An anti-mesh abnormal detection system and method

Abstract

The present invention relates to the technical field of abnormal detection of protective nets, and specifically to an abnormal detection system and method for protective nets, including a three-dimensional display unit, a first warning unit, a second warning unit, a real-time alarm unit, an image acquisition unit, and a defect classification unit. It further includes: a model construction unit, which is used to perform three-dimensional modeling on the target protective net through three-dimensional modeling technology to obtain a three-dimensional model of the target protective net, and transmit the three-dimensional model of the target protective net to the area matching unit and the three-dimensional display unit. The present invention obtains the state data of multiple positions of the target protective net in real time through the state monitoring unit and the warning unit, and issues a warning according to a preset state threshold table. This real-time nature can help detect possible abnormal situations of the protective net in a timely manner.

Description

Technical Field

[0001] The present invention relates to the technical field of abnormal detection of protective nets, and particularly relates to an abnormal detection system and method for protective nets. Background Art

[0002] A protective net generally refers to a mesh structure or barrier for safety purposes, and protective nets are commonly used in industrial facilities, construction sites, stadiums and other places to prevent people or objects from accidentally falling or entering dangerous areas and protect the safety of workers and the public.

[0003] Traditional systems usually rely on manual inspections and monitoring. Operators need to regularly check the status of the protective net, and the monitoring of traditional systems often depends on the subjective judgment and experience of the operators, which may lead to misjudgment or missed detection of subtle abnormalities or hidden defects. Moreover, traditional systems usually provide status updates in the form of regular inspections or reports. Therefore, after a problem occurs, it may take a long time to be discovered and reported. Summary of the Invention

[0004] The object of the present invention is to propose an abnormal detection system and method for protective nets in view of the problems existing in the background art.

[0005] The technical solution of the present invention: An abnormal detection system for a protective net, including a three-dimensional display unit, a first warning unit, a second warning unit, a real-time alarm unit, an image acquisition unit and a defect classification unit, further including:

[0006] A model construction unit, which is used to perform three-dimensional modeling on the target protective net through three-dimensional modeling technology to obtain a three-dimensional model of the target protective net, and transmit the three-dimensional model of the target protective net to the area matching unit and the three-dimensional display unit;

[0007] An area matching unit, which receives the three-dimensional model of the target protective net transmitted by the model construction unit, obtains the geographical map of the location where the target protective net is located, matches the geographical map with the three-dimensional model of the target protective net to obtain an area mapping table, and transmits the area mapping table to the three-dimensional display unit;

[0008] A status monitoring unit, which is used to set fixed monitoring points at multiple positions of the target protective net, set a sensor group at the fixed monitoring points, monitor the status of multiple positions of the target protective net through the sensor group to obtain real-time status data of multiple positions of the target protective net, the real-time status data including temperature data, humidity data and vibration data, and transmit the real-time status data of multiple positions of the target protective net to the first warning unit and the second warning unit.

[0009] Preferably, the first warning unit receives the real-time status data of multiple positions of the target protection net transmitted by the status monitoring unit, matches the real-time status data of multiple positions of the target protection net with the status threshold table pre-stored in the database to obtain the real-time warning information of multiple positions of the target protection net, and transmits the real-time warning information of multiple positions of the target protection net to the three-dimensional display unit and the real-time alarm unit.

[0010] Preferably, the second warning unit receives the real-time status data of multiple positions of the target protection net transmitted by the status monitoring unit, predicts the real-time risks of multiple positions of the target protection net through a trained risk prediction model to obtain multiple risk positions of the target protection net, and transmits the multiple risk positions of the target protection net to the three-dimensional display unit, the real-time alarm unit and the image acquisition unit.

[0011] Preferably, the image acquisition unit receives the multiple risk positions of the target protection net transmitted by the second warning unit, acquires images of the multiple risk positions of the target protection net to obtain a target protection net image, and transmits the target protection net image to the defect classification unit.

[0012] Preferably, the defect classification unit receives the target protection net image transmitted by the image acquisition unit, classifies the defects of the target protection net image through a trained defect classification model to obtain the defect type corresponding to the target protection net image, thereby obtaining the defect type corresponding to the multiple risk positions of the target protection net, and transmits the defect type corresponding to the multiple risk positions of the target protection net to the three-dimensional display unit and the real-time alarm unit.

