CN118887615A - A thermal power plant security device with artificial intelligence identity authentication - Google Patents

Abstract

The invention relates to the technical field of security, in particular to a thermal power plant security device for artificial intelligent identity authentication, which comprises an image acquisition module, a security module and a security module, wherein the image acquisition module is used for acquiring image information of people entering and exiting in the thermal power plant; the image processing module is used for preprocessing the image information to obtain facial image information; the recognition module is used for extracting face features from the face image information and matching the face features with known features in the database to obtain a recognition result; the track analysis module is used for tracking and recording the moving path of unidentified personnel in the thermal power plant; and the alarm module is used for triggering an alarm when the moving path is found to relate to the sensitive area. Therefore, the identity of the mobile personnel in the thermal power plant can be automatically monitored in real time in a video monitoring mode, and the monitoring effect is better.

Description

A thermal power plant security device with artificial intelligence identity authentication

Technical Field

The present invention relates to the field of security technology, and in particular to a thermal power plant security device with artificial intelligence identity authentication.

Background Art

The security system of a thermal power plant is a key component to ensure the safe operation of the power plant, integrating modern information technology and security management practices. The system usually covers multiple subsystems such as video surveillance, intrusion alarm, access control, perimeter protection and fire warning, and realizes comprehensive monitoring and security management of the entire plant area through intelligent means.

Among them, the existing access control is mainly carried out by means of access control cards and identity information cards. The security effect of this method is limited, and corresponding monitoring cannot be carried out in non-access control areas.

Summary of the invention

The purpose of the present invention is to provide a thermal power plant security device with artificial intelligence identity authentication, which aims to automatically monitor the identities of mobile personnel in the thermal power plant in real time by means of video surveillance, thereby achieving better monitoring effects.

To achieve the above-mentioned purpose, the present invention provides a thermal power plant security device with artificial intelligence identity authentication, comprising an image acquisition module, which is arranged in the thermal power plant to collect image information of people entering and leaving;

An image processing module is used to pre-process the image information to obtain facial image information;

The recognition module is used to extract facial features from facial image information and match them with known features in the database to obtain recognition results;

Trajectory analysis module, used to track and record the movement paths of unidentified personnel in the thermal power plant;

The alarm module is used to trigger an alarm when it is found that the moving path involves sensitive areas.

Among them, the thermal power plant security device with artificial intelligence identity authentication also includes a user interface module, and the user interface module is used for security management personnel to view real-time monitoring images.

Wherein, the image acquisition module includes an image acquisition unit, an image transmission unit and an image storage unit;

The image acquisition unit is used to be distributed in the thermal power plant to collect image information of people entering and leaving;

The image transmission unit is used to transmit the collected image information to the host computer;

The image storage unit is used to store image information.

Wherein, the image processing module includes a correction unit, an enhancement unit, a noise reduction unit and a facial image extraction unit;

The correction unit is used to correct the image to eliminate lens distortion;

The enhancement unit is used to enhance the contrast and brightness of the image;

The noise reduction unit is used to remove noise in the image by using Gaussian filtering;

The facial image extraction unit is used to extract facial sub-images from the image after noise removal using a Haar cascade classifier.

Wherein, the facial image extraction unit includes a loading subunit, a scanning subunit, a screening subunit, and a cropping subunit;

The loading subunit is used to load the pre-trained Haar cascade classifier model from the OpenCV library;

The scanning subunit is used to scan the preprocessed image using the loaded Haar cascade classifier;

The screening subunit is used to set a threshold to filter out candidate areas that do not meet the conditions and retain the rectangular frame representing the face;

The cropping subunit is used to crop a facial sub-image from the original image according to the rectangular frame.

Wherein, the recognition module includes a high-dimensional feature extraction unit, a similarity calculation unit and a matching unit;

The high-dimensional feature extraction unit is used to extract high-dimensional feature vectors on the cropped face image using the VGGFace algorithm;

The similarity calculation unit is used to calculate the Euclidean distance between feature vectors;

The matching unit is used to set a Euclidean distance threshold. When the similarity between the newly extracted feature and the feature of a person in the database exceeds this threshold, it is determined that the match is successful, that is, the identity of the person is successfully identified; if the match fails, the person is marked as unknown.

