CN-122024309-A - Visual identification-based flue gas dioxin sample collection and supervision method and system

CN122024309ACN 122024309 ACN122024309 ACN 122024309ACN-122024309-A

Abstract

The application discloses a method and a system for collecting and supervising a flue gas dioxin sample based on visual identification, wherein the method for collecting and supervising the flue gas dioxin sample based on visual identification comprises the steps of constructing a flue gas dioxin sample collecting field model; training and optimizing the flue gas dioxin sample collection site model to obtain a mature model, deploying the mature model on an actual flue gas dioxin sample collection site to dynamically identify the flue gas dioxin sample sampling behavior, and carrying out reflux collection on the data of the dynamically identified flue gas dioxin sample sampling behavior to continuously iterate and update a database. The method has the advantages of realizing full-time and full-direction monitoring of the flue gas dioxin sample collection process, timely identifying and early warning the abnormal situation and improving the sample collection standardization and accuracy.

Inventors

CHEN SIYANG
LIU YIYANG
ZHONG YONGCAI

Assignees

深圳希诺检测有限公司

Dates

Publication Date: 20260512
Application Date: 20251229

Claims (10)

1. The method for collecting and supervising the fume dioxin sample based on visual recognition is characterized by comprising the following steps of: constructing a flue gas dioxin sample collection site model; training and optimizing the flue gas dioxin sample collection field model to obtain a mature model; Disposing the mature model on the actual flue gas dioxin sample collection site to dynamically identify the sampling behavior of the flue gas dioxin sample, and And (3) carrying out reflux collection on the data of the sampling behaviors of the dynamically identified flue gas dioxin samples, and continuously and iteratively updating the database.
2. The visual identification-based flue gas dioxin sample collection and supervision method according to claim 1, wherein the method comprises the following steps of: The construction of the flue gas dioxin sample collection field model comprises the following steps: Collecting site data and modeling to create a site three-dimensional model of the flue gas dioxin sample collection; Collecting standard flue gas dioxin sample sampling requirement information; Collecting sampling requirement information of a non-normative flue gas dioxin sample; marking the normalized flue gas dioxin sample sampling requirement information on the flue gas dioxin sample collection site three-dimensional model to finish normalized operation node marking, marking the non-normalized flue gas dioxin sample sampling requirement information on the flue gas dioxin sample collection site three-dimensional model to finish non-normalized operation node marking so as to generate a typical flue gas dioxin sample collection site model, and And simulating a typical flue gas dioxin sample acquisition field model by using simulation software to generate a simulation video of a sampling field simulation scene, wherein the sampling field simulation scene comprises a normative sample acquisition simulation scene and a non-normative sample acquisition simulation scene.
3. The visual identification-based flue gas dioxin sample collection and supervision method according to claim 2, wherein the method comprises the following steps of: the generating a simulation video of a sampling field simulation scene comprises the following steps: Dividing the generated simulation video of the standardized sample acquisition simulation scene according to a sampling process to form a plurality of video clips; marking the personnel behaviors and the object states contained in the video clips in corresponding scenes, and And storing the marked video fragments into a data set, and randomly dividing the video fragments into a training set, a testing set and a verification set according to a preset proportion.
4. The visual identification-based flue gas dioxin sample collection and supervision method according to claim 3, wherein the method comprises the following steps of: The training and tuning of the flue gas dioxin sample collection site model to obtain a mature model comprises the following steps: The method comprises the steps of training a flue gas dioxin sample collection site model by using a training set, monitoring and optimizing performance of the flue gas dioxin sample collection site model by using a verification set in the training process, evaluating the flue gas dioxin sample collection site model by using a test set after training, calculating average precision map and Recall ratio Recall of the flue gas dioxin sample collection site model on the test set, and judging that the flue gas dioxin sample collection site model meets the standard if the average precision map is more than 95% and the Recall ratio Recall is more than 99%, and judging that the standard flue gas dioxin sample collection site model is a mature model.
5. The visual recognition-based flue gas dioxin sample collection and supervision method according to any one of claims 1 to 4, wherein: the disposing of the mature model on the actual flue gas dioxin sample collection site to dynamically identify the flue gas dioxin sample sampling behavior comprises: Monitoring a video of a flue gas dioxin sampling site in real time; analyzing the monitoring video acquired in real time through the mature model, identifying the sampling information of the flue gas dioxin contained in the monitoring video, and identifying whether the sampling behavior of the flue gas dioxin sample is standard or not, and And reminding and recording according to the identification result.
6. The visual identification-based flue gas dioxin sample collection and supervision method according to claim 5, wherein the method comprises the following steps of: the identifying whether the sampling behavior of the flue gas dioxin sample is standard comprises the following steps: if the mature model detects that the state of the article and the personnel behavior in the current flue gas dioxin sample collection site are completely consistent with the state of the article and the personnel behavior contained in the database, the flue gas dioxin sample collection site is identified as known information; Comparing the sampling rule with the known information according to the predefined sampling rule of the normalized smoke dioxin sample and the predefined sampling rule of the non-normalized smoke dioxin sample, identifying the sampling operation as normalized sampling operation if the sampling rule of the normalized smoke dioxin sample accords with the predefined sampling rule of the normalized smoke dioxin sample, and identifying the sampling operation as non-normalized sampling operation if the sampling rule of the normalized smoke dioxin sample is the same as the predefined sampling rule of the non-normalized smoke dioxin sample, and reminding and recording.
