Search

CN-121996984-A - Multispectral optical signal acquisition and AI intelligent analysis device and method

CN121996984ACN 121996984 ACN121996984 ACN 121996984ACN-121996984-A

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. The core defines a multispectral optical signal intelligent analysis depth network system (MSS-IDNS), which takes 300-1000nm multispectral optical signals as input carriers 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 six steps of optical signal acquisition, data preprocessing, feature extraction, AI analysis, model iteration and intelligent response, so that intelligent analysis and biological state identification of the optical signals are completed. The invention breaks through human visual perception boundary, adapts to multiple scenes, has core hardware which is the existing mature industrial product, can be produced in batch, and has extremely strong civil mass production potential and commercial value.

Inventors

  • ZHAO WEI

Assignees

  • 赵卫

Dates

Publication Date
20260508
Application Date
20260213

Claims (10)

  1. 1. The device for multispectral optical signal acquisition and AI intelligent analysis is characterized by comprising a multispectral optical signal acquisition terminal and an AI intelligent analysis module (2), wherein the AI intelligent analysis module (2) is a hardware floor carrier of a multispectral optical signal deep analysis system (MSSDAS), and the multispectral optical signal acquisition terminal (1) and the AI intelligent analysis module (2) realize two-way communication through a data transmission module; The multispectral optical signal depth analysis system (MSSDAS) takes multispectral optical signals (300-1000 nm ultraviolet/visible light/near infrared+polarized light+narrow-band visible light) as input carriers, takes optical signal characteristics extraction, fusion and analysis as core targets, can be integrated with a full-dimension, expandable and calculation-force-unbounded deep learning model comprehensive system of various deep learning models, can carry out modularized splitting, fusion or deployment according to application scenes and calculation-force conditions, and supports OTA online iterative optimization; The multispectral optical signal acquisition terminal (1) comprises a multispectral sensor module (1.1), an auxiliary optical component (1.2) and a data transmission and storage module (1.3), wherein the multispectral sensor module (1.1) adopts a multispectral movement, the sensitization range covers 300-1000nm, a polarized light sensor is matched with the multispectral optical sensor to subdivide a visible light wave band into narrow wave bands, the auxiliary optical component (1.2) adopts an 'active light source+narrow-band filtering' combined structure, a miniature dark field illumination module is integrated, a high signal-to-noise ratio polarized light sensor with a signal-to-noise ratio more than or equal to 1000:1 is matched with the auxiliary optical component (1.2), the data transmission and storage module (1.3) is provided with an industrial-level high-speed transmission chip, a 16-32GB local cache is configured, and a lightweight wireless communication module is provided for providing data transmission and storage support for the multispectral optical signal depth analysis system (MSSDAS); The intelligent AI analysis module (2) is internally provided with a data preprocessing unit (2.1), a feature extraction unit (2.2), an intelligent analysis unit (2.3) and a model iteration unit (2.4), and is a core landing unit of the multispectral optical signal deep analysis system (MSSDAS), the data preprocessing unit (2.1) integrates an 'digital virtual filter' AI algorithm, the feature extraction unit (2.2) can flexibly carry various deep learning models to realize optical signal feature extraction, the intelligent analysis unit (2.3) carries a spectral unmixing algorithm and a customizable fusion deep learning model, and the model iteration unit (2.4) adopts a model quantization and pruning technology to support OTA online iteration of the multispectral optical signal deep analysis system (MSSDAS).
  2. 2. The device according to claim 1, wherein the photosensitive range of the multispectral sensor module (1.1) comprises ultraviolet light, full visible light and near infrared light bands, can capture the subtle differences of multiband narrow-band light signals and polarized light signals released by target organisms, and provides high-quality raw input data for a multispectral light signal deep analysis system (MSSDAS).
  3. 3. The device according to claim 1, wherein the data preprocessing unit (2.1) has functions of spectral data real-time preview, noise reduction, normalization, batch processing and channel registration, can automatically complete format conversion, effectively filter ambient noise interference, output a high-quality spectral data source, and provide data support for feature extraction and intelligent analysis in a multi-spectral optical signal deep analysis system (MSSDAS).
  4. 4. The device according to claim 1, wherein the spectral unmixing algorithm of the intelligent analysis unit (2.3) can separate overlapping light signal features, and the customized fused deep learning model can autonomously mine causal or strong correlation relation between the light signals and biological states, wherein the biological states comprise biological demands, emotional states and physical health conditions, so that the full-dimension intelligent analysis of the light signals by the multispectral light signal deep analysis system (MSSDAS) is realized.
