CN-121682227-B - Target detection double-threshold self-adaptive regulation and control method based on nerve feedback state machine

CN121682227BCN 121682227 BCN121682227 BCN 121682227BCN-121682227-B

Abstract

The invention discloses a target detection double-threshold self-adaptive regulation and control method based on a nerve feedback state machine, belongs to the technical field of brain-computer fusion, and solves the problems of low synergistic efficiency and poor robustness of the existing brain-computer fusion system in a dynamic environment. The method comprises the steps of carrying out target detection on an image flow task, outputting a target detection result of each target in each frame of image, wherein the target detection result comprises a category label of the target and a confidence score corresponding to the category label, receiving an electroencephalogram signal generated by an operator in the process of responding to the image flow task, carrying out multidimensional feature extraction on the electroencephalogram signal, fusing the result of multidimensional feature extraction to obtain a comprehensive quality index, carrying out self-adaptive regulation and control on a target detection double threshold according to the relation between the comprehensive quality index and a high threshold and a low threshold of an electroencephalogram quality state by a nerve feedback state machine, and respectively judging and outputting each target according to the relation between the self-adaptive regulated target detection double threshold and the confidence score.

Inventors

ZHANG LIJIAN
LIU YIXI
BAI PENGYING
HAN JIN
LIU HAO
Fan Xinan
CHEN YUANFANG
YI WEIBO
LIU YANG

Assignees

北京机械设备研究所

Dates

Publication Date: 20260508
Application Date: 20260205

Claims (7)

1. A target detection dual-threshold adaptive regulation and control method based on a nerve feedback state machine, which is characterized by comprising the following steps: Performing target detection on the image streaming task, and outputting a target detection result of each target in each frame of image, wherein the target detection result comprises a category label of the target and a confidence score corresponding to the category label; receiving an electroencephalogram signal generated by an operator in the process of responding to an image flow task, and extracting multidimensional features of the electroencephalogram signal; the multi-dimensional feature extraction comprises amplitude, signal-to-noise ratio and information entropy, wherein the signal-to-noise ratio Expressed as: (1) Wherein, the To analyze the average power of the electroencephalogram signal over a time window, Mean power of the brain electrical signal during pre-stimulation baseline; The nerve feedback state machine carries out self-adaptive regulation and control on the target detection double threshold according to the relation between the comprehensive quality index and the high and low thresholds of the brain electricity quality state; the target detection double threshold value is adaptively regulated and controlled, and the following steps are executed: Order the The comprehensive quality index is represented by the formula, 、 Respectively representing high and low thresholds of the brain electrical quality state; When (when) When the target detection threshold upper limit is regulated down, the target detection threshold lower limit is synchronously regulated up; When (when) When the target detection threshold upper limit is increased, the target detection threshold lower limit is synchronously decreased; When (when) When the target detection double threshold is maintained unchanged; according to the relation between the target detection double threshold value and the confidence coefficient score after self-adaptive regulation, respectively judging and outputting each target; respectively judging and outputting each target, and executing: if the confidence coefficient score C meets C not less than Judging the current target as a confident target, and outputting a target detection result of the current target; if the confidence coefficient score C meets C less than or equal to Judging that the current target is a confident non-target, and directly filtering; If the confidence score C satisfies <C< Judging the current target as a difficult target; Wherein, the 、 Respectively representing the lower limit and the upper limit of the target detection threshold after self-adaptive regulation.
2. The method for adaptive modulation and control of object detection dual threshold based on a neurofeedback state machine according to claim 1, wherein when And when the target detection threshold upper limit is regulated down, the target detection threshold lower limit is synchronously regulated up, and the following steps are executed: Target detection threshold upper limit after self-adaptive regulation Expressed as: (2) Wherein, the For the current target detection threshold upper limit, A first adjustment step for a target detection threshold; target detection threshold lower limit after self-adaptive regulation Expressed as: (3) Wherein, the For the current target detection threshold lower limit, A second adjustment step for the target detection threshold.
3. The method for adaptive modulation and control of object detection dual threshold based on a neurofeedback state machine according to claim 2, wherein when And when the target detection threshold upper limit is adjusted up, the target detection threshold lower limit is adjusted down synchronously, and the steps are executed: Target detection threshold upper limit after self-adaptive regulation Expressed as: (4) Wherein, the A third adjustment step for the target detection threshold; target detection threshold lower limit after self-adaptive regulation Expressed as: (5) Wherein, the A fourth adjustment step for the target detection threshold.
4. A neural feedback state machine based target detection dual-threshold adaptive regulation and control method according to any one of claims 1-3, wherein the receiving the electroencephalogram signal generated by the operator in the process of responding to the image flow task performs multidimensional feature extraction on the electroencephalogram signal, and performs: Receiving an electroencephalogram signal generated by an operator in the process of responding to an image flow task, and preprocessing the received electroencephalogram signal to obtain a preprocessed electroencephalogram signal; and carrying out multidimensional feature extraction on the preprocessed electroencephalogram signals.
5. The method for adaptive adjustment and control of target detection based on a neural feedback state machine according to claim 4, wherein the step of fusing the results of the multi-dimensional feature extraction to obtain a comprehensive quality index is performed by: And respectively carrying out normalization processing on the extracted amplitude, the signal-to-noise ratio and the information entropy, and carrying out weighted fusion on the amplitude, the signal-to-noise ratio and the information entropy after normalization processing to obtain the comprehensive quality index.
6. The method for adaptive dual threshold modulation for target detection based on a biofeedback state machine as claimed in claim 5, wherein the overall quality index is Expressed as: (6) Wherein, the 、、 Weight coefficients respectively representing amplitude, signal-to-noise ratio and information entropy; 、、 Respectively representing the amplitude, the signal-to-noise ratio and the information entropy after normalization processing.
7. The method for adaptive modulation and control of object detection based on a neurofeedback state machine as claimed in claim 6, wherein, 。

