CN-121999428-A - Flame morphology stage real-time research and judgment method, system, equipment and medium

CN121999428ACN 121999428 ACN121999428 ACN 121999428ACN-121999428-A

Abstract

A method, a system, equipment and a medium for real-time research and judgment of flame morphology stage comprise the steps of obtaining fire information to be identified, substituting the fire information to be identified into a pre-trained flame stage identification model to obtain a primary identification result, extracting a chromaticity value corresponding to the fire information by adopting a chromaticity identification algorithm, and correcting the primary identification result based on a chromaticity threshold range corresponding to the primary identification result to obtain a stage corresponding to the fire information, wherein the chromaticity threshold range is pre-trained and comprises chromaticity ranges corresponding to all stages of flame. The invention solves the problems of insufficient fire detection precision, contradiction between speed and light weight, weak anti-interference capability, poor generalization performance and the like of the traditional cable corridor by fusing bimodal technical paths of computer vision and chromaticity characteristic analysis.

Inventors

LIU CHANG
WANG DIE
ZHOU YIQI
LI PENG
YANG ZHI
LI JUNHUI
JI KUNPENG
HAN JINGSHAN
ZHAO BIN
ZHANG MINGHAO

Assignees

国网电力工程研究院有限公司

Dates

Publication Date: 20260508
Application Date: 20251120

Claims (16)

1. A flame morphology stage identification method, comprising: acquiring fire information to be identified; substituting fire information to be identified into a pre-trained flame stage identification model to obtain a primary identification result; extracting a chromaticity value corresponding to the fire information by adopting a chromaticity recognition algorithm; correcting the preliminary identification result based on the chromaticity threshold range corresponding to the preliminary identification result to obtain a stage corresponding to the fire information; the range of the chromaticity threshold is pre-trained and comprises a chromaticity range corresponding to each stage of flame.
2. The method of claim 1, further comprising, prior to the flame phase identification model and the pre-computed chromaticity threshold: Acquiring fire information historical data; dividing fire information historical data into three parts, labeling a stage label and a flame stage area for the first part of data, constructing a data set A, labeling a flame area for the second part of data, constructing a data set B, and constructing a data set C by the third part of data.
3. The method of claim 2, wherein training of the flame phase identification model comprises: Training a machine vision algorithm by the data marked with the stage labels and the flame stage areas in the data set A to obtain a trained flame stage identification model.
4. The method of claim 2, wherein training the range of chrominance thresholds comprises: Training a machine vision algorithm by the data marked with the flame region in the data set B to obtain a trained flame region identification model; Substituting the data in the data set C into a pre-trained flame region identification model, identifying the flame region, extracting the chromaticity value of the flame region by combining a chromaticity identification algorithm, and setting the chromaticity threshold range corresponding to each stage of the flame based on the chromaticity value of the extracted flame region.
5. The method of claim 4, wherein substituting the data in the data set C into the flame region identification model, identifying the flame region, extracting a chromaticity value of the flame region in combination with a chromaticity identification algorithm, and setting a chromaticity threshold range corresponding to each stage of the flame based on the extracted chromaticity value of the flame region, comprises: substituting the data in the data set C into the flame region identification model in sequence, and outputting the data marked with the flame region in sequence; extracting the chromaticity value of the flame area by adopting a chromaticity recognition algorithm, sequencing the chromaticity value according to the time sequence, and constructing a chromaticity value sequence; analyzing the chromaticity dynamic changes of different combustion stages through a chromaticity value sequence, and dividing the chromaticity threshold range corresponding to each stage of flame through statistical learning; The statistical learning comprises unsupervised clustering, probability distribution fitting and interval estimation.
6. The method of claim 1, wherein the step of correcting the preliminary identification result based on the chromaticity threshold range corresponding to the preliminary identification result to obtain the fire information corresponds to the fire information comprises: Judging the flame stage to which the colorimetric value belongs based on the colorimetric threshold value range of each stage of flame; calculating the ratio of the chromaticity value to the upper limit value of the chromaticity threshold value range corresponding to the flame stage; Judging the confidence level corresponding to the ratio and the primary identification result, if the ratio is large, correcting the primary identification result by the flame phase to which the chromaticity value belongs, taking the flame phase to which the chromaticity value belongs as the phase identification result, otherwise, taking the primary identification result as the phase identification result.
7. The method of claim 1, wherein extracting a chromaticity value corresponding to the fire information using a chromaticity recognition algorithm comprises: identifying the chromaticity value of the flame area by adopting a chromaticity identification algorithm; The flame area is framed by a pre-trained flame stage identification model when identifying fire information to be identified.
8. The method of claim 2, wherein the fire information history data is obtained based on real fire scenes simulated by an actual cable corridor scaled full-size tunnel experiment platform.
9. A flame morphology stage identification system, comprising: The parameter acquisition module is used for acquiring fire information to be identified; The primary identification module is used for substituting fire information to be identified into a pre-trained flame stage identification model to obtain a primary identification result; The chroma recognition module is used for extracting chroma values corresponding to the fire information by adopting a chroma recognition algorithm; the result correction module is used for correcting the preliminary identification result based on the chromaticity threshold range corresponding to the preliminary identification result to obtain a stage corresponding to the fire information; the range of the chromaticity threshold is pre-trained and comprises a chromaticity range corresponding to each stage of flame.
10. The system of claim 9, further comprising a parameter processing module for: Acquiring fire information historical data; dividing fire information historical data into three parts, labeling a stage label and a flame stage area for the first part of data, constructing a data set A, labeling a flame area for the second part of data, constructing a data set B, and constructing a data set C by the third part of data.
11. The system of claim 10, further comprising a model training module to: Training a machine vision algorithm by the data marked with the stage labels and the flame stage areas in the data set A to obtain a trained flame stage identification model.
12. The system of claim 10, further comprising a threshold training module to: Training a machine vision algorithm by the data marked with the flame region in the data set B to obtain a trained flame region identification model; Substituting the data in the data set C into a pre-trained flame region identification model, identifying the flame region, extracting the chromaticity value of the flame region by combining a chromaticity identification algorithm, and setting the chromaticity threshold range corresponding to each stage of the flame based on the chromaticity value of the extracted flame region.
13. The system of claim 12, wherein the specific implementation steps of substituting the data in the data set C into a pre-trained flame region identification model in the threshold training module, identifying the flame region, extracting the chromaticity value of the flame region in combination with the chromaticity identification algorithm, and setting the chromaticity threshold range corresponding to each stage of the flame based on the extracted chromaticity value of the flame region include: substituting the data in the data set C into the flame region identification model in sequence, and outputting the data marked with the flame region in sequence; extracting the chromaticity value of the flame area by adopting a chromaticity recognition algorithm, sequencing the chromaticity value according to the time sequence, and constructing a chromaticity value sequence; analyzing the chromaticity dynamic changes of different combustion stages through a chromaticity value sequence, and dividing the chromaticity threshold range corresponding to each stage of flame through statistical learning; The statistical learning comprises unsupervised clustering, probability distribution fitting and interval estimation.
14. The system of claim 9, wherein the result modification module is specifically configured to: Judging the flame stage to which the colorimetric value belongs based on the colorimetric threshold value range of each stage of flame; calculating the ratio of the chromaticity value to the upper limit value of the chromaticity threshold value range corresponding to the flame stage; Judging the confidence level corresponding to the ratio and the primary identification result, if the ratio is large, correcting the primary identification result by the flame phase to which the chromaticity value belongs, taking the flame phase to which the chromaticity value belongs as the phase identification result, otherwise, taking the primary identification result as the phase identification result.
15. The electronic equipment is characterized by comprising at least one processor and a memory, wherein the memory and the processor are connected through a bus; The memory is used for storing one or more programs; A flame morphology stage identification method as claimed in any one of claims 1 to 8 when the one or more programs are executed by the at least one processor.
16. A readable storage medium having stored thereon an execution program which, when executed, implements a flame morphology stage identification method according to any one of claims 1 to 8.