[0013] Preferably, the three-dimensional display unit receives the three-dimensional model of the target protection net transmitted by the model construction unit, the area mapping table transmitted by the area matching unit, the real-time warning information of multiple positions of the target protection net transmitted by the first warning unit, the multiple risk positions of the target protection net transmitted by the second warning unit, and the defect type corresponding to the multiple risk positions of the target protection net transmitted by the defect classification unit, maps the real-time warning information of multiple positions of the target protection net to the corresponding area in the three-dimensional model of the target protection net according to the area mapping table, displays the real-time warning information in the corresponding area in the three-dimensional model of the target protection net, matches the multiple risk positions of the target protection net with the area mapping table to obtain the corresponding area of the multiple risk positions of the target protection net in the three-dimensional model of the target protection net, and highlights the display at the corresponding area, and displays the defect type corresponding to the risk position at the corresponding area.

[0014] Preferably, the real-time alarm unit receives the real-time early warning information of multiple positions of the target protection network transmitted by the first early warning unit, the multiple risk positions of the target protection network transmitted by the second early warning unit, and the defect types corresponding to the multiple risk positions of the target protection network transmitted by the defect classification unit, and performs abnormal alarms on the real-time early warning information of multiple positions of the target protection network, the multiple risk positions of the target protection network, and the corresponding defect types through an alarm device.

[0015] Preferably, the risk prediction model includes an encoding module, a state evolution perception module, and a risk prediction module. The encoding module is used to encode the position information, temperature data, humidity data, and vibration data of the target protection network to obtain a low-dimensional feature set, and perform embedding fusion on the low-dimensional feature set to obtain a low-dimensional fusion feature. The state evolution perception module uses a global recurrent memory network, and the global recurrent memory network is used to process the low-dimensional fusion features of the position information, temperature data, humidity data, and vibration data of the target protection network to obtain a monitoring state representation of the target protection network. The risk prediction module is used to jointly calculate the weight of the current node based on the hidden layer states of all nodes before the current time step, and fuse the monitoring state representation of the target protection network based on the weight of the current node, and classify the fused monitoring state representation through a Sigmoid classifier to obtain multiple risk positions of the target protection network.

[0016] Preferably, the expression of the low-dimensional feature set is as follows:

[0017]

[0018] Wherein, and respectively represent the low-dimensional features of position information, temperature data, humidity data, and vibration data, and M loc , M tem , M hum and M vib respectively represent the embedding matrices of position information, temperature data, humidity data, and vibration data to be learned, and respectively represent the one-hot encodings of position information, temperature data, humidity data, and vibration data;

[0019] The expression of the low-dimensional fusion feature is as follows:

[0020]

[0021] Wherein, x i represents the low-dimensional fusion feature, ReLU represents the activation function, and W iDenote the potential impact parameter for measuring dependence and temporal variation effects, b i Denote the bias vector;

[0022] The expression of the monitored state representation is as follows:

[0023]

[0024] Wherein, Denote the hidden state at the i-th time step, and Denote the cell states at the i-th and i-1-th time steps respectively, LSTM i Denote the global recurrent memory network.

[0025] Preferably, the defect classification model includes a feature extraction module and a feature metric module. The feature extraction module includes 4 residual blocks. The residual block includes a convolutional layer, a BN layer, and an activation function layer. After the target security net image passes through 4 residual blocks, a tensor feature map is obtained, and the tensor feature map is flattened to obtain a set of local descriptors. The feature metric module calculates the similarity between the set of local descriptors and the feature pooling pool, and compares the similarity between the two with a preset similarity threshold. If it is greater than the preset similarity threshold, the target security net image is classified as the type of the template image in the corresponding feature pooling pool.

[0026] Preferably, the expression of the local descriptor is as follows:

[0027] Ψ(Y)=(y 1 ,...,y i ,...,y m );

[0028] Wherein, Ψ(Y) denotes the set of local descriptors, and y i Denotes the i-th local descriptor of the tensor feature map.

