Device and method for multispectral optical signal acquisition and AI intelligent analysis

Abstract

The invention discloses a multispectral optical signal acquisition and AI intelligent analysis device and method, and belongs to the technical field of optical signal detection and artificial intelligence. A multi-spectral optical signal intelligent analysis deep network system (MSS-IDNS) is defined as a core, and the system takes a 300-1000nm multi-spectral optical signal as an input carrier and can be fused with various deep learning models. The device comprises a multispectral optical signal acquisition terminal and an AI intelligent analysis module. The method comprises the six steps of optical signal collection, data preprocessing, feature extraction, AI analysis, model iteration and intelligent response, so that intelligent analysis and biological state recognition of optical signals are completed. The method breaks through the human visual perception boundary and adapts to multiple scenes, core hardware is an existing mature industrial product, batch production can be achieved, and the method has extremely high civilian mass production potential and commercial value.

Description

Technical Field

[0001] This invention relates to the fields of optical signal detection, multispectral analysis, and deep learning in artificial intelligence. Specifically, it relates to a device and method for multispectral optical signal acquisition and AI intelligent analysis, which can be widely applied to the identification and intelligent response of living beings in various scenarios such as intelligent pet management, human health monitoring, and intelligent family companionship. The core relies on the Multispectral Optical Signal Deep Analysis System (MSSDAS) to achieve full-dimensional intelligent analysis of optical signals. Background Technology

[0002] The human visual system has evolved over a long period of time to form a light signal receiving system with red, green and blue cone cells as the core. It can only capture light signals in the visible light band and form visual images through mixing and analysis by the brain. It cannot perceive ultraviolet light, infrared light, polarized light and other band signals, nor can it distinguish the narrow band differences within the core band of visible light. It has an innate limitation of perceptual boundaries. In existing technologies, products related to life form status recognition, health monitoring, and intelligent service robots mainly rely on signals such as images, voice, and body movements for status judgment and command response. These signals have inherent defects such as being easily forged, easily interfered with, and subjectively spoofed, leading to low judgment accuracy and poor scenario adaptability. Some spectral detection devices can only acquire light signals in a single or limited band and are not deeply integrated with artificial intelligence technology, failing to achieve accurate analysis of light signals and intelligent association with life form status. Furthermore, existing AI analysis models are mostly single-network models, capable of extracting features in specific dimensions and relying on human experience to replicate these relationships. They solidify the correlation between external features and states summarized by humans into algorithms, failing to autonomously uncover the potential intrinsic connections between light signals and life form status. There is also no unified deep learning model system to perform full-dimensional analysis of various light signal features; its performance is limited by the boundaries of human cognition and experience. To address the aforementioned technical shortcomings, there is currently no integrated technical solution that can break through the boundaries of human visual perception, achieve accurate acquisition of multi-band narrowband optical signals, and autonomously mine the intrinsic correlation between optical signals and the state of living organisms using a unified deep learning model system. This solution cannot meet the core requirements of intelligent devices for signal authenticity, detection accuracy, and intelligent analysis. Therefore, a new technical solution is urgently needed to solve these problems. Summary of the Invention

