Multi-dimensional image analysis method and system of Internet of Things video monitoring large scene

Abstract

The invention relates to a multi-dimensional image analysis method and system of an internet of things video monitoring large scene, and belongs to the technical field of image analysis. The method comprises the following steps: acquiring a historical monitoring picture of personnel protection equipment, and performing image preprocessing to obtain a historical monitoring standard picture of the personnel protection equipment; constructing a personnel protection equipment monitoring data set and obtaining a personnel protection equipment area bounding box set through a boundary detection algorithm; a comprehensive monitoring picture feature set is obtained through multi-dimensional feature extraction and fusion according to the personnel protection equipment area bounding box set, and a personnel protection equipment recognition model is obtained through classification decision algorithm training and testing; and monitoring a personnel protection equipment monitoring video in real time, obtaining a personnel protection equipment classification label through the personnel protection equipment identification model, and sending a protection equipment wearing early warning prompt to a terminal, thereby realizing multi-dimensional protection equipment image analysis, intelligent detection of the protection equipment wearing condition and early warning, and ensuring personnel safety.

Description

Technical Field

[0001] The present invention belongs to the technical field of image analysis, and in particular relates to a multi-dimensional image analysis method and system for large-scale Internet of Things video surveillance scenes. Background Art

[0002] With the acceleration of urbanization and the rapid development of information technology, the application of large-scene image analysis technology of IoT video surveillance in public security, intelligent transportation, environmental monitoring and other fields has become increasingly important. However, with the increasing complexity of monitoring scenes and the increasing diversification of monitoring needs, the existing large-scene image analysis technology of IoT video surveillance faces a series of challenges.

[0003] In high-risk workplaces such as construction sites and factories, the detection of hard hats and protective equipment is particularly critical, but the safety monitoring technology in these places faces the following major challenges: when monitoring cameras capture large-scale scenes, it is often difficult to obtain a sufficiently high image resolution, resulting in the inability to clearly identify small-sized or long-distance targets; the changing environment of the construction site, such as lighting changes, weather effects, dust and smoke, will interfere with the image quality and affect the detection effect; the background in the monitoring scene is complex and changeable, such as building materials, mechanical equipment, etc., these elements may be confused with the monitoring target, increasing the difficulty of detection; in a busy construction site, workers may block each other, causing part or all of the hard hats and protective equipment to be blocked, affecting the accuracy of recognition. Due to the limitations of image resolution and environmental interference, it is particularly difficult to capture high-quality image details in a wide monitoring area, which seriously affects further image analysis and processing; and with the increasing surge in the amount of monitoring data, the need for real-time analysis of monitoring data has become more urgent. Existing image processing technology is difficult to meet the rapidly growing data processing needs, and the improvement of data processing speed has become an urgent problem to be solved; at the same time, with the increasing complexity of monitoring scenes, single-dimensional image analysis can no longer meet the comprehensive evaluation of complex information.

[0004] Therefore, an efficient multi-dimensional image analysis method and system are urgently needed to cope with the challenges brought about by the increasing complexity of large-scale IoT video surveillance scenes. Summary of the invention

[0005] In order to solve the above problems existing in the prior art, the present invention proposes a multi-dimensional image analysis method and system for large-scale IoT video surveillance.

[0006] The purpose of the present invention can be achieved by the following technical solutions: A multi-dimensional image analysis method for large-scale IoT video surveillance scenes, comprising: Acquire historical monitoring pictures of personnel protective equipment, and obtain historical monitoring standard pictures of personnel protective equipment through image preprocessing according to the historical monitoring pictures of personnel protective equipment; Constructing a personnel protective equipment monitoring data set according to the historical monitoring standard image of the personnel protective equipment, and obtaining a personnel protective equipment area boundary frame set according to the personnel protective equipment monitoring data set by using a boundary detection algorithm; A comprehensive monitoring image feature set is obtained by multi-dimensional feature extraction and fusion according to the personnel protective equipment area boundary frame set, and a personnel protective equipment recognition model is obtained by classification decision algorithm training and testing according to the comprehensive monitoring image feature set; Monitor the personal protective equipment surveillance video in real time, obtain the personal protective equipment classification label through the personal protective equipment recognition model according to the personal protective equipment surveillance video, and send the protective equipment wearing warning prompt to the terminal according to the personal protective equipment classification label.

