making method based on an SNN model, comprising: pre-constructing an SNN model and configuring its initial parameters; acquiring environmental data, encoding the environmental data, and generating a pulse stream; inputting the pulse stream into the SNN model, whereby the SNN model performs calculations based on an event-driven mechanism and outputs signal processing features; determining a local cognitive decision based on the signal processing features; transmitting the local cognitive decision to edge nodes for feature matching and verification via a collaborative communication module, and feeding back verification signals; and generating a final global decision based on the verification signals fed back by each edge node through a global fusion module. This invention processes changing environmental data using a pre-constructed SNN model to generate a pulse stream, performs data matching and verification based on the pulse stream, and generates a final global decision based on the feedback verification signals, thereby reducing redundant computing power and lowering system power consumption. Attached Figure Description

[0011] The present invention will be further described below with reference to the accompanying drawings and embodiments. In the accompanying drawings: Figure 1 This is a schematic diagram of an application environment for a distributed cognitive decision-making method based on an SNN model, according to one embodiment of the present invention. Figure 2 This is a flowchart illustrating an embodiment of the distributed cognitive decision-making method based on the SNN model of the present invention. Figure 3 This is a schematic diagram of the functional modules of a preferred embodiment of the distributed cognitive decision-making device based on the SNN model of the present invention. Figure 4 This is a schematic diagram of the structure of a computer device according to an embodiment of the present invention; Figure 5 This is another structural schematic diagram of a computer device according to one embodiment of the present invention. Detailed Implementation

[0012] It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0013] The distributed cognitive decision-making method based on the SNN model provided in this invention can be applied to, for example... Figure 1In this application environment, the user terminal communicates with the server via a network. The server can pre-build an SNN model from the user terminal and configure its initial parameters; acquire environmental data, encode the data, and generate a pulse stream; input the pulse stream into the SNN model, which performs calculations based on an event-driven mechanism and outputs signal processing features; determine local cognitive decisions based on these features; transmit the local cognitive decisions to edge nodes for feature matching and verification via a collaborative communication module, and provide feedback verification signals; and generate a final global decision based on the verification signals from each edge node via a global fusion module. This invention processes changing environmental data using a pre-built SNN model to generate a pulse stream, performs data matching and verification based on the pulse stream, and generates a final global decision based on the feedback verification signals, reducing redundant computing power and lowering system power consumption. The user terminal can be, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices. The server can be implemented using a standalone server or a server cluster consisting of multiple servers. The invention will be described in detail below through specific embodiments.

[0014] Please see Figure 2 , Figure 2 This is a flowchart illustrating an embodiment of the distributed cognitive decision-making method based on an SNN model provided by the present invention. It should be noted that although the logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than that shown here.

[0015] like Figure 2 As shown, the distributed cognitive decision-making method based on the SNN model proposed in this invention includes the following steps: S100. Pre-build the SNN model and configure the initial parameters of the SNN model; S200: Acquire environmental data, encode the environmental data, and generate a pulse stream; S300. The pulse stream is input into the SNN model, which calculates based on an event-driven mechanism and outputs signal processing features. S400, Determine local cognitive decisions based on signal processing characteristics; S500: The local cognitive decision is transmitted to the edge node for feature matching and verification through the collaborative communication module, and a verification signal is fed back. S600 generates the final global decision through the global fusion module based on the verification signals fed back by each edge node.

[0016] In this embodiment, pre-constructing the SNN model and configuring initial parameters is the fundamental preparatory step for the entire decision-making method. First, the model needs to be designed based on the computing power of the actual deployed hardware platform. For example, on different platforms such as the Intel Loihi neuromorphic chip, Raspberry Pi+FPGA, or STM32 MCU, an appropriate neuron model (such as the Leaky Integrate-and-Fire (LIF) model or the Izhikevich neuron model) and learning mechanism (such as STDP (Spike-Timing Dependent Plasticity) or Hebbian rules) should be selected to ensure the model can run efficiently on the target hardware. Next, according to function or perceptual modality, the complete SNN model is broken down into multiple sub-networks, such as the SNN-Vis sub-network responsible for visual recognition, the SNN-Audio sub-network for processing sound signals, the SNN-Gas sub-network for detecting gas information, and the SNN-Fusion decision fusion sub-network that summarizes the results of each sub-network. Each sub-network is deployed on different edge nodes; for example, SNN-Vis is deployed on an edge node with a camera, and SNN-Audio is deployed on an edge node connected to a microphone. Finally, initial parameters are configured for each subnetwork, including initial synaptic weights, neuron firing thresholds, synaptic pruning parameters (such as correlation coefficients for judging synaptic importance and pruning thresholds) and pulse firing threshold adjustment coefficients for optimizing power consumption. At the same time, an event-aware mapping table is constructed for each subnetwork to clarify the types of events it can handle, pulse map template matching conditions, and fast path loading mechanisms, in preparation for subsequent model operation.

[0017] Acquiring environmental data and encoding it to generate pulse streams is a crucial step in transforming real-world environmental information into signals that can be processed by an SNN model. This process relies on multiple sensors, such as cameras, microphones, temperature sensors, and smoke sensors. These sensors monitor the surrounding environment in real time, but do not continuously collect data. Data acquisition only begins when changes in environmental data are detected, such as temperature fluctuations, the appearance of new sound signals, or edge changes in images. When there are no changes, they remain in a low-power sleep state to reduce unnecessary energy consumption. After acquiring the environmental data, it is first classified, and then the corresponding encoding method is selected according to the different data types to convert it into a pulse stream that the SNN model can recognize. Specifically, temperature and gas data use threshold change encoding, for example, triggering a pulse for every 1°C increase in temperature; image data uses Difference-of-Gaussian (DoG) pulse encoding, generating pulses at image edges; sound data undergoes frequency domain analysis first, and then Poisson encoding is used to trigger pulse emission in the high-frequency energy range. Through this differentiated encoding, it is ensured that different types of environmental information can be accurately converted into pulse signals.