  5. 5. The apparatus according to claim 1, wherein the wireless communication module of the data transmission and storage module (1.3) is configured to perform OTA iteration of a data synchronization and multispectral optical signal deep parsing system (MSSDAS), so as to implement millisecond optical signal data transmission without delay, and adapt to deployment requirements of different computing power scenarios.
  6. 6. The apparatus according to claim 1, wherein the model iteration unit (2.4) may deploy the customized fusion model trained in the multispectral optical signal deep parsing system (MSSDAS) to a local chip of the device, a cloud server or a computing center, and form a closed loop iteration system of "data acquisition-model optimization-experience improvement" by continuously optimizing model system parameters through actual use of data by a user.
  7. 7. A multispectral optical signal acquisition and AI intelligent analysis method based on the device of any one of claims 1-6 and a multispectral optical signal depth analysis system (MSSDAS), comprising the steps of: The method comprises the steps of S1, accurately collecting multiband optical signals, capturing 300-1000nm band optical signals and polarized optical signals released by a target organism (3) through a multispectral optical signal collecting terminal (1), and filtering ambient light interference by an auxiliary optical component (1.2) to provide original input data for a multispectral optical signal deep analysis system (MSSDAS); Step S2, spectrum data preprocessing, a data transmission and storage module (1.3) transmits original optical signal data to an AI intelligent analysis module (2), a data preprocessing unit (2.1) of a multispectral optical signal depth analysis system (MSSDAS) carries out noise reduction, normalization and channel registration processing on the original data, and an effective spectrum channel is screened through a digital virtual filter to output a high-quality spectrum data source; s3, extracting core features of the optical signals, carrying a corresponding deep learning model by a feature extraction unit (2.2) of a multi-spectrum optical signal deep analysis system (MSSDAS), extracting features of the preprocessed spectrum sequence data, and excavating core features of band offset, intensity change and polarization mode of the optical signals; Step S4, AI intelligent analysis and biological state recognition, wherein an intelligent analysis unit (2.3) of a multispectral optical signal deep analysis system (MSSDAS) separates optical signal core features through a spectrum unmixing algorithm, and a customized fused deep learning model is utilized to match the association relationship between the optical signal and the biological state, so as to output a recognition result; S5, model iterative optimization, namely realizing OTA online iteration of a multi-spectrum optical signal depth analysis system (MSSDAS) by continuously optimizing model parameters of the system through actual use data of a user by a model iterative unit (2.4) of the multi-spectrum optical signal depth analysis system (MSSDAS); And S6, outputting an identification result and responding to the intelligent action, and triggering the corresponding action of the execution terminal (4) according to the identification result output by the multispectral optical signal deep analysis system (MSSDAS) to realize targeted response.
  8. 8. The method according to claim 7, wherein the optical signals in step S1 include ultraviolet, visible and near infrared optical signals, and the auxiliary optical assembly (1.2) implements ambient light interference filtering by means of an "active light source+narrow band filtering" and a micro dark field illumination module, so as to ensure the purity and accuracy of the original input data provided for the multi-spectral optical signal depth analysis system (MSSDAS).
  9. 9. The method according to claim 7, wherein the feature extraction unit (2.2) in step S3 can flexibly carry various deep learning models according to the computational power conditions and the precision requirements, directly extract features for the spectrum sequence data, avoid pixel-level operation redundancy, and improve the feature extraction efficiency and the precision of the multispectral optical signal deep analysis system (MSSDAS).
  10. 10. The method of claim 7, wherein the execution terminal (4) in the step S6 is flexibly configured according to application scenarios, including but not limited to a feeding device, an interaction module, a health management APP for human health monitoring, a voice module and an action module of a home companion robot of the smart pet robot, and the method is adaptable to multiple scenarios such as smart pet management, human health monitoring and home smart companion, and the multispectral optical signal deep analysis system (MSSDAS) can be split, fused or deployed in a modularized manner according to different scenario requirements and calculation conditions (end chip/cloud server/computing center).