Wherein, the trajectory analysis module includes a temporary ID allocation unit, a cross-shot matching unit and a trajectory generation unit;

The temporary ID allocation unit is used to allocate a temporary ID to the person when the person is unknown;

The cross-lens matching unit is used to match personnel across cameras using a DeepSort algorithm, and obtain the personnel position based on the camera position;

The trajectory generating unit is used to construct the movement trajectory of each unidentified person by using the assigned temporary ID and the person's position.

The specific steps of using the DeepSort algorithm to match pedestrians across cameras include:

The pedestrian image is encoded using a feature extraction network to generate a fixed-length vector representing the pedestrian's appearance;

Find the most similar match by calculating the cosine similarity of pedestrian feature vectors between different frames;

Predict the position of pedestrians in the next frame to reduce matching loss caused by short-term occlusion.

The invention discloses a thermal power plant security device with artificial intelligence identity authentication. The image acquisition module is deployed at all key entrances, exits and important internal channels of the thermal power plant. The camera with high definition and wide angle lens is used to capture dynamic image information of people entering and leaving the thermal power plant without blind spots around the clock. These cameras can maintain high-quality image output under various lighting conditions and complex environments to ensure that the collected images are clear and usable. The image processing module performs a series of preprocessing operations on the received original image information, such as denoising, brightness and contrast adjustment, image enhancement, etc., to optimize the image quality. Then, the advanced image segmentation technology is used to accurately separate the face image from the complex background to prepare for the subsequent face recognition. The recognition module uses the face recognition technology with deep learning algorithm as the core to accurately extract the face feature points from the processed face image, such as the geometric position and feature value of the face contour, eyes, nose, mouth, etc. These features are then quickly and accurately matched with the face feature templates of known employees or authorized visitors stored in a highly encrypted database to achieve identity authentication. For faces that cannot be matched, the system will mark them as unrecognized persons. The trajectory analysis module uses intelligent video analysis technology, and the system can continuously track the activity trajectories of unidentified or unauthorized personnel in the thermal power plant. Through multi-camera linkage and spatiotemporal analysis, a detailed movement path report is formed. The alarm module is used to trigger the alarm mechanism immediately once the movement path of unidentified personnel touches the preset sensitive areas, such as the control room, chemical warehouse or important equipment area. This includes but is not limited to sound and light alarms, sending instant alarm information to the security center, and even directly contacting security personnel to ensure a quick response. This artificial intelligence identity authentication thermal power plant security device realizes efficient identification of personnel identities, close monitoring of sensitive areas and timely response, making the monitoring effect better.

BRIEF DESCRIPTION OF THE DRAWINGS

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

FIG1 is a structural diagram of a thermal power plant security device with artificial intelligence identity authentication according to the first embodiment of the present invention.

FIG. 2 is a structural diagram of an image acquisition module according to a second embodiment of the present invention.

FIG. 3 is a structural diagram of an image processing module according to a second embodiment of the present invention.

FIG. 4 is a structural diagram of a facial image extraction unit according to a second embodiment of the present invention.

FIG. 5 is a structural diagram of a recognition module according to a second embodiment of the present invention.

FIG. 6 is a structural diagram of a trajectory analysis module according to a second embodiment of the present invention.

FIG. 7 is a structural diagram of an alarm module according to a second embodiment of the present invention.

Image acquisition module 101, image processing module 102, recognition module 103, trajectory analysis module 104, alarm module 105, user interface module 106, image acquisition unit 201, image transmission unit 202, image storage unit 203, correction unit 204, enhancement unit 205, noise reduction unit 206, facial image extraction unit 207, loading subunit 208, scanning subunit 209, screening subunit 210, cropping subunit 211, high-dimensional feature extraction unit 212, similarity calculation unit 213, matching unit 214, temporary ID allocation unit 215, cross-shot matching unit 216, trajectory generation unit 217, marking unit 218, comparison unit 219, alarm unit 220.

DETAILED DESCRIPTION

Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and should not be construed as limiting the present invention.