7. The visual identification-based flue gas dioxin sample collection and supervision method according to claim 6, wherein the method comprises the following steps of: Whether discernment flue gas dioxin sample sampling action is standard still includes: If the mature model detects that the object state and the personnel behavior in the current flue gas dioxin sample sampling site are not completely consistent with the object state and the personnel behavior contained in the database, extracting features of object state and personnel behavior images in the current flue gas dioxin sample sampling site monitoring video, and comparing the feature extracted features with sampling images in the database to obtain similarity; According to predefined rules such as normative flue gas dioxin sampling information and category thereof and non-normative flue gas dioxin sampling information and category rules thereof, classifying each article state and personnel behavior related to the current flue gas dioxin sampling into normative categories and non-normative categories according to the similarity, and reminding and recording if the similarity of each article state and the non-normative category related to the current flue gas dioxin sampling is greater than a similarity threshold interval or the similarity of each article state and the normative category related to the current flue gas dioxin sampling is less than a similarity threshold interval.
8. The visual identification-based flue gas dioxin sample collection and supervision method according to claim 7, wherein the method comprises the following steps of: The data reflux collection of the dynamically identified sampling behaviors of the flue gas dioxin sample is continuously iterated to update the database, and the method comprises the following steps: If the non-normative category similarity of each article state and personnel behavior related to the current smoke dioxin sampling is smaller than a similarity threshold interval, marking the current monitoring video and corresponding article state and personnel behavior information thereof, storing the marked monitoring video and corresponding article state and personnel behavior information in a database, taking the marked monitoring video and corresponding article state and personnel behavior information as potential novel non-normative condition samples, regularly rechecking and analyzing the novel non-normative condition samples, and judging whether the novel non-normative condition samples belong to novel non-normative behavior modes or abnormal article state conditions or not; If the normalized category similarity of each article state and personnel behavior related to the current smoke dioxin sampling is larger than a similarity threshold interval, marking the current monitoring video and corresponding article state and personnel behavior information thereof, storing the marked monitoring video and corresponding article state and personnel behavior information in a database, taking the marked monitoring video and corresponding article state and personnel behavior information as potential novel normalized condition samples, regularly rechecking and analyzing the novel normalized condition samples, judging whether the novel normalized condition samples belong to novel normalized behavior modes or article state compliance, supplementing the novel normalized condition samples to normalized smoke dioxin sampling rules if the novel normalized condition samples are judged to be yes, updating the database content, and deleting the corresponding monitoring video and corresponding article state and personnel behavior information thereof from the database if the novel normalized condition samples are judged to be no.
9. The visual identification-based flue gas dioxin sample collection and supervision method according to claim 8, wherein the method comprises the following steps of: if the non-normative class similarity of each article state and personnel behavior related to the current flue gas dioxin sampling is larger than the similarity threshold interval, or the normative class similarity of each article state and personnel behavior related to the current flue gas dioxin sampling is smaller than the similarity threshold interval, then: comparing the image of the monitoring video with the corresponding image in the database, extracting static characteristics, acquiring the characteristics of a flue gas dioxin sampling environment and flue gas dioxin sampling equipment, outputting a static difference result, and storing the result into the database; extracting dynamic characteristics from the behaviors of sampling personnel in a monitoring video, dividing action units according to process types, constructing action behavior sequences, comparing the action sequences with normalized dioxin sampling action sequences in a database, checking whether action deletion, order reversal or additional insertion abnormal actions exist, outputting dynamic difference results, and storing the results in the database; The method comprises the steps of calling cases that the similarity of the non-normative category of each article state and personnel behavior related to the historical smoke dioxin sampling in a database is larger than a similarity threshold interval, or the similarity of the normative category of each article state and personnel behavior related to the smoke dioxin sampling is smaller than a similarity threshold interval, comparing various characteristics of a current case with those of a historical case, analyzing the similarity and the difference of the article state, the personnel behavior and the environmental factors, and referring to a final judging result and a processing mode of the historical case to obtain a historical comparison result; According to the static difference result and the dynamic difference result, the non-normative category similarity or the normative category similarity, the historical comparison result, carrying out risk classification on the current case, and judging whether the current case belongs to non-normative behaviors or whether the article state is in an abnormal condition; The method comprises the steps of storing the cases belonging to the abnormal condition of the nonstandard behavior or the article state into a database, and marking the cases, and Based on the newly added non-standard behavior or the abnormal cases of the article state in the database, a machine learning algorithm or a manual evaluation strategy is applied to dynamically adjust the similarity threshold interval.
10. A smoke dioxin sample collection and supervision system based on visual recognition, for implementing the smoke dioxin sample collection and supervision method based on visual recognition according to any one of claims 1 to 9, comprising: the three-dimensional modeling module is used for constructing a flue gas dioxin sample acquisition field model; the training and optimizing module is used for training and optimizing the flue gas dioxin sample collection field model so as to obtain a mature model; A deployment identification module for deploying the mature model on the actual flue gas dioxin sample collection site to dynamically identify the sampling behavior of the flue gas dioxin sample, and And the data management module is used for carrying out reflux collection on the data of the sampling behaviors of the dynamically identified flue gas dioxin samples, and continuously and iteratively updating the database.