[0003] (a) Purpose of the invention The purpose of this invention is to overcome the shortcomings of existing technologies and provide a device and method for multispectral optical signal acquisition and AI intelligent analysis. This invention breaks through the human visible light perception boundary and achieves accurate acquisition of multi-band narrowband optical signals in the 300-1000nm ultraviolet, visible, near-infrared, and polarized light ranges. Relying on the Multi-Spectral Signal Intelligent Decoding Deep Network System (MSSDAS), it autonomously mines the causal or strong correlation between optical signals and the state of living organisms, enabling accurate identification and intelligent response to the needs, emotional states, and physical health of living organisms. This solves the technical problems of existing technologies, such as easy signal forgery, low detection accuracy, reliance on human experience for analysis, and the lack of a unified deep learning model system. (II) Technical Solution To achieve the aforementioned objectives, this invention first defines a Multispectral Optical Signal Deep Analysis System (MSSDAS): This system uses multispectral optical signals (300-1000nm ultraviolet / visible / near-infrared + polarized light + narrowband visible light) as the sole input carrier, with optical signal feature extraction, fusion, and analysis as its core objectives. It can integrate various deep learning models, including but not limited to one-dimensional convolutional neural networks (1D-CNN / EDCNN), one-dimensional residual networks (1D-ResNet), gated recurrent units (GRU), temporal convolutional networks (TCN), attention mechanism networks, etc., forming a comprehensive, scalable, and computationally unlimited deep learning model system. This system can be modularly split, fused, or deployed according to various application scenarios (such as pet management / health monitoring / family companionship) and computing power conditions (edge ​​chips / cloud servers / computing centers), and supports OTA online iterative optimization. Based on the above system, this invention provides a device for multispectral optical signal acquisition and AI intelligent analysis. The core includes a multispectral optical signal acquisition terminal and an AI intelligent analysis module. The AI ​​intelligent analysis module is the hardware implementation of the Multispectral Optical Signal Deep Resolution System (MSSDAS). The multispectral optical signal acquisition terminal and the AI ​​intelligent analysis module achieve bidirectional communication through a data transmission module. The main structure is as follows: 1. The multispectral optical signal acquisition terminal is the core carrier for optical signal acquisition in the device. It includes a multispectral sensor module, auxiliary optical components, and a data transmission and storage module. These three components work together to achieve accurate acquisition, anti-interference processing, and high-speed transmission of multi-band optical signals, providing high-quality, high-purity optical signal input data for the Multispectral Optical Signal Deep Resolution System (MSSDAS). 1.1 Multispectral sensor module: It adopts a multispectral core with a full light-sensing range covering the 300-1000nm band, including ultraviolet, full visible light, and near-infrared light bands. It is equipped with a polarized light sensor and further subdivides the human visible light band into narrow bands, which can accurately capture the subtle differences in multi-band narrow-band light signals and polarized light signals emitted by target life forms. 1.2 Auxiliary Optical Components: The combination design of "active light source + narrowband filter" replaces the traditional unfiltered structure. While reducing signal loss and shrinking the size of the device, it effectively filters ambient light interference and improves the purity of light signal acquisition. It integrates a miniature dark field illumination module and is paired with a high signal-to-noise ratio polarization light sensor with a signal-to-noise ratio of ≥1000:1. It can adapt to complex lighting scenarios such as strong light, low light, and multiple light sources superimposed indoors and outdoors, ensuring that it can accurately capture the characteristic light signals of the target object in various environments. 1.3 Data Transmission and Storage Module: Equipped with a high-speed transmission chip, it supports real-time uploading of spectral data and is configured with 16-32GB of local cache to meet the storage requirements of the device side, achieving millisecond-level optical signal data transmission without delay, and accurately adapting to the rapid response requirements of devices such as intelligent robots; at the same time, it is equipped with a lightweight wireless communication module for data synchronization and OTA iteration of the Multispectral Optical Signal Deep Resolution System (MSSDAS), avoiding communication delays or interruptions from affecting the normal operation of the device. 2. The AI ​​intelligent analysis module communicates bidirectionally with the multispectral optical signal acquisition terminal and is the core implementation unit of the Multispectral Optical Signal Deep Analysis System (MSSDAS). It integrates a data preprocessing unit, a feature extraction unit, an intelligent analysis unit, and a model iteration unit. These four units work together to process optical signal data, extract features, perform intelligent analysis, and iteratively optimize the model system. It can flexibly integrate various deep learning models according to computing power conditions, adapting to different computing scenarios such as edge devices, cloud computing, and computing centers. 