[0018] The core step in data processing is inputting the pulse stream into the SNN model and calculating the output signal processing features based on the event-driven mechanism. The pulse stream is first transmitted to the local SNN model of the corresponding edge node. The Event Aware Map (EAM) in the model first matches the pulse features to determine if the pulse stream matches the event category and pulse pattern template that the current sub-network can process. If the match is successful, the corresponding sub-network and the most responsive group of neurons (i.e., the Top-K neuron group) are quickly activated, while irrelevant modules and neuron groups remain silent to avoid redundant computation. If the match fails, the sub-network does not start computation and continues in a low-power state. The activated sub-network then performs computation based on the event-driven mechanism, processing the pulse stream through neuronal integration (such as the accumulation process of membrane potential in the LIF model) and synaptic transmission (combined with the initially configured synaptic weights). Simultaneously, it utilizes synaptic plasticity mechanisms (such as STDP) to fine-tune the synaptic weights online, continuously optimizing the model's processing capabilities. Ultimately, it generates signal processing features that reflect key information about environmental events, such as identifying feature signals corresponding to events like "fire" or "voice commands."

[0019] Determining local cognitive decisions based on signal processing features is a crucial step for edge nodes to achieve autonomous cognition. Each edge node independently makes a preliminary judgment on current environmental events based on the signal processing features output by its local SNN model, combined with its own preset decision rules. For example, if the signal processing features output by the SNN-Vis sub-network match the "flame" feature, the edge node will determine that there is a fire risk; if the features output by the SNN-Audio sub-network match the features of the "start movement" voice command, it will determine that a movement control command needs to be triggered. For judgments of ordinary events (such as normal temperature fluctuations or meaningless background noise), the edge node only records relevant information locally and does not initiate subsequent collaborative communication; however, when the judgment result is a critical event (such as a danger warning or an important control command), the edge node will trigger the next collaborative verification process to further confirm the authenticity of the event and formulate a global response strategy.

[0020] Transmitting local cognitive decisions and acquiring verification signals via a collaborative communication module is a crucial step in achieving multi-node collaborative judgment. When an edge node determines that a local cognitive decision is a critical event, it encapsulates the decision into a timestamped pulse packet. This pulse packet contains key information such as the source node identifier, event timestamp, event weight, and decision label, facilitating collaborative judgment by other nodes. Subsequently, the timestamped pulse packet is broadcast to other edge nodes via a collaborative communication module (such as a LoRa / mesh network or Spike Message Bus (SMB) mechanism). This communication process only occurs when a critical event is triggered; otherwise, the communication module remains off to conserve energy. Upon receiving the pulse packet, other edge nodes perform feature matching and verification against the critical event information and their own local sensor-collected environmental data and SNN model processing results. If the local data supports the critical event, it sends back a "verification passed" pulse signal; if the local data contradicts the critical event or cannot clearly support the event, it sends back a "pending confirmation" pulse signal, forming a multi-node collaborative verification feedback for the critical event.

[0021] The final global decision is generated through the global fusion module based on the verification signals fed back from each edge node. This final step in the entire decision-making process ensures the accuracy and reliability of the decision. The global fusion module first collects the verification signals from all edge nodes, and then selects an appropriate decision-making mechanism based on the actual application scenario to generate the final global decision. If a majority voting mechanism is selected, the percentage of nodes that return a "verification passed" signal will be tallied. When this percentage exceeds a preset threshold (e.g., 60%), the critical event is deemed valid, and a decision is generated to trigger corresponding response actions (e.g., activating an alarm system or controlling execution devices). If a confidence-weighted mechanism is used, a confidence weight will be assigned to each node based on factors such as historical decision accuracy, hardware performance, and deployment environment stability. Then, a total confidence score (the sum of the products of each node's confidence weight and its corresponding signal value, with a "verification passed" signal value set to 1 and a "pending confirmation" signal value set to 0) will be calculated. When the total score exceeds a preset threshold equal to the sum of the total weights, the event is deemed valid, and a decision is generated. If a master node coordination mechanism is selected, a pre-designated master node (usually a node with strong computing power and low communication latency) will combine its own local SNN model's judgment of the critical event with the verification signals from other nodes to directly generate the final global decision. The generated global decision will be transmitted to the corresponding execution module to trigger specific response actions and will also be fed back to each edge node to update the node's local SNN model parameters, providing better support for subsequent event processing.

[0022] In one embodiment, step S100 includes: S101. Configure the neuron model and learning mechanism for the SNN model based on the platform's computing power; S102. Divide the SNN model into several sub-networks according to function or perception modality; S103. Configure initial synaptic weights, neuron firing thresholds, synaptic pruning parameters, and pulse firing threshold adjustment coefficients for each sub-network, and construct an event perception table for the sub-network.

[0023] In this embodiment, configuring the neuron model and learning mechanism for the SNN model according to the platform's computing power is a fundamental prerequisite for ensuring the model's efficient operation on the target hardware. This requires close integration of hardware performance and application scenario requirements. The hardware platforms explicitly adapted in this invention include neuromorphic chips (such as Intel Loihi and SpiNNaker), resource-constrained MCUs (such as STM32), and Raspberry Pi+FPGA, etc., with significant differences in computing power across different platforms. For example, Intel Loihi supports on-chip SNN execution with low power consumption, making it suitable for deploying complex neuron models; STM32 MCUs have limited computing power, requiring the selection of lightweight models. Therefore, the neuron model must be selected specifically during configuration: on neuromorphic chip or FPGA platforms with strong computing power, the Leaky Integrate-and-Fire (LIF) model or the Izhikevich neuron model can be used. These models can combine synaptic delay and plasticity to achieve precise timing processing, adapting to complex environmental perception tasks; on resource-constrained MCU platforms, the neuron model structure can be simplified, prioritizing basic signal processing capabilities. The choice of learning mechanism also depends on the platform's computing power. Nodes generally use STDP (Spike-Timing Dependent Plasticity) or Hebbian rules for online updates. Among them, STDP rules can dynamically adjust synaptic weights based on the timing of spike firing, which is more in line with bio-inspired intelligent mechanisms and is suitable for collaborative learning on platforms with sufficient computing power. Hebbian rules have relatively simple logic and can be quickly deployed on platforms with limited computing power, ensuring that the model has basic adaptive capabilities. The overall configuration needs to balance model performance and hardware carrying capacity to avoid hardware resource overload due to excessive model complexity.