[0007] Preferably, the acquiring of the historical monitoring picture of the personnel protective equipment and obtaining the historical monitoring standard picture of the personnel protective equipment by image preprocessing according to the historical monitoring picture of the personnel protective equipment comprises: The image preprocessing includes image denoising and illumination correction; The image denoising removes noise in the historical monitoring image of the personnel protective equipment by median filtering to improve the image quality; The illumination correction adjusts the brightness and contrast of the historical monitoring pictures of the personnel protective equipment through an adaptive illumination correction technology to reduce the impact of illumination changes.

[0008] Preferably, the adaptive illumination correction technology includes extracting illumination components of historical monitoring images of the personnel protective equipment and dynamically adjusting the brightness using a two-dimensional gamma function.

[0009] Preferably, constructing a personnel protective equipment monitoring data set according to the historical monitoring standard image of the personnel protective equipment, and obtaining a personnel protective equipment area boundary frame set by a boundary detection algorithm according to the personnel protective equipment monitoring data set includes: According to the personnel protective equipment monitoring data set, a personnel protective equipment area boundary box detection model is obtained by training and testing a boundary detection algorithm; The personnel protection equipment area bounding box set is obtained through the personnel protection equipment area bounding box detection model. The mathematical expression of the personnel protection equipment area bounding box set is: , in, are the coordinates and size of the jth bounding box, c j is the sample category, and M is the total number of bounding boxes.

[0010] Preferably, the boundary detection algorithm includes introducing ECA attention in the YOLOv7 backbone network CSPDarknet to enhance feature expression. Specifically, the convolution kernel size of the starting convolution layer is 3x3, the output channel is 32, the step size is 1, and the padding is 1, followed by a CSP1x module, a CSP2x module, and a CSP3x module. The CSP1x module is a convolution layer with an output channel of 64, the CSP2x module includes 2 convolution layers with an output channel of 128, and the CSP3x module includes 8 convolution layers with an output channel of 256. ECA is embedded after each CSP module. The attention module calculates channel weights through global average pooling, 1D convolution, and Sigmoid activation to achieve feature importance weighting. The detection head adopts the PANet structure, which effectively integrates feature information of different scales through top-down path enhancement and bottom-up feature fusion. The output layer results are the predicted score of personnel protective equipment, the horizontal coordinate of the center point of the bounding box, the vertical coordinate of the center point of the bounding box, the width of the bounding box, and the height of the bounding box. During the training process, the cross entropy-smoothed L1 joint loss function is used and the cosine annealing strategy is used to optimize the learning rate. The model parameters are updated through back propagation through the Adam optimizer according to the total loss; The mathematical expression of the cross entropy-smoothed L1 joint loss function is: , Among them, C is the number of categories, y i is the true label of the category, p i is the predicted probability, s i is the predicted bounding box coordinate, z i is the real bounding box coordinate, and smoothL1 is the smooth L1 function.

[0011] Preferably, the comprehensive monitoring image feature set obtained by extracting and fusing the personnel protective equipment area boundary frame set through multi-dimensional features includes: The multi-dimensional feature extraction and fusion includes extracting color features through a color histogram to obtain image color features, extracting texture features through a gray-level co-occurrence matrix to obtain image texture features, extracting deep-level features of the image through a convolutional neural network, and fusing the image color features, the image texture features, and the image deep-level features to obtain a multi-dimensional comprehensive monitoring picture feature representation; Performing the multi-dimensional feature extraction and fusion processing on all the pictures in the personnel protective equipment monitoring data set to obtain a comprehensive monitoring picture feature set; The mathematical expression of the comprehensive monitoring image feature set is: , Among them, f color (b j) is the color feature vector of the jth bounding box area, f texture (b j ) is the texture feature vector of the jth bounding box area, f deep (b j ) is the deep feature vector of the jth bounding box area, c j is the sample category, and M is the total number of bounding boxes.