2.1 Data Preprocessing Unit: This is the preprocessing unit of the Multispectral Optical Signal Deep Analysis System (MSSDAS). It supports real-time preview, noise reduction, normalization, and batch processing of spectral data, providing high-quality and high-purity data sources for various deep learning models within the system. It also integrates a "digital virtual filter" AI algorithm to replace traditional optical filtering components. Through neural networks, it autonomously selects effective spectral channels to further filter environmental noise and significantly improve the signal acquisition accuracy under complex lighting conditions. 2.2 Feature Extraction Unit: This is the core feature extraction module of the Multispectral Optical Signal Deep Analysis System (MSSDAS). It can be flexibly equipped with various deep learning models such as 1D-CNN (One-dimensional Convolutional Neural Network), 1D-ResNet (One-dimensional Residual Network), and 1D-MobileNetV3 (One-dimensional Lightweight Mobile Network) according to the accuracy and computing power requirements. It directly extracts features based on the characteristics of spectral sequence data, avoiding the pixel-level computational redundancy of traditional image models, greatly improving the data processing speed and efficiency, and accurately extracting core features of optical signals such as band shift, intensity change, and polarization mode. 2.3 Intelligent Analysis Unit: This is the core analysis module of the Multispectral Optical Signal Deep Analysis System (MSSDAS). It can integrate spectral unmixing algorithms with various deep learning models. The spectral unmixing algorithm can effectively separate overlapping optical signal features, avoiding misjudgment caused by similar signals. The customized integrated deep learning model can autonomously discover the causal or strong correlation between the core features of optical signals and the state of living organisms, without relying on human experience to replicate them, thus achieving accurate identification of the needs, emotional state, and physical health of living organisms. 2.4 Model Iteration Unit: This is the iterative optimization module of the Multispectral Optical Signal Deep Resolution System (MSSDAS). It adopts lightweight technologies such as model quantization and pruning, and can deploy the customized fusion model trained within the system to the device's local chip, cloud server, or computing center. It also has OTA online iteration function, continuously optimizing various model parameters within the system based on real data generated during actual user use, improving the model system's scene adaptability and recognition accuracy, and forming a closed-loop iterative system of "data acquisition - model optimization - experience improvement". This invention also provides a multispectral optical signal acquisition and AI intelligent analysis method based on the above-mentioned device and the Multispectral Optical Signal Deep Resolution System (MSSDAS), comprising the following steps: Step S1: Precise acquisition of multi-band optical signals The multispectral sensor module of the multispectral optical signal acquisition terminal captures ultraviolet, visible, near-infrared, and polarized light signals in the 300-1000nm band emitted by the target organism. The auxiliary optical components filter ambient light interference through "active light source + narrowband filtering" and a miniature dark field illumination module to ensure the purity and accuracy of the light signal acquisition, providing raw input data for the Multispectral Optical Signal Deep Resolution System (MSSDAS). Step S2: Spectral data preprocessing The data transmission and storage module transmits the collected raw optical signal data to the AI ​​intelligent analysis module in real time at millisecond speeds. The data preprocessing unit of the Multispectral Optical Signal Deep Analysis System (MSSDAS) sequentially performs noise reduction, normalization, and channel registration on the raw data. At the same time, the AI ​​algorithm of "digital virtual filter" autonomously selects effective spectral channels, filters out environmental noise interference, and outputs high-quality, high-purity spectral data sources for subsequent feature extraction and intelligent analysis within the system. Step S3: Extraction of core features of optical signal The feature extraction unit of the Multispectral Optical Signal Deep Analysis System (MSSDAS) can be flexibly equipped with corresponding deep learning models according to computing power and accuracy requirements. It can directly extract features from preprocessed spectral sequence data, accurately mine core features such as band shift, intensity change, and polarization mode of optical signals, avoid pixel-level computational redundancy, improve the efficiency and accuracy of feature extraction, and provide core feature data for intelligent analysis within the system. Step S4: AI Intelligent Analysis and Living Organism Status Recognition The intelligent analysis unit of the Multispectral Optical Signal Deep Analysis System (MSSDAS) first separates the core features of the extracted optical signal through a spectral unmixing algorithm to avoid misjudgment caused by overlapping features. Then, it fuses the corresponding deep learning model according to the needs of the scene. Using the customized fusion model that has been trained, it autonomously matches the correlation between the core features of the optical signal and the state of the living organism, accurately identifies the needs, emotional state, and physical health status of the target living organism, and outputs the recognition results. Step