[0024] Decomposing an SNN model into several sub-networks based on function or perception modality is a core design for achieving distributed cognitive decision-making. This allows different modules to focus on processing specific types of information, improving computational efficiency and system flexibility. The decomposition strategy must be tailored to the perception and decision-making needs of the actual application scenario. Common sub-network divisions include visual recognition sub-networks (SNN-Vis), sound recognition sub-networks (SNN-Audio), gas sensing sub-networks (SNN-Gas), and decision fusion sub-networks (SNN-Fusion). Each sub-network has a clear functional positioning and deployment location: SNN-Vis processes image information, deployed at edge nodes connected to cameras, taking image pulse-coded streams as input and outputting target labels such as objects and flames; SNN-Audio focuses on sound signal processing, deployed at nodes equipped with microphones, recognizing voice commands or alarm statuses through audio pulse sequences; SNN-Gas detects changes in gas concentration, deployed at gas detection edge nodes, determining the presence of smoke or toxic gases based on odor concentration pulses; SNN-Fusion, as the decision core, is deployed at upper-layer aggregation or collaboration nodes, receiving intermediate outputs from each sub-network and generating the global decision result. This decomposition method allows each sub-network to operate independently, activating the corresponding module only when needed, reducing redundant computation, and facilitating the optimization of the model structure for the functional requirements of different sub-networks. For example, it can enhance the spatial feature processing capability of SNN-Vis and improve the temporal signal parsing efficiency of SNN-Audio, which aligns with the innovative point of this invention: "event-driven distributed SNN architecture".

[0025] Configuring initial parameters and building an event-aware table for each sub-network is crucial for ensuring accurate response to target events and low-power operation. Initial parameter configuration must cover the core elements of sub-network operation: initial synaptic weights should be preset according to the sub-network's function; for example, in SNN-Vis, the initial weights of synapses responsible for edge detection can be set to highlight image gradient changes, while in SNN-Audio, the weights of synapses corresponding to the speech command frequency band can be appropriately increased; neuron firing thresholds need to be set in conjunction with the signal strength range processed by the sub-network to avoid frequent false triggers due to excessively low thresholds or missed signal detection due to excessively high thresholds; synaptic pruning parameters, including correlation coefficients and pruning thresholds for judging synaptic importance, need to be initially determined during model training to lay the foundation for subsequent optimization of redundant connections and reduction of energy consumption; the impulse firing threshold adjustment coefficient is used to dynamically adapt to changes in input signal strength, increasing the threshold to reduce meaningless impulse responses at high input strengths. Meanwhile, the event-aware table (event-aware mapping table) constructed for each sub-network must contain three core components: the types of events that can be processed (such as "flame recognition" and "target detection" in SNN-Vis), pulse map template matching conditions (such as pulse sequences of specific time patterns), and a fast path loading mechanism (prioritizing the activation of synaptic groups related to the current event). The event-aware table allows the sub-network to quickly determine whether it matches its processing range after receiving a pulse stream, and only initiates forward propagation when the activation conditions are met, without waking up irrelevant modules, significantly reducing standby power consumption, and perfectly aligning with the technical principles of "event-driven" and "pulse sparsity optimization" of this invention.

[0026] In one embodiment, step S200 includes: S201. Monitor environmental data through multi-source sensors, and collect environmental data when the environmental data changes; S202. Classify the environmental data and select the encoding method of the environmental data according to the environmental type; S203. Encode the environmental data using an encoding method to generate a pulse stream.

[0027] In this embodiment, monitoring environmental data through multi-source sensors and collecting data as it changes is a fundamental step in the entire decision-making system's acquisition of environmental information, strictly adhering to the "event-driven" principle to achieve low power consumption. The multi-source sensors encompass various types, configured according to the application scenario. For example, in a forest fire monitoring scenario, temperature sensors, smoke sensors, and image sensors (cameras) are deployed; in an elderly companion robot scenario, accelerometers, microphones, and temperature sensors are used. These sensors do not continuously collect data but are in a real-time monitoring state, only initiating data collection when changes in environmental data within the monitoring range are detected. For example, a temperature sensor detects a temperature fluctuation exceeding a preset range, an image sensor captures a significant change in the edge of the image or color, or a microphone detects a new sound signal. When there are no data changes, the system remains in a low-power sleep mode. This data collection method minimizes unnecessary energy consumption, aligning with the core requirement of "low-power operation" in this invention, while ensuring that data is acquired only when the environment is abnormal or undergoes critical changes, providing effective information support for subsequent decision-making.

[0028] Classifying the collected environmental data and selecting the encoding method according to the type is a key step in transforming real-world environmental information into signals that can be processed by the SNN model. The classification is mainly based on the data's perceptual modality and feature attributes. Common data classifications include temperature / gas data (such as ambient temperature, smoke concentration, and toxic gas content), image data (such as scene images and target object images), and sound data (such as voice commands and abnormal sounds). Different types of environmental data have different characteristics, requiring corresponding encoding methods to accurately retain key information and adapt to the pulse signal processing logic of SNN models. For example, temperature / gas data are continuously changing physical quantities, suitable for "threshold change encoding," where a pulse is generated every time the data reaches a preset change threshold (e.g., temperature increases by 1°C, gas concentration increases by a certain unit), and the pulse frequency reflects the rate of data change. Image data needs to highlight spatial features, so "Difference-of-Gaussian (DoG) pulse encoding" is used to generate pulses at key feature locations such as image edges and textures, reducing the transmission and processing of redundant pixel information. Sound data requires analysis of frequency and energy characteristics. First, effective frequency bands (such as the high-frequency energy band of voice commands) are extracted through frequency domain analysis, and then "Poisson encoding" is combined to trigger pulse emission in the high-frequency energy band, so that the pulse sequence can reflect the temporal and energy changes of sound. This differentiated encoding strategy ensures that the key information of different types of environmental data can be accurately converted into pulse signals.