[0012] Preferably, the mathematical expression of the classification decision algorithm is: , Among them, α i is the Lagrange multiplier, b is the bias term, y i is the label of the support vector, N train is the number of training samples, F(b i ) is the feature vector of the comprehensive monitoring image of the ith training sample, F is the feature vector of the comprehensive monitoring image of the sample to be classified, K is the kernel function, which is used to calculate the inner product of two feature vectors in the high-dimensional space, indicating the similarity between the ith support vector and the sample to be classified, and the sign() function is used to judge the positive or negative of its internal calculation result and convert the output of the discriminant function into a classification label.

[0013] A multi-dimensional image analysis system for large-scale IoT video surveillance scenes, applied to the multi-dimensional image analysis method for large-scale IoT video surveillance scenes, including an image preprocessing module, a personnel protective equipment area detection module, a personnel protective equipment recognition model construction module, and a personnel protective equipment analysis and early warning module; The image preprocessing module is used to obtain historical monitoring pictures of personnel protection equipment, and obtain historical monitoring standard pictures of personnel protection equipment through image preprocessing according to the historical monitoring pictures of personnel protection equipment; The personnel protection equipment area detection module is used to construct a personnel protection equipment monitoring data set according to the personnel protection equipment historical monitoring standard picture, and obtain a personnel protection equipment area boundary frame set according to the personnel protection equipment monitoring data set through a boundary detection algorithm; The personnel protection equipment recognition model construction module is used to obtain a comprehensive monitoring image feature set through multi-dimensional feature extraction and fusion according to the personnel protection equipment area boundary frame set, and obtain a personnel protection equipment recognition model through classification decision algorithm training and testing according to the comprehensive monitoring image feature set; The personal protective equipment analysis and early warning module is used to monitor the personal protective equipment monitoring video in real time, obtain the personal protective equipment classification label through the personal protective equipment recognition model according to the personal protective equipment monitoring video, and send the protective equipment wearing early warning prompt to the terminal according to the personal protective equipment classification label.

[0014] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the multi-dimensional image analysis method for large scenes of IoT video surveillance is implemented.

[0015] A storage medium containing computer executable instructions, which are used to perform the above-mentioned multi-dimensional image analysis method for large-scale Internet of Things video surveillance when executed by a computer processor.

[0016] The beneficial effects of the present invention are: (1) The bounding box of the personal protective equipment area is detected through the boundary detection algorithm. The ECA attention mechanism is introduced and the loss function is combined to effectively improve the detection accuracy of the bounding box of the personal protective equipment area. The learning rate is optimized through the cosine annealing strategy, which accelerates the convergence speed of the model and reduces the training time. It can adapt to the detection needs in different scenarios and complex environments.

[0017] (2) A comprehensive monitoring image feature set is obtained through multi-dimensional feature extraction and fusion processing. By integrating color, texture and deep-level features, the obtained feature representation is more comprehensive and can better capture the diverse information of the image. Multi-dimensional feature fusion improves the expressiveness of the feature and helps to improve the accuracy and reliability of the personnel protective equipment recognition model. The rich feature information helps to train a more powerful recognition model and improve the generalization ability of the model.

[0018] (3) The personal protective equipment area is detected through the boundary detection algorithm, and the personal protective equipment recognition model is constructed by multi-dimensional feature extraction and fusion combined with the classification decision algorithm. It can timely detect abnormal wearing of personal protective equipment and issue early warnings, which helps to ensure personnel safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to facilitate understanding by those skilled in the art, the present invention is further described below with reference to the accompanying drawings.

[0020] Figure 1 The present invention is a flowchart of a multi-dimensional image analysis method for large-scale Internet of Things video surveillance. DETAILED DESCRIPTION

[0021] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation methods, structures, features and effects of the present invention are described in detail below in conjunction with the accompanying drawings and preferred embodiments.

[0022] See also Figure 1 , a multi-dimensional image analysis method for large scenes of IoT video surveillance, comprising: S1: Acquire a historical monitoring picture of personnel protective equipment, and obtain a historical monitoring standard picture of personnel protective equipment through image preprocessing according to the historical monitoring picture of personnel protective equipment; S2: constructing a personnel protective equipment monitoring data set according to the historical monitoring standard image of the personnel protective equipment, and obtaining a personnel protective equipment area boundary frame set according to the personnel protective equipment monitoring data set by using a boundary detection algorithm; S3: obtaining a comprehensive monitoring image feature set through multi-dimensional feature extraction and fusion according to the personnel protective equipment area boundary frame set, and obtaining a personnel protective equipment recognition model through classification decision algorithm training and testing according to the comprehensive monitoring image feature set; S4: monitor the personnel protective equipment monitoring video in real time, obtain the personnel protective equipment classification label through the personnel protective equipment recognition model according to the personnel protective equipment monitoring video, and send the protective equipment wearing warning prompt to the terminal according to the personnel protective equipment classification label.