[0029] The technical solution of the present invention: A method for warning of abnormalities in a security net, which is applicable to the described security net abnormality detection system, includes the following steps:

[0030] S1. Use 3D modeling technology to perform 3D modeling on the target security net to obtain a 3D model of the target security net, obtain a geographical map of the location where the target security net is located, perform matching of the geographical map and the 3D model, and generate a regional mapping table;

[0031] S2. Set fixed monitoring points at multiple positions of the target security net, set a sensor group at each monitoring point, monitor data such as temperature, humidity, and vibration, obtain real-time status data of multiple positions of the target security net, and match the real-time status data with a preset status threshold table to generate real-time warning information;

[0032] S3. Use the trained risk prediction model to predict the risk positions of the target protection net, collect images of the risk positions to obtain the image data of the target protection net, and use the trained defect classification model to classify the defects in the images to obtain the defect types corresponding to each risk position.

[0033] S4. Map the real-time warning information to the corresponding three-dimensional model area for display, highlight the risk positions, and issue an abnormal alarm through the alarm device.

[0034] Compared with the prior art, the above technical solution of the present invention has the following beneficial technical effects:

[0035] 1. The present invention obtains the status data of multiple positions of the target protection net in real time through the status monitoring unit and the warning unit, and issues a warning according to the preset status threshold table. This real-time nature can help detect possible abnormal situations of the protection net in time, such as abnormal temperature, humidity or vibration, as well as possible risk positions. And the second warning unit uses the trained risk prediction model to predict the positions that may have risks and transmits these positions to the three-dimensional display unit and the real-time alarm unit. The defect classification unit then collects images and classifies defects of the risk positions to further confirm the risk types.

[0036] 2. The present invention uses the three-dimensional model of the target protection net provided by the model construction unit and the area mapping table provided by the area matching unit through the three-dimensional display unit to accurately map the real-time warning information and risk positions to the corresponding three-dimensional model area and highlight them. This display method helps the operator intuitively understand the abnormal situations and defect types of each part of the protection net. The real-time alarm unit issues an abnormal alarm through the alarm device according to the received real-time warning information, risk positions and defect types. This mechanism ensures that the operator can respond quickly and take necessary measures to prevent the problems that may be caused by defects or risk positions of the protection net from deteriorating further. Description of the Drawings

[0037] Figure 1 It is a schematic flow chart of the overall system in an embodiment proposed by the present invention;

[0038] Figure 2 It is a schematic flow chart of the overall method in an embodiment proposed by the present invention.

[0039] Reference Numerals: 1. Model Construction Unit; 2. Area Matching Unit; 3. Three-Dimensional Display Unit; 4. Status Monitoring Unit; 5. First Warning Unit; 6. Second Warning Unit; 7. Real-Time Alarm Unit; 8. Image Acquisition Unit; 9. Defect Classification Unit. Detailed Embodiments

[0040] Example 1, as Figure 1 shown, an abnormal detection system for a protective net proposed by the present invention includes a three-dimensional display unit 3, a first warning unit 5, a second warning unit 6, a real-time alarm unit 7, an image acquisition unit 8, and a defect classification unit 9, and further includes:

[0041] A model construction unit 1, which is used to perform three-dimensional modeling on the target protective net through three-dimensional modeling technology to obtain a three-dimensional model of the target protective net, and transmit the three-dimensional model of the target protective net to the area matching unit 2 and the three-dimensional display unit 3;

[0042] An area matching unit 2, which receives the three-dimensional model of the target protective net transmitted by the model construction unit 1, obtains the geographical map of the location where the target protective net is located, matches the geographical map with the three-dimensional model of the target protective net to obtain an area mapping table, and transmits the area mapping table to the three-dimensional display unit 3;

[0043] A status monitoring unit 4, which is used to set fixed monitoring points at multiple positions of the target protective net, set a sensor group at the fixed monitoring points, monitor the status of multiple positions of the target protective net through the sensor group to obtain real-time status data of multiple positions of the target protective net, the real-time status data includes temperature data, humidity data, and vibration data, and transmits the real-time status data of multiple positions of the target protective net to the first warning unit 5 and the second warning unit 6.