Description

Visual identification-based flue gas dioxin sample collection and supervision method and system Technical Field The application belongs to the technical field of environment detection, and particularly relates to a smoke dioxin sample collection and supervision method and system based on visual identification. Background Fixed pollution sources such as waste incineration power plants have flue gas dioxin, and flue gas dioxin samples need to be collected. Dioxin is a pollutant with extremely strong carcinogenicity, teratogenicity and severe toxicity, and the content of dioxin in smoke is extremely low, so that the sample collection process needs to be highly accurate and standardized. At present, the collection of the dioxin sample in the flue gas generally needs to last for more than two hours, and in industries with special requirements such as garbage incineration, three samples are required to be collected in parallel, and the total time is at least six hours. Sampling personnel need work for a long time in high-temperature, narrow and high-altitude operation environment, and the working condition is abominable. However, the existing flue gas dioxin sample collection process is highly dependent on manual operation, lacks an effective on-line monitoring means, and the sampling personnel often keep away from the direct supervision range of the detection mechanism to which the sampling personnel belong, and also lacks an effective supervision capability on the sample collection quality for pollution source enterprises themselves. This results in difficulty in tracing or proving out the irregular operations that may exist during the sampling process when the accuracy and reliability of the final detection data are affected. In recent years, although detection institutions, regulatory authorities and sewage enterprises begin to pay attention to standardization of sample collection processes and take a series of on-site monitoring measures, such as taking pictures of time and place by using a watermark camera, taking on-site sampling videos by wearing a law enforcement recorder, and installing real-time video monitoring and shooting on-site sampling videos, the methods still have obvious defects. Firstly, the hysteresis is strong, most monitoring measures are post verification, the sampling process is irreversible when a problem is found, secondly, the efficiency is low, law enforcement recorders and monitoring videos need to consume a large amount of manpower to examine each frame of picture one by one, thirdly, the monitoring measures are passively responded, the review is usually started after the detection result is questioned, and the irregular operation cannot be actively prevented. Disclosure of Invention Aiming at the problems, the application provides a smoke dioxin sample collection monitoring method and system based on visual identification, which aim to realize the full-time and full-direction monitoring of a sample collection process, timely identify and early warn the abnormal situation, improve the standardization and accuracy of sample collection, effectively manage the monitoring image and environment of a sample collection site, and early warn and record when the monitoring image and environment deviate from the standardized situation, thereby solving the problems of lag, inefficiency and the like of the sample collection monitoring means in the prior art. Specifically, the application provides a smoke dioxin sample collection and supervision method based on visual identification, which comprises the following steps: constructing a flue gas dioxin sample collection site model; training and optimizing the flue gas dioxin sample collection field model to obtain a mature model; Disposing the mature model on the actual flue gas dioxin sample collection site to dynamically identify the sampling behavior of the flue gas dioxin sample, and And (3) carrying out reflux collection on the data of the sampling behaviors of the dynamically identified flue gas dioxin samples, and continuously and iteratively updating the database. Further, the constructing a flue gas dioxin sample collection site model includes: Collecting site data and modeling to create a site three-dimensional model of the flue gas dioxin sample collection; Collecting standard flue gas dioxin sample sampling requirement information; Collecting sampling requirement information of a non-normative flue gas dioxin sample; marking the normalized flue gas dioxin sample sampling requirement information on the flue gas dioxin sample collection site three-dimensional model to finish normalized operation node marking, marking the non-normalized flue gas dioxin sample sampling requirement information on the flue gas dioxin sample collection site three-dimensional model to finish non-normalized operation node marking so as to generate a typical flue gas dioxin sample collection site model, and And simulating a typical flue gas dioxin sample acquisi