[0023] In this embodiment, the acquisition of the historical monitoring picture of the personnel protective equipment and the acquisition of the historical monitoring standard picture of the personnel protective equipment by image preprocessing are specifically implemented by the following steps: The image preprocessing includes image denoising and illumination correction; The image denoising removes noise in the historical monitoring image of the personnel protective equipment by median filtering to improve the image quality; The illumination correction adjusts the brightness and contrast of the historical monitoring pictures of the personnel protective equipment through an adaptive illumination correction technology to reduce the impact of illumination changes.

[0024] It should be noted that the adaptive illumination correction technology includes extracting the illumination component of the historical monitoring picture of the personnel protective equipment and dynamically adjusting the brightness using a two-dimensional gamma function.

[0025] In this embodiment, the construction of a personnel protective equipment monitoring data set according to the historical monitoring standard image of the personnel protective equipment, and the acquisition of a personnel protective equipment area boundary frame set by a boundary detection algorithm according to the personnel protective equipment monitoring data set are specifically implemented by the following steps: S201: Obtaining a personnel protective equipment area boundary box detection model through boundary detection algorithm training and testing according to the personnel protective equipment monitoring data set; S202: The boundary detection algorithm includes introducing ECA attention into the YOLOv7 backbone network CSPDarknet to enhance feature expression. In terms of specific structure, the convolution kernel size of the starting convolution layer is 3x3, the output channel is 32, the step size is 1, and the padding is 1, followed by a CSP1x module, a CSP2x module, and a CSP3x module. The CSP1x module is a convolution layer with an output channel of 64, the CSP2x module includes 2 convolution layers with an output channel of 128, and the CSP3x module includes 8 convolution layers with an output channel of 256. ECA is embedded after each CSP module. The attention module calculates channel weights through global average pooling, 1D convolution and Sigmoid activation to achieve feature importance weighting. The detection head adopts the PANet structure, which effectively integrates feature information of different scales through top-down path enhancement and bottom-up feature fusion. The results of the output layer are the predicted score of personnel protective equipment, the horizontal coordinate of the center point of the bounding box, the vertical coordinate of the center point of the bounding box, the width of the bounding box, and the height of the bounding box. During the training process, the cross entropy-smoothed L1 joint loss function is used and the cosine annealing strategy is used to optimize the learning rate. The model parameters are updated through back propagation through the Adam optimizer according to the total loss.

[0026] The ECA attention is a lightweight channel attention mechanism that can enhance the model's emphasis on key feature channels. The mathematical expression is: , Among them, X is the input feature map, GlobalAvgPool is the global average pooling operation, W is the weight learned by the ECA module, σ is the Sigmoid activation function, and ⊙ is the element-by-element multiplication symbol.

[0027] The mathematical expression of the cross entropy-smoothed L1 joint loss function is: , Among them, C is the number of categories, y i is the true label of the category, p i is the predicted probability, s i is the predicted bounding box coordinate, z i is the real bounding box coordinate, smoothL1 is the smooth L1 function; The mathematical expression of the smooth L1 function is: . S203: Acquire a personnel protection equipment area bounding box set through the personnel protection equipment area bounding box detection model; The mathematical expression of the bounding box set of the personnel protective equipment area is: , in, are the coordinates and size of the jth bounding box, c j is the sample category, and M is the total number of bounding boxes.