[0044] In the present invention, three-dimensional modeling technology is the process of using computer software or tools to create three-dimensional models of objects or scenes. It is usually used in fields such as engineering, design, animation production, and scientific research. Three-dimensional modeling technology is used to generate a three-dimensional model of the target protective net for subsequent analysis, design, or display; a geographical map is a map that describes the geographical features, terrain, and location of the earth's surface, and is used to display and analyze geographical information of a specific area on the earth, such as roads, buildings, terrain, etc., so as to determine the specific geographical coordinates and environmental characteristics of the location where the target protective net is located; an area mapping table is a table or data structure that maps the three-dimensional data in the model to the actual geographical space. It contains the matching relationship between the three-dimensional model of the target protective net generated by the model construction unit and the geographical map information. This mapping is to ensure that the three-dimensional model can accurately reflect the position and layout in the actual geographical environment; a sensor group refers to a set of sensor devices installed at multiple positions of the target protective net. These sensors are used to monitor the status information of the target protective net, such as temperature, humidity, and vibration.

[0045] In an optional embodiment, the first warning unit 5 receives the real-time status data of multiple positions of the target protection net transmitted by the status monitoring unit 4, and matches the real-time status data of multiple positions of the target protection net with the status threshold table pre-stored in the database to obtain the real-time warning information of multiple positions of the target protection net, and transmits the real-time warning information of multiple positions of the target protection net to the three-dimensional display unit 3 and the real-time alarm unit 7.

[0046] It should be noted that the status threshold table is a pre-defined database that contains the preset thresholds of different status parameters for each position of the target protection net. These status parameters can include temperature, humidity, vibration intensity, etc., depending on the specific characteristics and environmental conditions that need to be monitored for the target protection net.

[0047] In an optional embodiment, the second warning unit 6 receives the real-time status data of multiple positions of the target protection net transmitted by the status monitoring unit 4, and predicts the real-time risks of multiple positions of the target protection net through a trained risk prediction model to obtain multiple risk positions of the target protection net, and transmits the multiple risk positions of the target protection net to the three-dimensional display unit 3, the real-time alarm unit 7 and the image acquisition unit 8.

[0048] In an optional embodiment, the image acquisition unit 8 receives the multiple risk positions of the target protection net transmitted by the second warning unit 6, and performs image acquisition on the multiple risk positions of the target protection net to obtain a target protection net image, and transmits the target protection net image to the defect classification unit 9.

[0049] In an optional embodiment, the defect classification unit 9 receives the target protection net image transmitted by the image acquisition unit 8, and classifies the defects of the target protection net image through a trained defect classification model to obtain the defect types corresponding to the target protection net image, so as to obtain the defect types corresponding to the multiple risk positions of the target protection net, and transmits the defect types corresponding to the multiple risk positions of the target protection net to the three-dimensional display unit 3 and the real-time alarm unit 7.

[0050] In an alternative embodiment, the 3D display unit 3 receives the 3D model of the target protective net transmitted by the model construction unit 1, the area mapping table transmitted by the area matching unit 2, the real-time warning information of multiple positions of the target protective net transmitted by the first warning unit 5, the multiple risk positions of the target protective net transmitted by the second warning unit 6, and the defect types corresponding to the multiple risk positions of the target protective net transmitted by the defect classification unit 9. The real-time warning information of multiple positions of the target protective net is mapped to the corresponding area in the 3D model of the target protective net according to the area mapping table, and the real-time warning information is displayed in the corresponding area in the 3D model of the target protective net. The multiple risk positions of the target protective net are matched with the area mapping table to obtain the corresponding areas of the multiple risk positions of the target protective net in the 3D model of the target protective net, and are highlighted in the corresponding areas, and the defect types corresponding to the risk positions are displayed in the corresponding areas.

[0051] In an alternative embodiment, the real-time alarm unit 7 receives the real-time warning information of multiple positions of the target protective net transmitted by the first warning unit 5, the multiple risk positions of the target protective net transmitted by the second warning unit 6, and the defect types corresponding to the multiple risk positions of the target protective net transmitted by the defect classification unit 9, and issues an abnormal alarm for the real-time warning information of multiple positions of the target protective net, the multiple risk positions of the target protective net, and the corresponding defect types through an alarm device.

[0052] It should be noted that the alarm device generally refers to a device used to emit sound or visual signals, and its purpose is to issue an alarm to the operator or relevant personnel when the monitoring system detects an abnormality or warning condition, so as to take necessary actions in a timely manner.