[0029] Encoding environmental data using a selected encoding method to generate a pulse stream is the final step in completing the "environmental information → pulse signal" conversion. The generated pulse stream will serve as the input data for the SNN model. During the encoding process, various sensors, based on their own characteristics and preset encoding rules, transform the collected analog or digital environmental data into discrete pulse sequences (i.e., pulse streams). Taking temperature data encoding as an example, if a pulse is triggered every 1°C increase in temperature, when the sensor detects the temperature gradually rising from 25°C to 30°C, it will generate 5 pulses sequentially, and the pulse interval can reflect the rate of temperature increase (shorter intervals for faster increases and longer intervals for slower increases). After DoG encoding, image data generates pulses only at the edges of objects and the boundary between light and dark areas in the image, forming a pulse stream that outlines the object's contour. After frequency domain analysis and Poisson encoding, key syllables in voice commands or high-frequency parts of abnormal sounds are converted into dense pulses, while background noise generates very few pulses or even no pulses. The generated pulse stream not only retains the core characteristics of the environmental data (such as trends, spatial structure, and temporal patterns) but also possesses the characteristic of "sparseness." Pulses are generated only when key information appears, and there are very few pulses for redundant information. This characteristic can significantly reduce the computational load of subsequent SNN models, further reduce system power consumption, and at the same time ensure that the model can quickly focus on key environmental information, improve decision-making efficiency, and fully conform to the technical principles of "pulse sparsity optimization" and "event-driven information processing" in this invention.

[0030] In one embodiment, step S300 includes: S301. Input the pulse stream into the SNN model and match the pulse features through the event-aware mapping table in the SNN model; S302. If the pulse feature matches the event type in the sub-network event perception table with the pulse map template, then the corresponding sub-network and neuron group are activated. S303: The activated subnetwork processes pulse flows through neuronal integration and synaptic transmission, generating signal processing features.

[0031] In this embodiment, after the pulse stream is input into the SNN model, pulse feature matching is first performed through the event-aware mapping table in the model. This is the crucial first step in realizing event-driven computation. The event-aware mapping table is a core functional module pre-built for each sub-network. It internally stores the event categories that the sub-network can handle, the corresponding pulse map template matching conditions, and information related to the fast path loading mechanism. For example, the SNN-Vis sub-network responsible for visual recognition records processable events such as "flame recognition" and "target detection" in its event-aware mapping table, and also stores the corresponding pulse map templates (such as specific pulse timing patterns formed by encoding flame images). When the pulse stream is input, the event-aware mapping table compares the features of the pulse stream (such as pulse frequency, timing arrangement, and energy distribution) with the template one by one to determine whether the current pulse stream belongs to the event category that the sub-network can handle. This process strictly follows the "sub-network fast activation strategy" in this invention to ensure that subsequent computation is only started when the match is successful, avoiding meaningless energy consumption.

[0032] If the event-aware mapping table determines that the impulse feature matches the event type and impulse map template in the sub-network's event-aware table, the system will immediately activate the corresponding sub-network and specific neuron groups, while keeping other unrelated sub-networks and neuron groups silent. Here, "activation" does not mean starting all the functions of the sub-network, but rather accurately locating the event-related functional modules and neuron groups based on the impulse feature. For example, when the SNN-Audio sub-network's event-aware mapping table matches the impulse feature of "voice command 'start'", only the neuron group responsible for voice command parsing will be activated, not the entire voice recognition sub-network. This precise activation strategy relies on the fast path loading mechanism in the event-aware mapping table, which can prioritize waking up synapses related to the current event, shortening response latency, and minimizing redundant computation. This perfectly aligns with the core innovation of the "event-driven distributed SNN architecture" in this invention, achieving a balance between low power consumption and high real-time performance.

[0033] The activated subnetwork processes the pulse stream through two core processes: neuronal integration and synaptic transmission, ultimately generating signal processing features. In the neuronal integration stage, activated neurons accumulate and integrate the input pulse signals according to their own characteristics (such as using the LIF model or the Izhikevich model). For example, in the LIF model, neurons gradually accumulate the membrane potential changes caused by the pulses, and only generate a firing response when the membrane potential reaches a preset threshold. This process simulates how neurons in the human brain process temporal signals, effectively capturing the dynamic features in the pulse stream. In the synaptic transmission stage, neurons transmit firing signals through synaptic connections. During transmission, pre-configured synaptic weights (some weights may have been optimized through pruning) are used to adjust the signal strength, highlighting the influence of key pulse information and suppressing irrelevant signal interference. Through the synergistic effect of these two processes, the subnetwork can extract key information reflecting the essence of the event from the pulse stream, forming signal processing features. For example, in a fire monitoring scenario, the sub-network extracts the combined features of "continuous high temperature + high smoke concentration" from temperature and smoke-related pulse streams, providing a basis for subsequent local cognitive decision-making. The entire processing process fully embodies the technical principle of "SNN local computation" in this invention, taking into account both processing accuracy and energy consumption control.

[0034] In one embodiment, step S500 includes: S501. When the local cognitive decision is a critical event, the local cognitive decision is packaged into a pulse packet with a timestamp. S502. The timestamped pulse packet is transmitted to the edge node via the collaborative communication module; S503, The edge node receives pulse packets with timestamps for feature matching and verification, and feeds back a verification signal.