[0028] In this embodiment, the comprehensive monitoring picture feature set is obtained by multi-dimensional feature extraction and fusion according to the personal protective equipment area boundary frame set, and the personal protective equipment recognition model is obtained by classification decision algorithm training and testing according to the comprehensive monitoring picture feature set. Specifically, the following steps are implemented: The multi-dimensional feature extraction and fusion includes extracting color features through a color histogram to obtain image color features, extracting texture features through a gray-level co-occurrence matrix to obtain image texture features, extracting deep-level features of the image through a convolutional neural network, and fusing the image color features, the image texture features, and the image deep-level features to obtain a multi-dimensional comprehensive monitoring picture feature representation; Performing the multi-dimensional feature extraction and fusion processing on all the pictures in the personnel protective equipment monitoring data set to obtain a comprehensive monitoring picture feature set; The mathematical expression of the comprehensive monitoring image feature set is: , Among them, f color (b j ) is the color feature vector of the jth bounding box area, f texture (b j ) is the texture feature vector of the jth bounding box area, f deep (b j ) is the deep feature vector of the jth bounding box area, c j is the sample category, and M is the total number of bounding boxes.

[0029] The mathematical expression of the classification decision algorithm is: , Among them, α i is the Lagrange multiplier, b is the bias term, y i is the label of the support vector, N train is the number of training samples, F(b i ) is the feature vector of the comprehensive monitoring image of the ith training sample, F is the feature vector of the comprehensive monitoring image of the sample to be classified, K is the kernel function, which is used to calculate the inner product of two feature vectors in the high-dimensional space, indicating the similarity between the ith support vector and the sample to be classified, and the sign() function is used to judge the positive or negative of its internal calculation result and convert the output of the discriminant function into a classification label.

[0030] Specifically, the category of the sample to be classified is finally determined by calculating the similarity between all support vectors and the sample to be classified in the high-dimensional space and taking weighted sum.

[0031] In this embodiment, the real-time monitoring of the personnel protective equipment monitoring video, obtaining the personnel protective equipment classification label through the personnel protective equipment recognition model according to the personnel protective equipment monitoring video, and sending the protective equipment wearing warning prompt to the terminal according to the personnel protective equipment classification label are specifically implemented by the following steps: S401: Install cameras at construction sites or factories to capture surveillance video of personnel wearing protective equipment; S402: accessing and processing the personnel protective equipment monitoring video in real time through the real-time data stream processing technology Apache Kafka, and intercepting the personnel protective equipment monitoring video frames at a preset frequency; S403: Obtaining a personal protective equipment classification label through the personal protective equipment recognition model according to the personal protective equipment monitoring video frame; The personal protective equipment classification label includes a normal label for wearing personal protective equipment and an abnormal label for wearing personal protective equipment; S404: When the personal protective equipment classification label is a personal protective equipment wearing abnormal label, a protective equipment wearing warning prompt is sent to the terminal.

[0032] A multi-dimensional image analysis system for large-scale IoT video surveillance includes an image preprocessing module, a personnel protection equipment area detection module, a personnel protection equipment recognition model construction module, and a personnel protection equipment analysis and early warning module; The image preprocessing module is used to obtain historical monitoring pictures of personnel protection equipment, and obtain historical monitoring standard pictures of personnel protection equipment through image preprocessing according to the historical monitoring pictures of personnel protection equipment; The personnel protection equipment area detection module is used to construct a personnel protection equipment monitoring data set according to the personnel protection equipment historical monitoring standard picture, and obtain a personnel protection equipment area boundary frame set according to the personnel protection equipment monitoring data set through a boundary detection algorithm; The personnel protection equipment recognition model construction module is used to obtain a comprehensive monitoring image feature set through multi-dimensional feature extraction and fusion according to the personnel protection equipment area boundary frame set, and obtain a personnel protection equipment recognition model through classification decision algorithm training and testing according to the comprehensive monitoring image feature set; The personal protective equipment analysis and early warning module is used to monitor the personal protective equipment monitoring video in real time, obtain the personal protective equipment classification label through the personal protective equipment recognition model according to the personal protective equipment monitoring video, and send the protective equipment wearing early warning prompt to the terminal according to the personal protective equipment classification label.

[0033] The computer storage medium of the embodiment of the present invention may adopt any combination of one or more computer-readable media. The computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples (non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this document, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in combination with an instruction execution system, device or device.

[0034] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, which carry computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0035] The program code included on the computer readable medium can be transmitted with any appropriate medium, including but not limited to wireless, electric wire, optical cable, RF, etc., or any suitable combination of the above. The computer program code for performing the operation of the present invention can be written in one or more programming languages ​​or their combinations, and the programming language includes object-oriented programming languages-such as Java, Smalltalk, C++, and also includes conventional procedural programming languages-such as "C" language or similar programming languages. The program code can be executed completely on the user's computer, partially on the user's computer, as an independent software package, partially on the user's computer and partially on the remote computer, or completely on the remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (for example, using an Internet service provider to connect through the Internet).