[0053] Embodiment 2, a protective net abnormality detection system proposed by the present invention. Compared with Embodiment 1, this embodiment further includes: The risk prediction model includes an encoding module, a state evolution perception module, and a risk prediction module. The encoding module is used to encode the position information, temperature data, humidity data, and vibration data of the target protective net to obtain a low-dimensional feature set, and perform embedded fusion on the low-dimensional feature set to obtain a low-dimensional fusion feature. The state evolution perception module uses a global recurrent memory network, and the global recurrent memory network is used to process the low-dimensional fusion feature of the position information, temperature data, humidity data, and vibration data of the target protective net to obtain a monitoring state representation of the target protective net. The risk prediction module is used to jointly calculate the weight of the current node based on the hidden layer states of all nodes before the current time step, and fuse the monitoring state representation of the target protective net based on the weight of the current node, and classify the fused monitoring state representation through a Sigmoid classifier to obtain multiple risk positions of the target protective net.

[0054] In this embodiment, the low-dimensional feature set is a feature set obtained by performing dimensionality reduction on the original data, and generally has higher information content and less redundancy; the global recurrent memory network is a neural network structure specifically used to process sequence data, which can remember and utilize the information of all past time steps, and helps to capture the long-term dependencies and dynamic changes in the data; the Sigmoid classifier is a commonly used classifier that maps the input data to a probability value between 0 and 1, and is usually used for binary classification problems.

[0055] In an alternative embodiment, the expression of the low-dimensional feature set is as follows:

[0056]

[0057] Wherein, and respectively represent the low-dimensional features of the position information, temperature data, humidity data, and vibration data, and M loc , M tem , M hum and M vib respectively represent the embedding matrices of the position information, temperature data, humidity data, and vibration data to be learned, and respectively represent the one-hot encodings of the position information, temperature data, humidity data, and vibration data;

[0058] The expression of the low-dimensional fusion feature is as follows:

[0059]

[0060] Wherein, x i represents the low-dimensional fusion feature, ReLU represents the activation function, W i represents the potential influence parameter that measures the dependence and the effect of temporal variation, and b i represents the bias vector;

[0061] The expression of the monitoring state representation is as follows:

[0062]

[0063] Wherein, represents the hidden state at the i-th time step, and respectively represent the cell states at the i-th and (i - 1)-th time steps, and LSTM i represents the global recurrent memory network.

[0064] In an optional embodiment, the defect classification model includes a feature extraction module and a feature metric module. The feature extraction module includes 4 residual blocks. Each residual block includes a convolutional layer, a BN layer, and an activation function layer. After the target security net image passes through the 4 residual blocks, a tensor feature map is obtained, and a flattening operation is performed on the tensor feature map to obtain a set of local descriptors. The feature metric module calculates the similarity between the set of local descriptors and the feature aggregation pool, and compares the similarity between the two with a preset similarity threshold. If it is greater than the preset similarity threshold, the target security net image is classified as the type of the template image in the corresponding feature aggregation pool.

[0065] It should be noted that the residual block is a commonly used building block in deep learning, which is used to solve the problems of gradient disappearance and gradient explosion in the training of deep neural networks; the convolutional layer is one of the core layers in deep learning, which is used to extract the spatial features in the input image or feature map; the BN layer is used to normalize the input of each layer during the training of deep neural networks, which helps to speed up the training speed and enhance the generalization ability of the model; the flattening operation converts the multi-dimensional tensor feature map into a one-dimensional vector so that it can be input into the subsequent fully connected layer or other classifiers; the feature aggregation pool refers to a feature template library used to aggregate and store known defect types, which is used to calculate the similarity with the local descriptors of the image to be classified; in the feature metric module, the similarity between the set of local descriptors and the templates in the feature aggregation pool is calculated, and the commonly used methods include Euclidean distance, cosine similarity, etc.; the preset similarity threshold is a predefined boundary used to determine whether the similarity between the local descriptors of the image to be classified and the templates in the feature aggregation pool is high enough to determine the classification type of the image.

[0066] In an optional embodiment, the local descriptor expression is as follows:

[0067] Ψ(Y) = (y 1 ,..., y i ,..., y m );

[0068] where Ψ(Y) represents the set of local descriptors, and y i represents the i-th local descriptor of the tensor feature map.