S5: Iterative Model Optimization The model iteration unit of the Multispectral Optical Signal Deep Analysis System (MSSDAS) uses the recognition results, feedback data and newly added optical signal data from the user's actual use as training samples to continuously optimize the parameters of various deep learning models within the system. At the same time, through the OTA online iteration function, the optimized model system is synchronized to the device-side chip, cloud server or computing center to realize the real-time update of the model system and continuously improve the recognition accuracy and scene adaptability. Step S6: Recognition Result Output and Intelligent Action Response Based on the life form status recognition results output by the Multispectral Optical Signal Deep Analysis System (MSSDAS), the corresponding actions of the matching execution terminal are triggered to achieve targeted intelligent response. The execution terminal can be flexibly configured according to the application scenario, including feeding devices and interactive modules for intelligent pet robots, health management APPs for human health monitoring, and voice and motion modules for family companion robots. (III) Beneficial Effects Compared with the prior art, the present invention has the following significant technical advantages: 1. Breaking through the boundaries of human visual perception to achieve precise acquisition of multi-band optical signals: This invention covers the 300-1000nm ultraviolet, visible, near-infrared and polarized light bands, while subdividing visible light into narrow bands with a spectral resolution ≥2.3nm. It can capture the latent light signals emitted by living organisms, and its detection range and accuracy far exceed those of existing technologies. It solves the problem of the inherent limitations of human visual perception and provides a rich and accurate optical signal data foundation for subsequent intelligent analysis. 2. Define a unified deep learning model system to achieve full-dimensional analysis of optical signals: For the first time, a multispectral optical signal deep analysis system (MSSDAS) is defined, which breaks through the technical limitations of existing single models. It can integrate various deep learning models to achieve full-dimensional extraction and analysis of optical signal features. Moreover, the system supports unbounded deployment of computing power and can be flexibly split and integrated according to different computing power conditions such as edge, cloud, and computing center to adapt to the computing power needs of multiple scenarios. 3. Relying on the absolute authenticity of light signals, false interference is avoided from the root: As an objective physical signal released by a living organism, light signals have the characteristics of being unforgeable and undisguised, unlike easily forged signals such as language, images, and actions. This invention uses light signals as the sole input carrier, thereby improving the authenticity and reliability of the identification of the state of living organisms from the root. 4. Breaking free from the constraints of human experience and achieving autonomous deep learning in AI: This invention utilizes the Multispectral Optical Signal Deep Analysis System (MSSDAS) to autonomously uncover the causal or strong correlation between optical signals and the state of living organisms. It does not require replicating human experience, breaking through the technical bottleneck of traditional AI models being limited by human cognition, resulting in higher recognition accuracy and a wider iteration limit. 5. Dual anti-interference design, adaptable to complex scenarios and contactless detection: It adopts a hardware anti-interference design of "active light source + narrowband filtering" and a software anti-interference design of "digital virtual filter", which can adapt to complex lighting scenarios such as strong light, low light and multiple light sources superimposed indoor and outdoor; at the same time, it can realize long-distance, contactless light signal detection, covering blank scenarios such as low light monitoring and privacy protection that existing products cannot adapt to. 6. Integrated closed-loop system, adaptable to multiple scenarios and with mass production potential: Constructing an integrated closed-loop system of "hardware acquisition - software processing - model system analysis - model iteration - intelligent response", it can be customized and lightweightly integrated according to different scenarios such as intelligent pet management, human health monitoring, and family companionship. The core hardware are all existing mature industrial products, and the model system supports full computing power scenario deployment and OTA iteration, which has extremely strong potential for civilian mass production and commercial value. Attached Figure Description Figure 1: Overall structural block diagram of the Multispectral Optical Signal Depth Resolution System (MSSDAS) device Note: The execution terminal (4) can be configured according to the application scenario, including but not limited to feeding devices, interactive modules, health management APPs, voice modules, etc. Figure 2: Hardware structure diagram of the multispectral optical signal acquisition terminal Figure 3: Functional unit block diagram of the Multispectral Optical Signal Depth Resolution System (MSSDAS) Figure 4: Schematic diagram of multispectral optical signal acquisition and AI intelligent analysis method Detailed Implementation The technical solution of the present invention will be clearly and completely described below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are only preferred embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of the present invention. Example 1: Application Scenario of Intelligent Pet Management Robot This embodiment applies the device and multispectral optical signal deep analysis system (MSSDAS) of the present invention to an intelligent pet management robot for felines, adopting an edge-side low-computing-power deployment mode. The system integrates a lightweight deep learning model, and the specific implementation is as follows: 1. Customized Hardware Integration: A lightweight multispectral optical signal acquisition terminal weighing less than 50g is integrated into the robot's head gimbal, paired with a miniature distance sensor to avoid disturbing pets during close-range detection; an entry-level MCU is selected as the hardware carrier for the AI ​​intelligent analysis module, carrying a lightweight fusion model of the Multispectral Optical Signal Deep Resolution System (MSSDAS) to control hardware costs; the multispectral sensor module has a light-sensing range of 300-1000nm and a spectral resolution of 2.3nm, and the auxiliary optical components adopt an "active light source + narrowband filtering" design to adapt to complex indoor lighting scenarios where pets are active. 2. Customized training and deployment of the model system: Construct a feline-specific light signal dataset, covering core states such as hunger, thirst, activity, stress, fear, and health; train the Multispectral Light Signal Deep Analysis System (MSSDAS) based on the dataset, integrate a lightweight deep learning model, and use transfer learning technology to optimize the adaptability to small sample scenarios, which can quickly identify the differences in light signals of felines of different breeds and ages; after the trained model system is lightweighted, it is fixed to the robot's end-side MCU. 3. Method Implementation Process: The multispectral optical signal acquisition terminal captures multi-band optical signals from felines from 360° without blind spots. After data preprocessing and feature extraction by the Multispectral Optical Signal Deep Analysis System (MSSDAS), the fusion model within the system accurately identifies the pet's status. If the pet is identified as hungry / thirsty, the feeding / watering device is automatically triggered; if it is identified as stressed / dangerous, intervention and protective measures are immediately initiated; if it is identified as active, the interactive module is activated to provide companionship and play. Simultaneously, through the system's model iteration unit, the model parameters are continuously optimized by combining auxiliary data such as the pet's eating time and activity level to offset the optical signal drift caused by sensor aging and changes in ambient temperature, ensuring long-term identification accuracy. Example 2: Human Health Monitoring Application Scenario (Edge-side + Cloud Hybrid Computing Power Deployment) This embodiment applies the device and Multispectral Optical Signal Deep Analysis System (MSSDAS) of the present invention to a portable human health monitoring terminal to achieve real-time detection of human health status and early warning of diseases. It adopts a hybrid computing power deployment mode of lightweight processing on the edge and high-precision analysis in the cloud. The edge integrates a lightweight model, and the cloud integrates a deep learning model with an attention mechanism. The specific implementation is as follows: 1. Customized Hardware Integration: Utilizing a portable multispectral light signal acquisition terminal design, it supports contact detection on areas such as the wrist and forehead. It integrates surface-enhanced spectroscopy technology to amplify and process trace bio-light signals from the human body. The multispectral sensor module is upgraded to a high-precision sensor with a spectral resolution of 1.5-2.0nm, accurately capturing characteristic light signals released by early trace lesions in the human body. The professional version is compatible with clinical-grade multispectral imagers, supporting simultaneous measurement of 10 channels of spectrum with a pixel count ≥12 million, meeting the needs of medical clinical testing. It is equipped with a low-computing-power chip on the edge and a communication module with the cloud server, enabling collaborative computing power between the edge and the cloud. 2. Customized training and deployment of the model system: Construct a large-scale human spectral database, perform hierarchical training of the Multispectral Optical Signal Deep Analysis System (MSSDAS) based on the dataset, deploy a lightweight model on the edge to achieve basic optical signal feature extraction and preliminary judgment of abnormal signals; deploy an attention mechanism fusion model in the cloud, integrate spectral data augmentation technology, and improve the disease prediction accuracy to over 95%; achieve linkage and data synchronization between the edge and cloud model systems. 