Description

Target detection double-threshold self-adaptive regulation and control method based on nerve feedback state machine Technical Field The invention relates to the technical field of brain-computer fusion, in particular to a target detection double-threshold self-adaptive regulation and control method based on a nerve feedback state machine. Background The brain-computer fusion technology aims at realizing deep synergy of human brain intelligence and machine intelligence by combining a brain-computer interface and artificial intelligence, and constructing a hybrid enhanced intelligent system with environmental adaptability and decision generalization capability. The brain-computer fusion system exhibits significant advantages in target detection tasks such as medical diagnosis, automatic driving, and the like, which have extremely high requirements on real-time and accuracy. The current brain-computer fusion target detection system mainly adopts a serial, parallel or serial-parallel combined architecture. In a serial architecture, an artificial intelligence system is used for initially detecting and screening images, then a suspicious target is handed to an operator for final discrimination, or the operator sends out an instruction through a brain-computer interface, and the artificial intelligence is responsible for execution. The architecture is fixed in flow and lacks real-time interaction and state adaptation between human and machine. In the parallel architecture, the artificial intelligence and brain-computer interface independently process information and output a final result in a weighting fusion mode and the like, wherein the mode is difficult to dynamically adjust according to the task process and the human brain state. The serial-parallel combined architecture is attempted to integrate the advantages of the two, but still does not fundamentally solve the problem of insufficient self-adaptive capacity of the system. The prior art has the following main defects: 1) The static task allocation and parameter setting are that key parameters such as a confidence threshold value, a learning rate and the like of an artificial intelligent module in the system are usually preset fixed values, and cannot be dynamically adjusted according to real-time task loads, environment changes and nerve states of operators. 2) The existing system has more single-way instruction transmission from brain to machine or from machine to brain, and lacks a central control mechanism capable of evaluating the state of a human-computer double system in real time and performing closed-loop tuning according to the state. 3) The cooperative efficiency of the system is limited because of parameter solidification and lack of feedback, the system cannot fully exert the automatic processing capability of artificial intelligence when the attention of an operator is concentrated and the quality of an electroencephalogram signal is high, and the system robustness can not be ensured by improving a machine decision threshold when the fatigue of the operator or the environmental interference is large, so that the cooperative amplification of 1+1>2 can not be realized on the overall performance. In recent years, the reasoning interfaces of YOLOv and other advanced target detection models support dynamic adjustment of core parameters such as confidence threshold values, and the like, which provides a technical basis for constructing an adaptive brain-computer fusion system. Meanwhile, the time domain and frequency domain characteristics of event related potential (such as P300 component) in the brain electrical signal are proved to be highly correlated with the attention level, cognitive load and decision certainty of an individual, and can be used as a reliable biological nerve feedback index for reflecting the current cooperative state of the system. Therefore, there is a need for a method for dynamically and adaptively controlling key parameters of an artificial intelligence system by using quantifiable electroencephalogram signal characteristics as real-time nerve feedback through a specific closed-loop intelligent interaction module, namely a nerve feedback state machine, so as to construct a truly intelligent, robust and efficient brain-computer collaborative sensing and decision-making system. Disclosure of Invention In view of the above analysis, the embodiment of the invention aims to provide a target detection dual-threshold self-adaptive regulation and control method based on a nerve feedback state machine, which is used for solving the problems of low cooperative efficiency and poor robustness of the existing brain-computer fusion system in a dynamic environment. The invention provides a target detection double-threshold self-adaptive regulation and control method based on a nerve feedback state machine, which comprises the following steps: Performing target detection on the image streaming task, and outputting a target detection resu