[0069] Embodiment 3, as Figure 2 shown, a security net anomaly warning method proposed by the present invention, which is applicable to the security net anomaly detection system described above, includes the following steps:

[0070] S1. Use 3D modeling technology to perform 3D modeling on the target security net to obtain a 3D model of the target security net, obtain the geographical map of the location where the target security net is located, perform the matching of the geographical map and the 3D model, and generate a regional mapping table;

[0071] S2. Set fixed monitoring points at multiple positions of the target protection net, set a sensor group at each monitoring point to monitor data such as temperature, humidity, and vibration, obtain the real-time status data of multiple positions of the target protection net, and match the real-time status data with a preset status threshold table to generate real-time warning information;

[0072] S3. Use the trained risk prediction model to predict the risk positions of the target protection net, collect images of the risk positions to obtain the image data of the target protection net, and use the trained defect classification model to classify the defects in the images to obtain the defect types corresponding to each risk position;

[0073] S4. Map the real-time warning information to the corresponding three-dimensional model area for display, highlight it at the risk positions, and give an abnormal alarm through an alarm device.

[0074] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited thereto. Various changes can be made without departing from the spirit of the present invention within the scope of knowledge possessed by those skilled in the art to which the present invention pertains.

Claims

1. A protection net anomaly detection system, comprising a model building unit (1), a region matching unit (2), a state monitoring unit (4), a three-dimensional display unit (3), a first warning unit (5), a second warning unit (6), a real-time alarm unit (7), an image acquisition unit (8) and a defect classification unit (9), characterized in that: The model building unit (1) is used to perform three-dimensional modeling on the target protection net by using a three-dimensional modeling technology to obtain a three-dimensional model of the target protection net, and transmit the three-dimensional model of the target protection net to the area matching unit (2) and the three-dimensional display unit (3); The region matching unit (2) receives the three-dimensional model of the target protection net transmitted by the model building unit (1), obtains a geographical map of the location of the target protection net, matches the geographical map with the three-dimensional model of the target protection net to obtain a region mapping table, and transmits the region mapping table to the three-dimensional display unit (3); The state monitoring unit (4) is used to set fixed monitoring points at multiple positions of the target protection net, set sensor groups at the fixed monitoring points, and monitor the state of the multiple positions of the target protection net through the sensor groups to obtain real-time state data of the multiple positions of the target protection net, wherein the real-time state data includes temperature data, humidity data and vibration data, and transmit the real-time state data of the multiple positions of the target protection net to the first early warning unit (5) and the second early warning unit (6); The first warning unit (5) receives the real-time status data of the multiple positions of the target protection net transmitted by the status monitoring unit (4), matches the real-time status data of the multiple positions of the target protection net with a status threshold table pre-stored in a database to obtain real-time warning information of the multiple positions of the target protection net, and transmits the real-time warning information of the multiple positions of the target protection net to the three-dimensional display unit (3) and the real-time alarm unit (7); The second early warning unit (6) receives the real-time status data of the multiple positions of the target protection net transmitted by the status monitoring unit (4), and predicts the real-time risks of the multiple positions of the target protection net using the trained risk prediction model to obtain multiple risk positions of the target protection net, and transmits the multiple risk positions of the target protection net to the three-dimensional display unit (3), the real-time alarm unit (7) and the image acquisition unit (8); The risk prediction model includes an encoding module, a state evolution perception module and a risk prediction module. The encoding module is used to encode the location information, temperature data, humidity data and vibration data of the target protection network to obtain a low-dimensional feature set, and embed and fuse the low-dimensional feature set to obtain a low-dimensional fusion feature. The state evolution perception module adopts a global recurrent memory network, and the global recurrent memory network is used to process the low-dimensional fusion features of the location information, temperature data, humidity data and vibration data of the target protection network to obtain the monitoring state representation of the target protection network. The risk prediction module is used to jointly calculate the weight of the current node based on the hidden layer states of all nodes before the current time step, and fuse the monitoring state representation of the target protection network based on the weight of the current node, and classify the fused monitoring state representation through a Sigmoid classifier to obtain multiple risk positions of the target protection network.

2. A protection net abnormality detection system according to claim 1, characterized in that: The image acquisition unit (8) receives the multiple risk positions of the target protection net transmitted by the second early warning unit (6), and performs image acquisition on the multiple risk positions of the target protection net to obtain an image of the target protection net, and transmits the image of the target protection net to the defect classification unit (9); The defect classification unit (9) receives the target protection net image transmitted by the image acquisition unit (8), and classifies the target protection net image by using a trained defect classification model to obtain the defect type corresponding to the target protection net image, thereby obtaining the defect types corresponding to the multiple risk positions of the target protection net, and transmitting the defect types corresponding to the multiple risk positions of the target protection net to the three-dimensional display unit (3) and the real-time alarm unit (7).