First embodiment

Please refer to Figure 1. The present invention provides a thermal power plant security device with artificial intelligence identity authentication, including an image acquisition module 101, which is arranged in the thermal power plant to collect image information of people entering and leaving; an image processing module 102, which is used to pre-process the image information to obtain facial image information; a recognition module 103, which is used to extract facial features from the facial image information and match them with known features in a database to obtain recognition results; a trajectory analysis module 104, which is used to track and record the movement path of unidentified people in the thermal power plant; an alarm module 105, which is used to trigger an alarm when it is found that the movement path involves a sensitive area.

The artificial intelligence identity authentication thermal power plant security device also includes a user interface module 106, and the user interface module 106 is used for security management personnel to view real-time monitoring images.

In this embodiment, the image acquisition module 101 is deployed at all key entrances, exits and important internal channels of the thermal power plant, and uses high-definition, wide-angle lens cameras to capture dynamic image information of people entering and leaving the power plant around the clock without blind spots. These cameras can maintain high-quality image output under various lighting conditions and complex environments to ensure that the collected images are clear and usable. The image processing module 102 performs a series of preprocessing operations on the received raw image information, such as denoising, brightness and contrast adjustment, image enhancement, etc., to optimize the image quality. Then, using advanced image segmentation technology, the face image is accurately separated from the complex background to prepare for subsequent face recognition. The recognition module 103 uses face recognition technology with deep learning algorithm as the core to accurately extract face feature points from the processed face image, such as the geometric position and feature values of the face contour, eyes, nose, mouth, etc. These features are then quickly and accurately matched with the facial feature templates of known employees or authorized visitors stored in a highly encrypted database to achieve identity authentication. For faces that cannot be matched, the system will mark them as unrecognized persons. The trajectory analysis module 104 uses intelligent video analysis technology, and the system can continuously track the activity trajectory of unidentified or unauthorized personnel in the thermal power plant. Through multi-camera linkage and spatiotemporal analysis, a detailed movement path report is formed. The alarm module 105 is used to trigger the alarm mechanism immediately once the movement path of the unidentified personnel touches the preset sensitive area, such as the control room, chemical warehouse or important equipment area. This includes but is not limited to sound and light alarms, sending instant alarm information to the security center, and even directly contacting security personnel to ensure a quick response. The user interface module 106 provides an intuitive and easy-to-use graphical interface for security managers, which can not only view the images of each monitoring point in real time, but also play back historical recordings, review the details of alarm events, and track personnel dynamics through interactive maps. In addition, the interface supports permission management to ensure that only authorized personnel can access sensitive information and operate the security system. In summary, this thermal power plant security device with artificial intelligence identity authentication realizes efficient identification of personnel identities, close monitoring of sensitive areas and timely response, making the monitoring effect better.

Second embodiment

Please refer to Figures 2 to 7. On the basis of the first embodiment, the present invention also provides a thermal power plant security device with artificial intelligence identity authentication. The image acquisition module 101 includes an image acquisition unit 201, an image transmission unit 202 and an image storage unit 203; the image acquisition unit 201 is used to collect image information of people entering and leaving the thermal power plant; the image transmission unit 202 is used to transmit the collected image information to the host computer; the image storage unit 203 is used to store the image information. The image acquisition unit 201 is composed of a series of high-resolution network cameras, which are distributed at all key entrances, exits and internal passages of the thermal power plant. These cameras use infrared night vision, wide dynamic range (WDR) and intelligent motion detection technology to ensure that the facial features, posture and behavioral details of people entering and leaving can be clearly captured whether in bright daytime or dim night. The camera is equipped with a weather-resistant shell, which can operate stably under extreme environmental conditions to ensure the continuity and integrity of image information. The image transmission unit 202 is responsible for transmitting the large amount of real-time data captured by the image acquisition unit 201 to the host computer or central processing server safely and losslessly through a high-speed and stable network connection. This link adopts the latest encryption transmission protocol, such as SSL/TLS, to prevent the data from being intercepted or tampered with during the transmission process, ensuring the security of the information. At the same time, the transmission unit has an intelligent flow control function, which can automatically adjust the data transmission rate according to the network conditions, avoid network congestion, and ensure the smoothness of image transmission. As a long-term storage for security data, the image storage unit 203 adopts a high-performance enterprise-level hard disk array, combined with advanced data compression and deduplication technology, to effectively manage and save massive image information. It not only provides sufficient storage capacity to archive historical data, but also supports rapid retrieval, which is convenient for subsequent tracing and analysis. In order to cope with hardware failures, the storage unit implements a RAID redundant backup strategy to ensure data reliability and disaster recovery capabilities.