[0035] In this embodiment, when a local cognitive decision generated by an edge node through its local SNN model is identified as a critical event, the decision must first be packaged into a timestamped pulse packet. This step is fundamental to ensuring accurate information transmission and traceability in distributed collaborative decision-making. Here, "critical events" typically refer to events with high timeliness or importance to the system response, such as "fire risk" in forest fire monitoring, "fall actions" detected by elderly companion robots, and "equipment failure warnings" in industrial scenarios. These events require multi-node collaborative verification to avoid misjudgments. During the encapsulation process, the pulse packet not only contains the core information of the local cognitive decision (such as event type and event feature description), but also adds a source node identifier (clearly indicating which edge node the decision originated from), a precise timestamp (recording the specific moment the event was determined, used to solve the timing deviation problem in multi-node data transmission), and an event weight (set according to the urgency of the event or the reliability of the node; for example, the weight of a fire event is higher than that of a normal environmental monitoring event). The addition of this information references the "timetamped pulse packet" design in this invention, ensuring that the receiving node can clearly understand the decision background and provide a complete basis for subsequent feature matching and verification.

[0036] After pulse packet encapsulation, it is transmitted to other edge nodes via a collaborative communication module, which is the key carrier for realizing multi-node information sharing and collaboration. The collaborative communication module does not employ a continuous data transmission mode, but rather follows the principle of "event-triggered communication," activating only when a critical event pulse packet needs to be transmitted, and maintaining a low-power state when no critical events occur. This aligns with the design philosophy of "reducing communication redundancy and power consumption" in this invention. The specific transmission carrier can be selected based on the application scenario. For example, in large-scale outdoor scenarios (such as forest monitoring), LoRa or mesh networks can be used to achieve long-distance, low-power transmission; in close-range industrial workshops or home scenarios, more efficient local communication protocols can be selected. During transmission, the module prioritizes the transmission of critical event pulse packets, preventing bandwidth consumption by other non-critical data, ensuring that multiple nodes can quickly receive the information requiring collaborative verification, laying the foundation for subsequent real-time response.

[0037] Upon receiving the timestamped pulse packet, other edge nodes immediately initiate a feature matching and verification process, a core step in verifying the authenticity of events in distributed collaborative decision-making. First, the receiving node parses the event information in the pulse packet to identify the key event types and features requiring verification. Then, it retrieves real-time or recently cached environmental data from its local sensors (such as temperature, images, sound, and gas concentration). This data is consistent with the sensor data source of the sending node, ensuring uniformity in verification standards. Next, a locally deployed lightweight SNN model compares the preset features of the key events in the pulse packet with the actual features extracted from the local sensor data. For example, when verifying a "fire risk" event, it compares whether the local temperature exceeds a threshold, whether smoke is detected, and whether there are flame edges in the image. If the local sensor data features match the key event features highly (i.e., in the "support" state), a "verification passed pulse signal" is fed back. If there is a significant conflict between the local data features and the key event features (e.g., the sending node determines "fire risk," but the receiving node has a normal temperature and no smoke), a "pending confirmation pulse signal" is fed back. In some scenarios, local data features will also be fed back simultaneously to provide more judgment basis for the sending node or the global fusion module. The entire verification process strictly relies on the "edge node autonomous cognition and local decision-making" capability in this invention to ensure that each node can independently participate in collaborative verification and improve the accuracy and reliability of global decision-making.

[0038] In one embodiment, step S503 includes: S5031. Perform feature matching and verification between key events and local sensor data; S5032. If the local sensor data matches the features of the key event in a supportive manner, then the feedback verification passes with a pulse signal. S5033. If the local sensor data contradicts the characteristics of the key event, a confirmation pulse signal is fed back.

[0039] In this embodiment, feature matching and verification of key events with local sensor data is a core step for edge nodes to participate in distributed collaborative decision-making. Its purpose is to verify the authenticity of key events a second time through local data, avoiding global decision-making bias caused by misjudgments from a single node. This process strictly relies on the lightweight SNN model deployed on the edge nodes and pre-configured event feature templates. Specifically, when an edge node receives a "key event pulse packet" (containing event type, feature description, timestamp, etc.) transmitted by other nodes through the collaborative communication module, it first retrieves environmental data collected in real-time or recently cached by its own local sensors. This data covers multimodal information such as temperature, images, sound, and gas concentration. For example, in a forest fire monitoring scenario, local sensor data may include the current area temperature value, smoke concentration readings, and frames captured by cameras. Subsequently, the edge node will activate the feature matching function of the local SNN model, comparing the preset features of key events (such as the feature combination of "temperature > 50℃ + smoke concentration > threshold + image containing flame edges" corresponding to the "fire" event) with the actual features extracted from the local sensor data one by one. During the comparison process, the SNN model's ability to process time-series signals will be used to analyze the dynamic changing trend of data features (such as whether the temperature continues to rise or whether the smoke concentration gradually increases), rather than relying solely on data from a single time point, to ensure that the matching results are more in line with the actual scenario.

[0040] If the actual features extracted from local sensor data completely or largely match the preset features of a critical event, indicating a "support" state, the edge node determines that the critical event also holds true in its local area. At this point, it generates and sends back a "verification passed pulse signal." For example, upon receiving a "fire risk" critical event, if the edge node's local temperature sensor detects a temperature consistently above 50°C, the smoke sensor reading exceeds the warning threshold, and the camera data, processed by an SNN model, identifies flame edge features, all three support a "fire" event. Therefore, it sends back a verification passed signal, which is encapsulated into a timestamped pulse packet and transmitted to the collaborative communication module, providing support for global decision-making. This "support" state determination does not require all sensor data to match. Instead, feature matching weights are set based on the importance of the event. For example, in fire monitoring, temperature and smoke data have higher matching weights than other data; as long as the core features match, it can be determined as support.