[0036] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment as above, it is not used to limit the present invention. Any technical personnel in this field can make some changes or modify the technical contents disclosed above into equivalent embodiments without departing from the scope of the technical solution of the present invention. However, any simple modification, equivalent change and modification made to the above embodiments according to the technical essence of the present invention without departing from the content of the technical solution of the present invention still fall within the scope of the technical solution of the present invention.

Claims

1. A multi-dimensional image analysis method for large scenes of IoT video surveillance, characterized in that: include: Acquire historical monitoring pictures of personnel protective equipment, and obtain historical monitoring standard pictures of personnel protective equipment through image preprocessing according to the historical monitoring pictures of personnel protective equipment; Constructing a personnel protective equipment monitoring data set according to the historical monitoring standard image of the personnel protective equipment, and obtaining a personnel protective equipment area boundary frame set according to the personnel protective equipment monitoring data set by using a boundary detection algorithm; A comprehensive monitoring image feature set is obtained by multi-dimensional feature extraction and fusion according to the personnel protective equipment area boundary frame set, and a personnel protective equipment recognition model is obtained by classification decision algorithm training and testing according to the comprehensive monitoring image feature set; Monitor the personal protective equipment surveillance video in real time, obtain the personal protective equipment classification label through the personal protective equipment recognition model according to the personal protective equipment surveillance video, and send the protective equipment wearing warning prompt to the terminal according to the personal protective equipment classification label.

2. The multi-dimensional image analysis method for large scenes of IoT video surveillance according to claim 1 is characterized in that: The obtaining of the historical monitoring picture of the personnel protective equipment and obtaining the historical monitoring standard picture of the personnel protective equipment by image preprocessing according to the historical monitoring picture of the personnel protective equipment include: The image preprocessing includes image denoising and illumination correction; The image denoising removes noise in the historical monitoring image of the personnel protective equipment by median filtering to improve the image quality; The illumination correction adjusts the brightness and contrast of the historical monitoring pictures of the personnel protective equipment through an adaptive illumination correction technology to reduce the impact of illumination changes.

3. The multi-dimensional image analysis method for large scenes of IoT video surveillance according to claim 2 is characterized in that: The adaptive illumination correction technique includes extracting illumination components of historical monitoring images of the personnel protective equipment and dynamically adjusting brightness using a two-dimensional gamma function.

4. The multi-dimensional image analysis method for large scenes of IoT video surveillance according to claim 1 is characterized in that: The step of constructing a personnel protective equipment monitoring data set according to the historical monitoring standard image of the personnel protective equipment, and obtaining a personnel protective equipment area boundary frame set according to the personnel protective equipment monitoring data set by using a boundary detection algorithm includes: According to the personnel protective equipment monitoring data set, a personnel protective equipment area boundary box detection model is obtained by training and testing a boundary detection algorithm; The personnel protection equipment area bounding box set is obtained through the personnel protection equipment area bounding box detection model. The mathematical expression of the personnel protection equipment area bounding box set is: ， in, are the coordinates and size of the jth bounding box, c j is the sample category, and M is the total number of bounding boxes.

5. The multi-dimensional image analysis method for large scenes of IoT video surveillance according to claim 4 is characterized in that: The boundary detection algorithm includes introducing ECA attention in the YOLOv7 backbone network CSPDarknet to enhance feature expression. In terms of specific structure, the convolution kernel size of the starting convolution layer is 3x3, the output channel is 32, the step size is 1, and the padding is 1, followed by CSP1x module, CSP2x module and CSP3x module. The CSP1x module is a convolution layer with an output channel of 64, the CSP2x module includes 2 convolution layers with an output channel of 128, and the CSP3x module includes 8 convolution layers with an output channel of 256. ECA attention is embedded after each CSP module. The force module calculates channel weights through global average pooling, 1D convolution and Sigmoid activation to achieve feature importance weighting. The detection head adopts the PANet structure, which effectively integrates feature information of different scales through top-down path enhancement and bottom-up feature fusion. The output layer results are the predicted score of personnel protective equipment, the horizontal coordinate of the center point of the bounding box, the vertical coordinate of the center point of the bounding box, the width of the bounding box, and the height of the bounding box. During the training process, the cross entropy-smoothed L1 joint loss function is used and the cosine annealing strategy is used to optimize the learning rate. The model parameters are updated through back propagation through the Adam optimizer according to the total loss; The mathematical expression of the cross entropy-smoothed L1 joint loss function is: ， Among them, C is the number of categories, y i is the true label of the category, p i is the predicted probability, s i is the predicted bounding box coordinate, z i is the real bounding box coordinate, and smoothL1 is the smooth L1 function.