The image processing module 102 includes a correction unit 204, an enhancement unit 205, a noise reduction unit 206 and a facial image extraction unit 207; the correction unit 204 is used to correct the image to eliminate lens distortion; the enhancement unit 205 is used to enhance the contrast and brightness of the image; the noise reduction unit 206 is used to remove noise in the image by using Gaussian filtering; the facial image extraction unit 207 is used to extract facial sub-images from the image after noise removal by using a Haar cascade classifier. The correction unit 204 is responsible for geometric correction processing of the original image to eliminate image distortion caused by camera lens characteristics, installation angle or shooting distance. By applying perspective transformation, lens correction algorithm and other technologies, the true shape of the image can be restored to ensure that the facial feature extraction in the subsequent processing steps is not affected by distortion, thereby improving the accuracy of recognition. The enhancement unit 205 is committed to improving the visual quality and recognizability of the image. By dynamically adjusting the contrast and brightness of the image, the facial features are made more distinct, and even images taken under low light or backlight conditions can be effectively improved. By using algorithms such as histogram equalization and adaptive brightness gain, the details of the face in the image are clearly layered, which helps to improve the robustness of the subsequent face recognition algorithm. The noise reduction unit 206 uses Gaussian filtering technology to perform denoising for random noise and interference in the image. Gaussian filtering is a smoothing algorithm widely used in image processing, which can effectively remove granular noise in the image, while retaining edge information as much as possible, maintaining the clarity of facial features, and providing purer image materials for subsequent facial detection and recognition. After the image is corrected, enhanced and denoised, the facial image extraction unit 207 uses the Haar cascade classifier, a classic face detection algorithm, to accurately locate and extract the face area from the processed image. The Haar cascade classifier can quickly and efficiently identify faces of different sizes and postures through a combination of a series of weak classifiers obtained through training, and can achieve stable face detection even in complex backgrounds. The extracted facial sub-image provides accurate input data for further feature extraction and identity authentication.

The facial image extraction unit 207 includes a loading subunit 208, a scanning subunit 209, a screening subunit 210, and a cropping subunit 211; the loading subunit 208 is used to load a pre-trained Haar cascade classifier model from the OpenCV library; the scanning subunit 209 is used to use the loaded Haar cascade classifier to scan the pre-processed image; the screening subunit 210 is used to set a threshold to filter out candidate areas that do not meet the conditions and retain the rectangular frame representing the face; the cropping subunit 211 is used to crop the facial sub-image from the original image according to the rectangular frame. The loading subunit 208 loads the pre-trained Haar cascade classifier model from the OpenCV library. Haar cascade is a face detection algorithm based on machine learning, which recognizes a series of simple features (such as edges, line segments, etc.) through training to efficiently detect faces in images. The loading subunit 208 ensures that the system can use these pre-trained models without having to train from scratch, which greatly improves the convenience of system deployment and operation efficiency.

After loading the Haar cascade classifier model, the scanning subunit 209 uses the model to scan the preprocessed image pixel by pixel. This process involves sliding window techniques of different scales, and the model attempts to match facial feature templates at different positions and sizes in the image to detect whether there is a face. The precision and speed of the scan are crucial to the system performance, ensuring that faces can be accurately detected even in complex backgrounds.

During the detection process, multiple candidate regions containing human faces are generated. The screening subunit 210 filters out candidate frames that are not representative of real human faces by setting a reasonable threshold. This step avoids the occurrence of false positives, reduces the burden of subsequent processing, and improves the accuracy of system recognition. The screening process involves a comprehensive evaluation of factors such as the size, shape, and confidence of the candidate frames to ensure that the retained rectangular frames do correspond to real human faces.