3. Method Implementation Process: Multi-band light signals emitted by human skin are collected via a portable terminal. These signals undergo basic preprocessing and feature extraction using the Multispectral Light Signal Deep Analysis System (MSSDAS) edge model to initially identify abnormal signals. If a basic abnormality is detected, the light signal data is uploaded to the cloud, where a cloud-based model system performs high-precision feature analysis and health status assessment. Real-time light signal data is compared with a historical human spectral database to accurately determine health status. If abnormalities such as band shifts or intensity changes are detected in the light signal of a certain part of the body, disease or lesion risks are immediately predicted, and a detection report is generated via a supporting health management app, marking risk points and pushing personalized prevention suggestions. Simultaneously, the cloud continuously optimizes the model system parameters using user health data, gradually verifying the causal relationship between light signals and human health status, and synchronizing the optimized lightweight model to the edge device to achieve closed-loop iteration of the model system. Example 3: Application scenario of home companion robot (deployment on the edge with medium computing power) This embodiment applies the device and Multispectral Optical Signal Deep Analysis System (MSSDAS) of the present invention to a home intelligent companion robot, realizing emotion recognition, health monitoring, and intelligent companionship for the elderly, children, and adults. It adopts a medium-computing-power edge deployment mode, and integrates an attention mechanism deep learning model within the system. The specific implementation method is as follows: 1. Customized Hardware Integration: The multispectral optical signal acquisition terminal is deeply integrated with the robot's original vision module, adopting a hidden design that balances functionality and aesthetics. It is paired with a body temperature sensing auxiliary module to achieve coordinated acquisition and analysis of optical and temperature signals, enhancing the accuracy of status judgment. A mid-range MCU is selected as the hardware carrier for the AI ​​intelligent analysis module, supporting the mid-computing power fusion model of the Multispectral Optical Signal Deep Analysis System (MSSDAS), adapting to the low-power, high-real-time requirements of home robots. Standardized hardware upgrade interfaces are also reserved for future configuration upgrades. 2. Customized training and deployment of the model system: Construct a multi-person emotional spectrum dataset covering different ages, genders, and emotional states, including core emotions such as irritability, depression, anger, joy, and calmness; train the Multispectral Optical Signal Deep Analysis System (MSSDAS) based on the dataset, integrate the attention mechanism deep learning model, and optimize the judgment logic by combining auxiliary data such as body temperature and behavioral characteristics to avoid misjudgments caused by a single optical signal; after quantizing and pruning the trained model system, deploy it to the MCU on the edge of the home robot. 3. Method Implementation Process: The robot captures the light signals released by the user without contact via a multispectral light signal acquisition terminal. Combined with temperature data from the body temperature sensing module, the data is transmitted to the Multispectral Light Signal Deep Analysis System (MSSDAS) for data preprocessing and feature extraction. The system's fusion model accurately identifies the user's emotional state, whether their body temperature is abnormal, and their physiological needs. If the user is identified as irritable / depressed, the system automatically matches soothing music and comforting words to regulate their emotions. If an abnormal body temperature is detected, an early warning message is immediately pushed to the associated terminal. If the user is identified as hungry / thirsty, food service is provided promptly. Simultaneously, the software integrates an ambient light adaptive adjustment module to adjust the hardware's filtering band and signal amplification factor in real time to adapt to different indoor lighting scenarios. The system's model iteration unit continuously optimizes model parameters based on user interaction feedback data, continuously improving recognition accuracy. (vi) Industrial applicability The device and method for multispectral optical signal acquisition and AI intelligent analysis described in this invention, relying on the Multispectral Optical Signal Deep Resolution System (MSSDAS), achieves full-dimensional intelligent analysis of optical signals and has significant industrial applicability. 1. The core hardware consists of existing mature industrial products. Multispectral sensors, optical components, industrial-grade chips, wireless communication modules, etc. can all be obtained through market procurement. The device can be mass-produced through modular integration, and the production process is simple and the cost is controllable. 2. The Multispectral Optical Signal Deep Analysis System (MSSDAS) can flexibly integrate various existing deep learning models. All algorithms can be implemented through existing software programming. The model system supports deployment and OTA online iteration in all computing scenarios, including edge devices, cloud, and computing centers. It does not rely on high-end computing equipment and is adapted to the production requirements of civilian smart devices. 3. The device and method can be flexibly customized according to different scenarios such as intelligent pet management, human health monitoring, and family companionship. The model system can be modularly split and integrated according to scenario requirements and computing power conditions, adapting to the research and development and production of various civilian intelligent devices, with broad market application scenarios; 4. The entire technical solution has a simple operation process, requires no professional personnel, meets the usage needs of civilian products, and is easily accepted by the market. In summary, this invention enables mass production and widespread application, possesses extremely high industrial applicability and commercial value, and can promote technological upgrading and product innovation in the field of intelligent devices.