Description

Multispectral optical signal acquisition and AI intelligent analysis device and method Technical Field The invention relates to the technical fields of optical signal detection, multispectral analysis and artificial intelligent deep learning, in particular to a device and a method for multispectral optical signal acquisition and AI intelligent analysis, which can be widely applied to intelligent pet management, human health monitoring, home intelligent companion and other multi-scene life body state identification and intelligent response, and the core realizes full-dimension intelligent analysis of optical signals by means of a multispectral optical signal deep analysis system (MSSDAS). Background The human visual system forms an optical signal receiving system taking red, green and blue cone cells as cores through long-term evolution, can only capture optical signals of visible light wave bands and form visual images through brain mixing analysis, cannot sense signals of wave bands such as ultraviolet light, infrared light, polarized light and the like, cannot distinguish narrow wave band differences in the visible light core wave bands, and has congenital sensing boundary limitation. In the prior art, related products such as life body state identification, health monitoring, intelligent service robots and the like are mainly used for carrying out state judgment and instruction response based on signals such as images, voices, limb actions and the like. The method has the advantages that the signals are easy to forge and interfere, inherent defects of subjective camouflage are caused, the judging precision is low, the scene suitability is poor, part of spectrum detection equipment can only realize optical signal acquisition of a single wave band or a limited wave band, the optical signal acquisition is not deeply fused with an artificial intelligence technology, the accurate analysis of the optical signal and the intelligent association of a life body state cannot be realized, meanwhile, the conventional AI analysis model is a single network model, only the feature extraction of a specific dimension can be realized, the association relation between the external features summarized by human and the state is solidified into an algorithm depending on the repeated carving of human experience, the potential inherent association between the optical signal and the life body state cannot be independently mined, and the full-dimensional analysis is carried out on various optical signal features by a unified deep learning model system, so that the upper performance limit is limited by the cognition and experience boundary of human. Aiming at the technical defects, a novel technical scheme capable of breaking through human visual perception boundaries, realizing accurate collection of multi-band narrow-band optical signals and autonomously excavating optical signals and internally associating life body states by combining a unified deep learning model system cannot meet the core requirements of intelligent equipment on signal authenticity, detection accuracy and analysis intellectualization, so that a novel technical scheme is needed to solve the problems. Disclosure of Invention Object of the invention The invention aims to overcome the defects of the prior art, provides a device and a method for multispectral light signal acquisition and AI intelligent analysis, breaks through human visible light perception boundary, realizes accurate acquisition of multiband narrow-band light signals of 300-1000nm ultraviolet light, visible light, near infrared light and polarized light, and realizes accurate identification and intelligent response of living body requirements, emotion states and physical health conditions by autonomously excavating a causal relation or a strong correlation relation between the light signals and a living body state by means of a multispectral light signal depth analysis system (MSSDAS, abbreviation of Multi-SPECTRAL SIGNAL INTELLIGENT Decoding Deep Network System), thereby solving the technical problems of easy counterfeiting of signals, low detection precision, analysis dependence on human experience and non-unified depth learning model system in the prior art. (II) technical scheme In order to achieve the above purpose, the invention firstly defines a multispectral optical signal deep analysis system (MSSDAS) which takes multispectral optical signals (300-1000 nm ultraviolet/visible light/near infrared+polarized light+narrow-band visible light) as unique input carriers, takes optical signal characteristics extraction, fusion and analysis as core targets, can be fused to comprise but not limited to a one-dimensional convolutional neural network (1D-CNN/EDCNN), a one-dimensional residual network (1D-ResNet), a gate-controlled loop unit (GRU), a Time Convolutional Network (TCN), an attention mechanism network and other deep learning model comprehensive systems with full dimension, expandable and u