Based on the selected rectangular frame representing the face, the cropping sub-unit 211 accurately crops the facial sub-image from the original image. This step is the basis for subsequent feature extraction and identity verification, because it ensures that the input to the face recognition algorithm is pure and only contains the facial features, excluding the influence of background and other interference factors. The cropped facial sub-image will be further processed to extract facial feature vectors and match them with known face data in the database, and finally realize the identity authentication of the person.

The recognition module 103 includes a high-dimensional feature extraction unit 212, a similarity calculation unit 213 and a matching unit 214; the high-dimensional feature extraction unit 212 is used to apply the VGGFace algorithm to extract a high-dimensional feature vector on the cropped face image; the similarity calculation unit 213 is used to calculate the Euclidean distance between feature vectors; the matching unit 214 is used to set a Euclidean distance threshold. When the similarity between the newly extracted features and the features of a person in the database exceeds this threshold, it is determined that the match is successful, that is, the identity of the person is successfully identified; if the match fails, the person is marked as unknown.

The core task of the high-dimensional feature extraction unit 212 is to use the VGGFace algorithm, which is a deep learning-based facial recognition technology that is well-known in the field of computer vision for its excellent recognition performance. The VGGFace algorithm uses a pre-trained deep neural network model to perform in-depth analysis on the cropped face image and extract high-dimensional feature vectors. These feature vectors can highly summarize the individual differences of the face, including but not limited to facial structure, texture, and expression, and are the basis for achieving accurate recognition.

The similarity calculation unit 213 then uses the Euclidean distance as an indicator to measure the difference between the two feature vectors. The Euclidean distance is simple and intuitive, and can directly reflect the straight-line distance between two sets of data points in a multidimensional space. It is often used to measure the similarity of feature vectors. By calculating the Euclidean distance between the feature vector to be identified and the feature vector of each known person in the database, the system can quantitatively compare the difference between the two.

Based on the result of similarity calculation, the matching unit 214 sets a reasonable Euclidean distance threshold as the criterion for whether the match is successful or not. When the Euclidean distance between the newly extracted feature vector and the feature vector of a person in the database is lower than the preset threshold, it indicates that the two are highly similar, and the system determines that the match is successful, that is, the identity of the person is successfully identified, and the corresponding authority is allowed to access or record his/her entry and exit information. On the contrary, if all the compared Euclidean distances exceed the threshold, the system marks the person as unknown, and the alarm module 105 is triggered for further processing, such as notifying security personnel to intervene, or further tracking and monitoring of unknown persons, to ensure that there are no loopholes in the safety management of the thermal power plant.

The trajectory analysis module 104 includes a temporary ID allocation unit 215, a cross-shot matching unit 216 and a trajectory generation unit 217; the temporary ID allocation unit 215 is used to allocate a temporary ID to the person when the person is unknown; the cross-shot matching unit 216 is used to use the DeepSort algorithm to match the person across cameras and obtain the person's position based on the camera position; the trajectory generation unit 217 is used to construct the motion trajectory of each unidentified person using the allocated temporary ID and the person's position.

The temporary ID allocation unit 215 is responsible for immediately assigning a unique temporary identity (temporary ID) to an unknown person when the system detects him/her for the first time. This mechanism ensures that even when the personal identity information is unknown, the activity trajectory can be continuously and independently tracked, laying the foundation for subsequent cross-camera association and trajectory reconstruction. The temporary ID not only improves the efficiency of data processing, but also protects personal privacy and avoids confusion between unidentified individuals.

The cross-lens matching unit 216 uses the DeepSort algorithm to achieve continuous tracking of people assigned temporary IDs under different camera perspectives. The DeepSort algorithm combines the advantages of deep learning appearance feature extraction and sorting association algorithms, and can achieve high-precision person re-identification in complex scenes, even in conditions of dense crowds, light changes or frequent occlusions. By integrating the geographic location information of each camera, the unit can not only determine the real-time location of the person, but also predict his or her movement path, providing important data support for security management and situational awareness.