Claims

1. A device for multispectral optical signal acquisition and AI intelligent analysis, characterized in that, It includes a multispectral optical signal acquisition terminal and an AI intelligent analysis module (2). The AI ​​intelligent analysis module (2) is the hardware carrier of the Multispectral Optical Signal Deep Analysis System (MSSDAS). The multispectral optical signal acquisition terminal (1) and the AI ​​intelligent analysis module (2) achieve bidirectional communication through a data transmission module. The Multispectral Optical Signal Deep Analysis System (MSSDAS) uses multispectral optical signals (300-1000nm ultraviolet / visible / near-infrared + polarized light + narrowband visible light) as input carriers, and takes optical signal feature extraction, fusion, and analysis as its core objectives. It is a comprehensive deep learning model system that can integrate various deep learning models in all dimensions, is scalable, and has unlimited computing power. This system can be modularly split, fused, or deployed according to application scenarios and computing power conditions, and supports OTA online iterative optimization. The multispectral optical signal acquisition terminal (1) includes a multispectral sensor module (1.1), an auxiliary optical component (1.2), and a data transmission and storage module (1.3). The multispectral sensor module (1.1) adopts a multispectral core with a light-sensing range covering 300-1000nm. It is paired with a polarization sensor to subdivide the visible light band into narrow bands. The auxiliary optical component (1.2) adopts an "active light source + narrowband filter" combination structure, integrates a micro dark field illumination module, and is paired with a high signal-to-noise ratio polarization sensor with a signal-to-noise ratio ≥1000:

1. The data transmission and storage module (1.3) is equipped with an industrial-grade high-speed transmission chip, configured with 16-32GB of local cache, and equipped with a lightweight wireless communication module to provide data transmission and storage support for the Multispectral Optical Signal Deep Resolution System (MSSDAS). The AI ​​intelligent analysis module (2) has a built-in data preprocessing unit (2.1), feature extraction unit (2.2), intelligent analysis unit (2.3), and model iteration unit (2.4), which is the core implementation unit of the Multispectral Optical Signal Deep Analysis System (MSSDAS). The data preprocessing unit (2.1) integrates the "digital virtual filter" AI algorithm. The feature extraction unit (2.2) can be flexibly equipped with various deep learning models to realize optical signal feature extraction. The intelligent analysis unit (2.3) is equipped with a spectral unmixing algorithm and a customizable fusion deep learning model. The model iteration unit (2.4) adopts model quantization and pruning technology to support the OTA online iteration of the Multispectral Optical Signal Deep Analysis System (MSSDAS).

2. The apparatus according to claim 1, characterized in that, The multispectral sensor module (1.1) has a light-sensing range that includes ultraviolet, full visible, and near-infrared light bands. It can capture the subtle differences in multi-band narrowband light signals and polarized light signals emitted by target organisms, providing high-quality raw input data for the Multispectral Light Signal Deep Resolution System (MSSDAS).

3. The apparatus according to claim 1, characterized in that, The data preprocessing unit (2.1) has functions such as real-time preview of spectral data, noise reduction, normalization, batch processing and channel registration. It can automatically complete format conversion, effectively filter environmental noise interference, and output high-quality spectral data sources, providing data support for feature extraction and intelligent analysis within the Multispectral Optical Signal Deep Analysis System (MSSDAS).

4. The apparatus according to claim 1, characterized in that, The spectral demixing algorithm of the intelligent analysis unit (2.3) can separate overlapping light signal features, and the customized fusion deep learning model can autonomously explore the causal relationship or strong correlation between light signals and biological states. The biological states include biological needs, emotional state, and physical health status, realizing the full-dimensional intelligent analysis of light signals by the Multispectral Light Signal Deep Analysis System (MSSDAS).