[0041] If the actual features extracted from local sensor data significantly conflict with the preset features of a critical event, i.e., a "contradictory" state is observed, the edge node determines that the authenticity of the critical event in the local area is questionable. In this case, it generates and feeds back a "pending confirmation pulse signal." For example, after receiving a "fire risk" critical event, if the edge node's local temperature sensor shows normal temperature and zero smoke concentration, and only the camera captures a suspected light spot (not a flame), these data contradict the preset features of "fire," and a pending confirmation signal will be fed back. Furthermore, if some local sensor data is missing or abnormal (e.g., a temporary sensor malfunction resulting in no readings), and the critical event features cannot be fully verified, it will also be classified as a "contradictory" state and a pending confirmation signal will be fed back to avoid misjudgment due to incomplete data. After feeding back the pending confirmation signal, the edge node enters a short-term data re-acquisition phase to re-acquire environmental data and perform secondary verification. Simultaneously, the re-acquisition results are synchronized to the collaborative communication module, providing a more comprehensive reference for global decision-making and ensuring that the system maintains caution and accuracy in decision-making even when faced with contradictory data.

[0042] In one embodiment, S600 includes: S601. When the decision-making mechanism is a majority voting mechanism, the percentage of edge nodes that pass the verification pulse signal in the statistical feedback verification signal is calculated. S602. If the proportion of the quantity exceeds the preset threshold, the key event is determined to be established, and the final global decision to execute the response action is generated. S603. When the decision-making mechanism is a confidence-weighted mechanism, a confidence weight is pre-configured for each edge node, and the total confidence score is calculated. S604. If the total score exceeds the preset threshold of the sum of total weights, the key event is determined to be true, and the final global decision is generated. S605. When the decision-making mechanism is the master node coordination mechanism, the master node is selected according to the preset node selection rules. S606 After receiving the verification signals from each edge node, the master node combines the judgment results of the SNN model on key events to generate the final global decision.

[0043] In this embodiment, when the decision-making mechanism is a majority voting mechanism, its core logic is to determine whether a critical event is valid by statistically analyzing the verification results of a majority of edge nodes. This mechanism is suitable for scenarios where the hardware performance and deployment environment of each edge node are similar, and the differences in decision reliability are small, such as large-scale forest fire monitoring and multi-area intelligent security. Specifically, the global fusion module first collects the verification signals fed back by all edge nodes participating in collaborative verification. These signals are mainly divided into two categories: "verification passed pulse signal" and "pending confirmation pulse signal". Subsequently, the module counts the number of edge nodes that fed back "verification passed pulse signal" and calculates the proportion of this number to the total number of nodes participating in verification. The "preset threshold" here needs to be set according to the reliability requirements of the actual application scenario, and is usually above 50%. For example, in a fire monitoring scenario where the accuracy of decision-making is highly critical, the threshold can be set to 60%. If the percentage of nodes that have passed verification exceeds the preset threshold, it indicates that most nodes recognize the authenticity of the critical event. At this point, the global fusion module will determine that the critical event is valid and generate a final global decision to execute a response action, such as triggering an alarm on the fire platform, starting a drone patrol, or turning on the on-site spraying system to ensure a rapid response to the critical event. If the percentage does not reach the preset threshold, the response action will not be executed for the time being. The node that has returned a "pending confirmation" signal may be instructed to re-collect environmental data and perform a second verification to further confirm the event situation.

[0044] When the decision-making mechanism is a confidence-weighted mechanism, its focus is on comprehensively judging key events by combining the actual reliability differences of each edge node. This is suitable for scenarios where the reliability of decisions varies significantly among edge nodes due to differences in hardware performance, deployment location, and environmental interference. Under this mechanism, a confidence weight must first be pre-configured for each edge node. The weight is set based on factors including the node's historical decision accuracy (e.g., the degree to which the node's judgment results match the actual situation over a past period), hardware performance (e.g., whether high-performance hardware such as neuromorphic chips is used), and deployment environment stability (e.g., whether it is located in an area with minimal signal interference and good data acquisition conditions). Nodes with higher reliability have a larger confidence weight. For example, a node deployed in a key monitoring area of ​​a forest, using an Intel Loihi neuromorphic chip, and with a historical decision accuracy of 95%, can have a confidence weight of 0.8, while a node deployed in an edge area, using a regular MCU, and with a historical accuracy of 80%, can have a weight of 0.5. Next, the global fusion module calculates the total confidence score based on the verification signals fed back by each node ("verification passed" signals are marked as valid, and "pending confirmation" signals are marked as invalid) and their corresponding confidence weights. This total confidence score is the sum of "confidence weight × signal validity" for all nodes. Simultaneously, a preset threshold based on the sum of the total weights (e.g., 60% of the total weights) is set. If the total confidence score exceeds this threshold, it indicates that the result after considering the reliability of each node supports the validity of the critical event, and the global fusion module generates the final global decision. If the threshold is not exceeded, further verification information from more nodes needs to be collected or the node weights need to be reassessed to ensure the accuracy of the decision.

[0045] When the decision-making mechanism is a master node coordination mechanism, its core is to rely on a pre-selected master node to quickly generate global decisions. This is suitable for scenarios with high real-time decision requirements and a need for rapid response to critical events, such as emergency response in elderly companion robots and early warning of industrial equipment failures. The selection of the master node must follow preset node selection rules. Typically, nodes with strong computing power (e.g., capable of efficiently running a complete SNN model), centrally located (facilitating rapid reception of signals from edge nodes), and low communication latency (reducing signal transmission time) are prioritized as master nodes, such as the master control node of an elderly companion robot and the central monitoring node in an industrial workshop. When generating global decisions, the master node first receives verification signals from all edge nodes to fully understand the judgment of each node on critical events. At the same time, the master node uses its own deployed SNN model (which is consistent with the pre-built SNN model system of the system and includes sub-networks adapted to the master node's functions, such as a sub-network for processing accelerometer sensor signals) combined with environmental data collected by its own sensors (e.g., the master node of an elderly companion robot detects whether there is a fall through the accelerometer sensor) to independently judge critical events. Finally, the master node integrates the verification signals from each edge node and the judgment results of its own SNN model to quickly generate the final global decision. For example, after determining that "a fall event has occurred", it immediately triggers emergency calls, plays reminder voice messages, and other response actions to ensure that the decision is completed and executed in a short time to meet the real-time requirements of the scenario.