Based on the temporary ID and the personnel position information provided by the cross-shot matching unit 216, the trajectory generation unit 217 constructs a complete motion trajectory of each unidentified person through spatiotemporal correlation analysis. This process involves efficient integration and smoothing of a large number of spatiotemporal data points to visualize the movement patterns of the personnel, such as the travel route, the stop area, and the speed change.

The specific steps of using the DeepSort algorithm to match pedestrians across cameras include: encoding pedestrian images using a feature extraction network to generate a fixed-length vector representing the appearance of pedestrians; finding the most similar match by calculating the cosine similarity of pedestrian feature vectors between different frames; predicting the position of the pedestrian in the next frame to reduce the loss of matching caused by short-term occlusion. Specifically, the pedestrian images captured from each camera are processed using a pre-trained feature extraction network (usually a deep convolutional neural network, such as ResNet, MobileNet, etc.). These networks can extract features that are highly recognizable to the identity of pedestrians from the original pixel data. Subsequently, these high-level features are compressed into a fixed-length vector, a process called appearance feature encoding. This vector can fully represent the appearance information of pedestrians while keeping the dimension low, which is convenient for subsequent rapid comparison and storage. After obtaining the appearance feature vectors of pedestrians between different frames, the algorithm calculates the cosine similarity between any two feature vectors. Cosine similarity is a method to measure the degree of proximity between the directions of two non-zero vectors, and is suitable for comparing the similarity of vectors in high-dimensional space. By pairing the pedestrian feature vectors captured in adjacent frames and across cameras, the most similar pedestrian feature vectors in each detection box in the current frame and the previous frame or another camera view are found. This step ensures that the correct association of the target is maintained even when the pedestrian's posture changes or is slightly occluded in the picture. To further enhance the robustness of tracking, the DeepSort algorithm integrates a Kalman filter to predict the position where the pedestrian may appear in the next frame. Kalman filtering is a recursive Bayesian estimation method that can predict the state of a target (such as position and speed) based on previous observations and motion models. When a pedestrian is temporarily blocked or leaves the picture, the future position predicted by the filter can effectively reduce the matching interruption caused by occlusion and ensure the continuity of tracking. Once the pedestrian reappears, the algorithm can quickly find the best match based on the predicted position, avoiding tracking loss caused by temporary disappearance.

In addition to direct similarity matching and position prediction, DeepSort also uses the Hungarian algorithm to solve the best match between the predicted state and the newly detected state, thereby reducing the exchange error of the target ID.

The alarm module 105 includes a marking unit 218, a comparison unit 219 and an alarm unit 220. The marking unit 218 is used to clearly mark the positions of all sensitive areas in the system. The comparison unit 219 is used to continuously receive personnel position updates from the camera and perform real-time comparison with the defined sensitive area boundaries. The alarm unit 220 is used to determine the existence of violations. The alarm module 105 immediately executes a preset response action.

The marking unit 218 allows the administrator to accurately outline the location of all sensitive areas on the system map through an advanced graphical user interface (GUI) or programming interface. These areas may include high-security areas with restricted access, areas around important facilities, or areas that are prohibited from passing during specific time periods. During the marking process, the system supports a variety of custom options, such as area shape, color coding, and time limits to meet the security needs of different scenarios.

The comparison unit 219 is tightly integrated with the video surveillance system, and continuously receives and analyzes the real-time position data stream of personnel captured by the camera. Using advanced image recognition and position tracking technology, the comparison unit 219 can accurately identify the position changes of personnel in the picture, and immediately compare these dynamic information with the sensitive area boundaries pre-defined by the marking unit 218 in real time at high speed.

Once a person is detected to have illegally entered or approached a sensitive area, the alarm unit 220 immediately initiates a preset response action. These responses can be in various forms, including but not limited to: on-site sound and light alarms to deter violators and attract the attention of surrounding personnel; automatically sending alarm information to the security control center with a real-time location image of the intruder; and even activating linked physical protection measures, such as closing specific access control doors, starting surveillance video recording, etc. In addition, the alarm unit 220 can also implement response strategies in a hierarchical manner according to the pre-set urgency level to ensure that the response is both rapid and appropriate.