Description

Flame morphology stage real-time research and judgment method, system, equipment and medium Technical Field The invention relates to the field of fire detection in computer vision in artificial intelligence, in particular to a method, a system, equipment and a medium for real-time research and judgment of flame morphology stage. Background The traditional cable corridor fire detection technology mainly relies on the monitoring of a sensor on the environment, and whether a fire occurs or not is judged by detecting characteristic changes when the fire occurs. Such techniques typically cover four types, light-sensing, temperature-sensing, smoke detection, and gas identification. With the rise of artificial intelligence technology, a fire detection technology based on computer vision is continuously innovated, the operation state of a cable corridor facility is monitored in real time and perceived in parameters, flames and smoke are identified by analyzing image or video data, the accurate identification of early fire and the dynamic prediction of fire evolution situation are realized, and the technology link is important to guaranteeing the safety persistence and the efficiency stability of an underground power transmission system. The existing fire detection algorithm has more technical defects in a cable gallery scene, a machine vision technology does not form an effective solution in a cable tunnel flame stage prediction task, a traditional machine vision model is weak in stage prediction processing capacity, anti-interference capacity and generalization performance in the scene, a large model can provide higher detection precision but tends to have slower running speed, and a small model has fast running speed but has insufficient precision. In addition, such algorithms tend to perform poorly in strong exposure or darkness environments, and model generalization capability is also less efficient on embedded devices. Disclosure of Invention In order to solve the problems that the traditional machine vision model has weaker stage prediction processing capability, anti-interference capability and generalization performance in a cable gallery scene, a large model can provide higher detection precision but has slower running speed, and a small model has lower precision and poorer performance in a strong exposure or dark environment although the running speed is high, and the model generalization capability is also lower on embedded equipment, the invention provides a flame form stage real-time research and judgment method, which comprises the following steps: acquiring fire information to be identified; substituting fire information to be identified into a pre-trained flame stage identification model to obtain a primary identification result; extracting a chromaticity value corresponding to the fire information by adopting a chromaticity recognition algorithm; Correcting the preliminary identification result based on the chromaticity threshold range corresponding to the preliminary identification result to obtain a stage corresponding to the fire information; the range of the chromaticity threshold is pre-trained and comprises a chromaticity range corresponding to each stage of flame. Optionally, before the flame phase identification model and the pre-computed chromaticity threshold, further comprising: Acquiring fire information historical data; dividing fire information historical data into three parts, labeling a stage label and a flame stage area for the first part of data, constructing a data set A, labeling a flame area for the second part of data, constructing a data set B, and constructing a data set C by the third part of data. Optionally, training of the flame phase identification model includes: Training a machine vision algorithm by the data marked with the stage labels and the flame stage areas in the data set A to obtain a trained flame stage identification model. Optionally, training of the chromaticity threshold range includes: Training a machine vision algorithm by the data marked with the flame region in the data set B to obtain a trained flame region identification model; Substituting the data in the data set C into a pre-trained flame region identification model, identifying the flame region, extracting the chromaticity value of the flame region by combining a chromaticity identification algorithm, and setting the chromaticity threshold range corresponding to each stage of the flame based on the chromaticity value of the extracted flame region. Optionally, substituting the data in the data set C into the flame region identification model, identifying the flame region, extracting a chromaticity value of the flame region in combination with a chromaticity identification algorithm, and setting a chromaticity threshold range corresponding to each stage of the flame based on the extracted chromaticity value of the flame region, including: substituting the data in the data set C into the flame regi