[0046] In one embodiment, a distributed cognitive decision-making device based on an SNN model is provided, which corresponds one-to-one with the distributed cognitive decision-making method based on an SNN model described in the above embodiments. (Refer to...) Figure 3 , Figure 3 This is a schematic diagram of the functional modules of a preferred embodiment of the distributed cognitive decision-making device based on an SNN model according to the present invention. The modules include a model construction module 10, a data encoding module 20, a pulse stream calculation module 30, a cognitive decision-making module 40, a data verification module 50, and a final decision-making module 60. Detailed descriptions of each functional module are as follows: Model building module 10 is used to pre-build the SNN model and configure the initial parameters of the SNN model; Data encoding module 20 is used to acquire environmental data, encode the environmental data, and generate a pulse stream; The pulse stream calculation module 30 is used to input the pulse stream into the SNN model, which calculates based on an event-driven mechanism and outputs signal processing features. Cognitive decision module 40 is used to determine local cognitive decisions based on signal processing characteristics; The data verification module 50 is used to transmit the local cognitive decision to the edge node for feature matching and verification through the collaborative communication module, and to provide a verification signal. The final decision module 60 is used to generate the final global decision through the global fusion module based on the verification signals fed back by each edge node.

[0047] In one embodiment, the model building module 10 includes: The model configuration unit is used to configure the neuron model and learning mechanism for the SNN model based on the platform's computing power. Model splitting unit, used to split an SNN model into several sub-networks according to functional or perceptual modality; The parameter configuration unit is used to configure the initial synaptic weights, neuron firing thresholds, synaptic pruning parameters, and pulse firing threshold adjustment coefficients for each sub-network, while also constructing the sub-network event perception table.

[0048] In one embodiment, the data encoding module 20 includes: The data acquisition unit is used to monitor environmental data through multi-source sensors, and to acquire environmental data when the environmental data changes. A data classification unit is used to classify the environmental data and select the encoding method of the environmental data according to the environmental type; The data encoding unit is used to encode the environmental data through an encoding method to generate a pulse stream.

[0049] In one embodiment, the pulse flow calculation module 30 includes: The feature matching unit is used to input the pulse stream into the SNN model and match pulse features through the event-aware mapping table in the SNN model. The network activation unit is used to activate the corresponding subnetwork and neuron group if the pulse feature matches the event type in the subnetwork event perception table and the pulse map template. The feature generation unit is used to activate subnetworks that process pulse streams through neuronal integration and synaptic transmission to generate signal processing features.

[0050] In one embodiment, the data verification module 50 includes: A data packaging unit is used to package the local cognitive decision into a timestamped pulse packet when the local cognitive decision is a critical event; The data transmission unit is used to transmit the timestamped pulse packets to the edge nodes via the cooperative communication module; The data verification unit is used by the edge node to receive pulse packets with timestamps, perform feature matching and verification, and feed back verification signals.

[0051] In one embodiment, the edge node receives timestamped pulse packets for feature matching and verification, and feeds back a verification signal, including: The data verification unit is used to perform feature matching and verification between key events and local sensor data; A data support unit is used to provide a feedback verification pulse signal if the local sensor data matches the features of the key event. The data contradiction unit is used to feed back a confirmation pulse signal if the local sensor data contradicts the features of the key event.

[0052] In one embodiment, the final decision module 60 includes: The majority voting mechanism unit is used to, when the decision-making mechanism is a majority voting mechanism, count the percentage of edge nodes that pass the verification pulse signal in the feedback verification signal; The voting decision unit is used to determine that a key event has occurred if the percentage of votes exceeds a preset threshold, and to generate a final global decision to execute a response action. The confidence weighting mechanism unit is used to pre-configure confidence weights for each edge node and calculate the total confidence score when the decision mechanism is a confidence weighting mechanism. The weighted decision unit is used to determine that the key event has occurred and generate the final global decision if the total score exceeds a preset threshold of the sum of the total weights. The master node coordination mechanism unit is used to select the master node according to a preset node selection rule when the decision mechanism is the master node coordination mechanism. The master node decision unit is used to generate the final global decision by combining the judgment results of the SNN model on key events after the master node receives the verification signals from each edge node.

[0053] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 4 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computational and control capabilities. The memory includes non-volatile and / or volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface is used for communication with external user terminals via a network connection. When the computer program is executed by the processor, it implements the functions or steps of a server-side distributed cognitive decision-making method based on an SNN model.

[0054] In one embodiment, a computer device is provided, which may be a user terminal, and its internal structure diagram may be as follows: Figure 5 As shown, the computer device includes a processor, memory, network interface, display screen, and input devices connected via a system bus. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used to communicate with an external server via a network connection. When the computer program is executed by the processor, it implements user-side functions or steps of a distributed cognitive decision-making method based on an SNN model. In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to perform the following steps: Pre-build the SNN model and configure its initial parameters; Acquire environmental data, encode the environmental data, and generate a pulse stream; The pulse stream is input into the SNN model, which calculates based on an event-driven mechanism and outputs signal processing features. Determine local cognitive decisions based on signal processing characteristics; The local cognitive decision is transmitted to the edge node for feature matching and verification through the collaborative communication module, and a verification signal is fed back. Based on the verification signals fed back by each edge node, the final global decision is generated through the global fusion module.