3. A protection net abnormality detection system according to claim 2, characterized in that: The three-dimensional display unit (3) receives the three-dimensional model of the target protection net transmitted by the model building unit (1), the area mapping table transmitted by the area matching unit (2), the real-time warning information of multiple positions of the target protection net transmitted by the first warning unit (5), the multiple risk positions of the target protection net transmitted by the second warning unit (6), and the defect types corresponding to the multiple risk positions of the target protection net transmitted by the defect classification unit (9), and maps the real-time warning information of the multiple positions of the target protection net to the corresponding areas in the three-dimensional model of the target protection net according to the area mapping table, and displays the real-time warning information in the corresponding areas in the three-dimensional model of the target protection net, matches the multiple risk positions of the target protection net with the area mapping table to obtain the corresponding areas of the multiple risk positions of the target protection net in the three-dimensional model of the target protection net, and highlights the corresponding areas, and displays the defect types corresponding to the risk positions in the corresponding areas; The real-time alarm unit (7) receives the real-time alarm information of the multiple positions of the target protection net transmitted by the first alarm unit (5), the multiple risk positions of the target protection net transmitted by the second alarm unit (6), and the defect types corresponding to the multiple risk positions of the target protection net transmitted by the defect classification unit (9), and issues an abnormal alarm for the real-time alarm information of the multiple positions of the target protection net, the multiple risk positions of the target protection net, and the corresponding defect types through the alarm device.

4. A protection net abnormality detection system according to claim 1, characterized in that: The low-dimensional feature set expression is as follows: ； in, , , and Represent the low-dimensional features of location information, temperature data, humidity data, and vibration data, respectively. , , and Respectively represent the embedding matrices of the location information, temperature data, humidity data, and vibration data to be learned, , , and Represents one-hot encoding of location information, temperature data, humidity data, and vibration data respectively; The low-dimensional fusion feature expression is as follows: ； in, represents low-dimensional fusion features, represents the activation function, represents the potential impact parameter measuring the effects of dependency and timing variation, represents the bias vector; The expression of the monitoring state is as follows: ； in, represents the hidden state at the i-th time step, and denote the cell states at the i-th and i-1-th time steps respectively, Represents a global recurrent memory network.

5. A protection net abnormality detection system according to claim 2, characterized in that: The defect classification model includes a feature extraction module and a feature measurement module. The feature extraction module includes 4 residual blocks. The residual blocks include a convolution layer, a BN layer and an activation function layer. After the target protection net image passes through the 4 residual blocks, a tensor feature map is obtained, and the tensor feature map is flattened to obtain a local descriptor set. The feature measurement module calculates the similarity between the local descriptor set and the feature aggregation pool, and compares the similarity between the two with a preset similarity threshold. If it is greater than the preset similarity threshold, the target protection net image is classified as the type of the corresponding template image in the feature aggregation pool.

6. A protection net abnormality detection system according to claim 5, characterized in that: The local descriptor expression is as follows: ； in, represents a local descriptor set, Represents the i-th local descriptor of the tensor feature map.

7. A protection net abnormality early warning method, which is applicable to a protection net abnormality detection system according to any one of claims 1 to 6, characterized in that: The following steps are involved: S1. Use 3D modeling technology to perform 3D modeling on the target protection network, obtain a 3D model of the target protection network, obtain a geographical map of the location of the target protection network, match the geographical map with the 3D model, and generate a regional mapping table; S2. Set fixed monitoring points at multiple locations of the target protection network, set a sensor group at each monitoring point to monitor temperature, humidity and vibration data, obtain real-time status data of multiple locations of the target protection network, and match the real-time status data with a preset status threshold table to generate real-time warning information; S3. Use the trained risk prediction model to predict the risk position of the target protection net, collect images of the risk position, obtain image data of the target protection net, and use the trained defect classification model to classify the images to obtain the defect type corresponding to each risk position; S4. Map the real-time warning information to the corresponding three-dimensional model area for display, highlight the risk location, and issue an abnormal alarm through the alarm device.

Images