What is disclosed above is only a preferred embodiment of the present invention, and it certainly cannot be used to limit the scope of rights of the present invention. Ordinary technicians in this field can understand that all or part of the processes of the above embodiment and equivalent changes made according to the claims of the present invention still fall within the scope of the invention.

Claims (8)

1. A thermal power plant security device with artificial intelligence identity authentication, characterized in that: It includes image acquisition module, image processing module, recognition module, trajectory analysis module and alarm module; The image acquisition module is used to be arranged in the thermal power plant to collect image information of people entering and leaving; The image processing module is used to pre-process the image information to obtain facial image information; The recognition module is used to extract facial features from facial image information and match them with known features in the database to obtain recognition results; The trajectory analysis module is used to track and record the movement path of unidentified personnel in the thermal power plant; The alarm module is used to trigger an alarm when it is found that the moving path involves a sensitive area. 2. The artificial intelligence identity authentication thermal power plant security device according to claim 1, characterized in that: The artificial intelligence identity authentication thermal power plant security device also includes a user interface module, which is used for security management personnel to view real-time monitoring images. 3. The artificial intelligence identity authentication thermal power plant security device according to claim 2, characterized in that: The image acquisition module includes an image acquisition unit, an image transmission unit and an image storage unit; The image acquisition unit is used to be distributed in the thermal power plant to collect image information of people entering and leaving; The image transmission unit is used to transmit the collected image information to the host computer; The image storage unit is used to store image information. 4. The artificial intelligence identity authentication thermal power plant security device according to claim 3, characterized in that: The image processing module includes a correction unit, an enhancement unit, a noise reduction unit and a facial image extraction unit; The correction unit is used to correct the image to eliminate lens distortion; The enhancement unit is used to enhance the contrast and brightness of the image; The noise reduction unit is used to remove noise in the image by using Gaussian filtering; The facial image extraction unit is used to extract facial sub-images from the image after noise removal using a Haar cascade classifier. 5. The artificial intelligence identity authentication thermal power plant security device according to claim 4, characterized in that: The facial image extraction unit includes a loading subunit, a scanning subunit, a screening subunit, and a cutting subunit; The loading subunit is used to load the pre-trained Haar cascade classifier model from the OpenCV library; The scanning subunit is used to scan the preprocessed image using the loaded Haar cascade classifier; The screening subunit is used to set a threshold to filter out candidate areas that do not meet the conditions and retain the rectangular frame representing the face; The cropping subunit is used to crop a facial sub-image from the original image according to the rectangular frame. 6. The artificial intelligence identity authentication thermal power plant security device according to claim 5, characterized in that: The recognition module includes a high-dimensional feature extraction unit, a similarity calculation unit and a matching unit; The high-dimensional feature extraction unit is used to extract high-dimensional feature vectors on the cropped face image using the VGGFace algorithm; The similarity calculation unit is used to calculate the Euclidean distance between feature vectors; The matching unit is used to set a Euclidean distance threshold. When the similarity between the newly extracted feature and the feature of a person in the database exceeds this threshold, it is determined that the match is successful, that is, the identity of the person is successfully identified; if the match fails, the person is marked as unknown. 7. The artificial intelligence identity authentication thermal power plant security device according to claim 6, characterized in that: The trajectory analysis module includes a temporary ID allocation unit, a cross-shot matching unit and a trajectory generation unit; The temporary ID allocation unit is used to allocate a temporary ID to the person when the person is unknown; The cross-lens matching unit is used to match personnel across cameras using a DeepSort algorithm, and obtain the personnel position based on the camera position; The trajectory generating unit is used to construct the movement trajectory of each unidentified person by using the assigned temporary ID and the person's position. 8. The artificial intelligence identity authentication thermal power plant security device according to claim 7, characterized in that: The specific steps of using the DeepSort algorithm to match pedestrians across cameras include: The pedestrian image is encoded using a feature extraction network to generate a fixed-length vector representing the appearance of the pedestrian; Find the most similar match by calculating the cosine similarity of pedestrian feature vectors between different frames; Predict the position of pedestrians in the next frame to reduce matching loss caused by short-term occlusion.