[0055] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor: Pre-build the SNN model and configure its initial parameters; Acquire environmental data, encode the environmental data, and generate a pulse stream; The pulse stream is input into the SNN model, which calculates based on an event-driven mechanism and outputs signal processing features. Determine local cognitive decisions based on signal processing characteristics; The local cognitive decision is transmitted to the edge node for feature matching and verification through the collaborative communication module, and a verification signal is fed back. Based on the verification signals fed back by each edge node, the final global decision is generated through the global fusion module.

[0056] It should be noted that the functions or steps that can be implemented by the computer-readable storage medium or computer device described above can be referred to the relevant descriptions on the server side and user side in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.

[0057] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0058] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0059] It should be noted that if any software tools or components not belonging to this company appear in the embodiments of this application, they are merely illustrative examples and do not represent actual use. The above-described embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A distributed cognitive decision-making method based on an SNN model, characterized in that, Includes the following steps: Pre-build the SNN model and configure its initial parameters; Acquire environmental data, encode the environmental data, and generate a pulse stream; The pulse stream is input into the SNN model, which calculates based on an event-driven mechanism and outputs signal processing features. Determine local cognitive decisions based on signal processing characteristics; The local cognitive decision is transmitted to the edge node for feature matching and verification through the collaborative communication module, and a verification signal is fed back. Based on the verification signals fed back by each edge node, the final global decision is generated through the global fusion module.

2. The distributed cognitive decision-making method based on the SNN model as described in claim 1, characterized in that, The pre-built SNN model and the configuration of its initial parameters include: Configure neuron models and learning mechanisms for SNN models based on the platform's computing power; The SNN model is split into several sub-networks according to its function or perceptual modality; Initial synaptic weights, neuron firing thresholds, synaptic pruning parameters, and pulse firing threshold adjustment coefficients are configured for each subnetwork, and an event perception table for the subnetwork is constructed.

3. The distributed cognitive decision-making method based on the SNN model as described in claim 1, characterized in that, The process of acquiring environmental data, encoding the environmental data, and generating a pulse stream includes: Environmental data is monitored through multi-source sensors, and environmental data is collected when the environmental data changes. The environmental data is classified, and the encoding method of the environmental data is selected according to the environmental type; The environmental data is encoded using an encoding method to generate a pulse stream.

4. The distributed cognitive decision-making method based on the SNN model as described in claim 1, characterized in that, The step involves inputting the pulse stream into an SNN model, where the SNN model is calculated based on an event-driven mechanism and outputs signal processing features, including: The pulse stream is input into the SNN model, and the pulse features are matched through the event-aware mapping table in the SNN model. If the pulse feature matches the event type in the sub-network event perception table with the pulse map template, then the corresponding sub-network and neuron group are activated; The activated subnetworks process the pulse flow through neuronal integration and synaptic transmission, generating signal processing features.

5. The distributed cognitive decision-making method based on the SNN model as described in claim 1, characterized in that, The step of transmitting the local cognitive decision to edge nodes for feature matching and verification via a collaborative communication module, and feeding back verification signals, includes: When the local cognitive decision is a critical event, the local cognitive decision is packaged into a timestamped pulse packet; The timestamped pulse packets are transmitted to the edge nodes via the collaborative communication module; The edge node receives timestamped pulse packets for feature matching and verification, and feeds back a verification signal.

6. The distributed cognitive decision-making method based on the SNN model as described in claim 5, characterized in that, The edge node receives timestamped pulse packets for feature matching and verification, and feeds back verification signals, including: Perform feature matching and verification between key events and local sensor data; If the local sensor data matches the features of the key event in a supportive manner, then the feedback verification is passed via a pulse signal; If the local sensor data does not match the characteristics of the key event, a confirmation pulse signal will be fed back.

7. The distributed cognitive decision-making method based on the SNN model as described in claim 1, characterized in that, The step of generating a final global decision based on the verification signals fed back by each edge node through the verification mechanism of the global fusion module includes: When the decision-making mechanism is a majority voting mechanism, the percentage of edge nodes that pass the verification pulse signal in the statistical feedback verification signal is calculated. If the proportion of the number exceeds the preset threshold, the key event is determined to be true, and the final global decision to execute the response action is generated. When the decision-making mechanism is a confidence-weighted mechanism, a confidence weight is pre-configured for each edge node, and the total confidence score is calculated. If the total score exceeds the preset threshold of the sum of total weights, the key event is determined to be true, and the final global decision is generated. When the decision-making mechanism is the master node coordination mechanism, the master node is selected according to the preset node selection rules. After receiving the verification signals from each edge node, the master node combines the judgment results of the SNN model on key events to generate the final global decision.

8. A distributed cognitive decision-making device based on an SNN model, characterized in that, The distributed cognitive decision-making device based on the SNN model includes: The model building module is used to pre-build the SNN model and configure the initial parameters of the SNN model; The data encoding module is used to acquire environmental data, encode the environmental data, and generate a pulse stream; The pulse stream calculation module is used to input the pulse stream into the SNN model, which calculates based on an event-driven mechanism and outputs signal processing features. The cognitive decision-making module is used to determine local cognitive decisions based on signal processing characteristics; The data verification module is used to transmit the local cognitive decision to the edge node for feature matching and verification through the collaborative communication module, and to provide a verification signal. The final decision module is used to generate the final global decision based on the verification signals fed back by each edge node through the global fusion module.

9. A computer device, characterized in that, The computer device includes a memory, a processor, and a distributed cognitive decision-making program based on an SNN model stored in the memory and executable on the processor. When the distributed cognitive decision-making program based on an SNN model is executed by the processor, it implements the steps of the distributed cognitive decision-making method based on an SNN model as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The storage medium stores a distributed cognitive decision-making program based on an SNN model, which, when executed by a processor, implements the steps of the distributed cognitive decision-making method based on an SNN model as described in any one of claims 1-7.

Images