6. The multi-dimensional image analysis method for large scenes of IoT video surveillance according to claim 1 is characterized in that: The comprehensive monitoring image feature set obtained by extracting and fusing the multi-dimensional feature set according to the personnel protective equipment area boundary frame set includes: The multi-dimensional feature extraction and fusion includes extracting color features through a color histogram to obtain image color features, extracting texture features through a gray-level co-occurrence matrix to obtain image texture features, extracting deep-level features of the image through a convolutional neural network, and fusing the image color features, the image texture features, and the image deep-level features to obtain a multi-dimensional comprehensive monitoring picture feature representation; Performing the multi-dimensional feature extraction and fusion processing on all the pictures in the personnel protective equipment monitoring data set to obtain a comprehensive monitoring picture feature set; The mathematical expression of the comprehensive monitoring image feature set is: ， Among them, f color (b j ) is the color feature vector of the jth bounding box area, f texture (b j ) is the texture feature vector of the jth bounding box area, f deep (b j ) is the deep feature vector of the jth bounding box area, c j is the sample category, and M is the total number of bounding boxes.

7. The multi-dimensional image analysis method for large scenes of IoT video surveillance according to claim 1 is characterized in that: The mathematical expression of the classification decision algorithm is: ， Among them, α i is the Lagrange multiplier, b is the bias term, y i is the label of the support vector, N train is the number of training samples, F(b i ) is the feature vector of the comprehensive monitoring image of the ith training sample, F is the feature vector of the comprehensive monitoring image of the sample to be classified, K is the kernel function, which is used to calculate the inner product of two feature vectors in the high-dimensional space, indicating the similarity between the ith support vector and the sample to be classified, and the sign() function is used to judge the positive or negative of its internal calculation result and convert the output of the discriminant function into a classification label.

8. A multi-dimensional image analysis system for large-scale video surveillance scenes of the Internet of Things, applied to the multi-dimensional image analysis method for large-scale video surveillance scenes of the Internet of Things as described in any one of claims 1-7, characterized in that: It includes image preprocessing module, personnel protective equipment area detection module, personnel protective equipment recognition model building module, and personnel protective equipment analysis and early warning module; The image preprocessing module is used to obtain historical monitoring pictures of personnel protection equipment, and obtain historical monitoring standard pictures of personnel protection equipment through image preprocessing according to the historical monitoring pictures of personnel protection equipment; The personnel protection equipment area detection module is used to construct a personnel protection equipment monitoring data set according to the personnel protection equipment historical monitoring standard picture, and obtain a personnel protection equipment area boundary frame set according to the personnel protection equipment monitoring data set through a boundary detection algorithm; The personnel protection equipment recognition model construction module is used to obtain a comprehensive monitoring image feature set through multi-dimensional feature extraction and fusion according to the personnel protection equipment area boundary frame set, and obtain a personnel protection equipment recognition model through classification decision algorithm training and testing according to the comprehensive monitoring image feature set; The personal protective equipment analysis and early warning module is used to monitor the personal protective equipment monitoring video in real time, obtain the personal protective equipment classification label through the personal protective equipment recognition model according to the personal protective equipment monitoring video, and send the protective equipment wearing early warning prompt to the terminal according to the personal protective equipment classification label.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the multi-dimensional image analysis method for large scenes of Internet of Things video surveillance is implemented as described in any one of claims 1-7.

10. A storage medium containing computer executable instructions, characterized in that: The computer executable instructions, when executed by a computer processor, are used to execute the multi-dimensional image analysis method for large scenes of Internet of Things video surveillance as described in any one of claims 1-7.

Images