5. The apparatus according to claim 1, characterized in that, The wireless communication module of the data transmission and storage module (1.3) is used for data synchronization and OTA iteration of the Multispectral Optical Signal Deep Resolution System (MSSDAS), realizing millisecond-level optical signal data transmission without delay, and adapting to the deployment requirements of different computing power scenarios.

6. The apparatus according to claim 1, characterized in that, The model iteration unit (2.4) can deploy the customized fusion model trained in the Multispectral Optical Signal Deep Resolution System (MSSDAS) to the device's local chip, cloud server, or computing center, and continuously optimize the model system parameters through actual user data, forming a closed-loop iterative system of "data acquisition - model optimization - experience improvement".

7. A method for multispectral optical signal acquisition and AI intelligent analysis based on the device and multispectral optical signal deep resolution system (MSSDAS) according to any one of claims 1-6, characterized in that, Includes the following steps: Step S1: Accurate acquisition of multi-band optical signals. The 300-1000nm band optical signals and polarized light signals released by the target organism (3) are captured by the multi-spectral optical signal acquisition terminal (1). The auxiliary optical components (1.2) filter ambient light interference and provide raw input data for the multi-spectral optical signal deep resolution system (MSSDAS). Step S2: Spectral data preprocessing. The data transmission and storage module (1.3) transmits the raw optical signal data to the AI ​​intelligent analysis module (2). The data preprocessing unit (2.1) of the Multispectral Optical Signal Deep Analysis System (MSSDAS) performs noise reduction, normalization, and channel registration on the raw data. The effective spectral channels are filtered through the "digital virtual filter" to output a high-quality spectral data source. Step S3: Extraction of core features of optical signal. The feature extraction unit (2.2) of the Multispectral Optical Signal Deep Analysis System (MSSDAS) is equipped with the corresponding deep learning model to extract features from the preprocessed spectral sequence data and mine the core features of optical signal such as band shift, intensity change and polarization mode. Step S4: AI intelligent analysis and biological state recognition. The intelligent analysis unit (2.3) of the Multispectral Optical Signal Deep Analysis System (MSSDAS) separates the core features of the optical signal through the spectral unmixing algorithm, and uses a customized fusion deep learning model to match the correlation between the optical signal and the biological state, and outputs the recognition result. Step S5: Model Iteration Optimization. The model iteration unit (2.4) of the Multispectral Optical Signal Deep Resolution System (MSSDAS) continuously optimizes the model parameters within the system based on actual user data, thereby achieving OTA online iteration of the Multispectral Optical Signal Deep Resolution System (MSSDAS). Step S6: Identification result output and intelligent action response. Based on the identification result output by the Multispectral Optical Signal Deep Analysis System (MSSDAS), the corresponding action of the execution terminal (4) is triggered to achieve a targeted response.

8. The method according to claim 7, characterized in that, The optical signals mentioned in step S1 include ultraviolet light, visible light, and near-infrared light signals. The auxiliary optical components (1.2) achieve ambient light interference filtering through "active light source + narrowband filtering" and micro dark field illumination module, ensuring the purity and accuracy of the raw input data provided to the Multispectral Optical Signal Deep Resolution System (MSSDAS).

9. The method according to claim 7, characterized in that, The feature extraction unit (2.2) described in step S3 can be flexibly equipped with various deep learning models according to computing power conditions and accuracy requirements, directly extracting features from spectral sequence data, avoiding pixel-level computational redundancy, and improving the feature extraction efficiency and accuracy of the Multispectral Optical Signal Deep Analysis System (MSSDAS).

10. The method according to claim 7, characterized in that, The execution terminal (4) mentioned in step S6 can be flexibly configured according to the application scenario, including but not limited to the feeding device and interactive module of the smart pet robot, the health management APP for human health monitoring, and the voice module and action module of the family companion robot; the method can be adapted to multiple scenarios such as smart pet management, human health monitoring, and family smart companionship. The multispectral optical signal deep analysis system (MSSDAS) can be modularly split, integrated, or deployed according to different scenario requirements and computing power conditions (edge ​​chip / cloud server / computing center).

Images