Full-scene feature recognition method and system for illegal behavior in port intelligent monitoring

Abstract

The full-scene feature recognition method and system for illegal behaviors in the intelligent monitoring of ports proposed in this application relate to the field of intelligent monitoring technology. First, the target port is grid-divided, the operation information and meteorological data in the first time interval in the future are processed, and the illegal behaviors with a high probability of occurring in the first time interval are predicted based on the machine learning model. Then, based on the illegal behaviors with a high probability of occurring, simulated violation rules are determined in advance, and a first sensor data set is determined. Different weight information is set for sensor data of different importance in the first sensor data set. Finally, multiple first violation information is predicted based on the first sensor data, and verified through the simulated violation rules determined in advance to determine the target violation information. The technical solution of the present invention analyzes and applies multi-source data to make fine-grained predictions of illegal behaviors in advance.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent monitoring technology, and in particular relates to a method and system for identifying full-scene features of illegal behaviors in intelligent port monitoring. Background Art

[0002] In the field of intelligent port monitoring technology, although existing monitoring systems already possess certain video analysis and anomaly detection capabilities, many technical bottlenecks remain that need to be overcome. Current systems often rely on real-time video streams to identify and alert violations, but lack comprehensive consideration of future operational plans and environmental changes, making it difficult to predict and proactively prevent potential violations. Especially in complex and changing port environments, multiple sources of information, such as weather conditions, vessel dynamics, and loading and unloading plans, have a significant impact on the occurrence of violations. However, existing technologies have not yet effectively integrated this future time-series information for intelligent reasoning and risk prediction. Summary of the Invention

[0003] The purpose of the present invention is to provide a method and system for identifying full-scene features of illegal behaviors in port intelligent monitoring, so as to solve the technical problem of inaccurate identification of port illegal behaviors in the prior art.

[0004] This application proposes a full-scene feature recognition method for illegal behavior in port intelligent monitoring, which includes:

[0005] S1: Preprocessing first operation information and first meteorological information within a first future time interval to obtain first multi-source operation information;

[0006] S2: inputting the first multi-source operation information into a first violation information determination model to obtain first violation information;

[0007] S3: Determine a plurality of first simulated violation rules based on the first violation information;

[0008] S4: determining first sensor weight information according to the first violation information, and determining a first sensor data set based on the first sensor weight information;

[0009] S5: inputting the first sensor data set into a first violation behavior determination model to output a plurality of first violation behaviors;

[0010] S6: Based on the plurality of first violations, determine a plurality of second simulation violation rules from a plurality of the first simulation violation rules, and verify the plurality of first violations using the plurality of the second simulation violation rules to obtain a plurality of target violations.

[0011] Preferably, the S1 includes the following sub-steps:

[0012] S11: Acquire first operation information and first weather information of a target port within a first time interval;

[0013] S12: Performing a first preprocessing and a first spatiotemporal alignment process on the first operation information and the first meteorological information to obtain first multi-source operation information.

[0014] Preferably, the first meteorological information includes: wind speed, wave height, visibility, precipitation probability, and lightning warning data.

[0015] Preferably, the first preprocessing specifically includes:

[0016] Converting the first operation information and the first meteorological information into a standardized JSON format with a unified timestamp;

[0017] The first meteorological information is resampled into a time series that matches an operation time granularity of the first operation information.

[0018] Preferably, the S12 further includes:

[0019] A two-dimensional grid matrix is ​​generated based on the data after the first preprocessing and the first spatiotemporal alignment processing. In the two-dimensional grid matrix, each grid corresponds to at least one specified first operation information, and each grid also corresponds to a meteorological data information.

[0020] Preferably, the first violation information determination model is obtained by:

[0021] S21: Acquire multiple pieces of first historical sample data of the target port;

[0022] S22: performing normalization processing on each piece of the first historical sample data to obtain multiple first historical sample vectors;

[0023] S23: performing a first clustering process on the plurality of first historical sample vectors to obtain a plurality of first cluster centers;

[0024] S24: Obtain the first violation information determination model based on the plurality of first cluster centers.

[0025] Preferably, S3 includes the following sub-steps:

[0026] S31: Determine multiple first violation events based on the first violation information;

[0027] S32: Determine a plurality of first simulated violation rules according to the plurality of first violation events and the first safety operation specification.

[0028] Preferably, the S4 includes the following sub-steps:

[0029] S41: Determine multiple first high-probability violation information based on the first violation information; the first high-probability violation information includes first high-probability violation behavior and first high-probability operating condition information;

[0030] S42: determining a plurality of first sensing data type information for each of the first high-probability violation information;

[0031] S43: for each piece of the first high-probability operating condition information, determine a plurality of second sensor data type information according to the first sensor distribution information of the target port;

[0032] S44: determining first sensor weight information according to the plurality of first sensing data type information and the plurality of second sensing data type information;

[0033] S45: Determine a first sensor data set based on the first sensor weight information and multiple second sensor data type information.

[0034] Preferably, the S6 includes the following sub-steps:

[0035] S61: for each first violation, determining a plurality of second simulated violation rules from a plurality of first simulated violation rules. If the determination is successful, proceed to S62; otherwise, proceed to S63.

[0036] S62: Verify the first violation using the second simulated violation rule. If the verification fails, determine that the first violation is a target violation. Otherwise, output a misjudgment result and proceed to S64.

[0037] S63: Perform a first violation determination on the first violation. If the determination is successful, the first violation is determined to be a target violation. Otherwise, a misjudgment result is output and the process proceeds to S64.

[0038] S64: Get multiple target violations and output them.

[0039] This application also proposes a full-scene feature recognition system for illegal behaviors in port intelligent monitoring, which is used to implement the above-mentioned full-scene feature recognition method for illegal behaviors in port intelligent monitoring.

[0040] The full-scene feature recognition method and system for illegal behaviors in the intelligent monitoring of ports proposed in this application relate to the field of intelligent monitoring technology. First, the target port is grid-divided, the operation information and meteorological data in the first time interval in the future are processed, and the illegal behaviors with a high probability of occurring in the first time interval are predicted based on the machine learning model. Then, based on the illegal behaviors with a high probability of occurring, simulated violation rules are determined in advance, and a first sensor data set is determined. Different weight information is set for sensor data of different importance in the first sensor data set. Finally, multiple first violation information is predicted based on the first sensor data, and verified through the simulated violation rules determined in advance to determine the target violation information. The technical solution of the present invention analyzes and applies multi-source data to make fine-grained predictions of illegal behaviors in advance. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for the embodiments or the description of the prior art. Obviously, the drawings described below are merely exemplary, and those skilled in the art can derive other implementation drawings based on the provided drawings without inventive effort.

[0042] Figure 1 It is an execution flow chart of the full-scene feature recognition method for illegal behavior in port intelligent monitoring in the present invention.

[0043] Figure 2 It is a schematic diagram of the grid coordinate system of the port operation area in the present invention. DETAILED DESCRIPTION

[0044] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0045] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. The exemplary embodiments and descriptions are only used to explain the present invention but are not intended to limit the present invention.

[0046] The following is a detailed description of the method and system for identifying full-scene features of violations in port intelligent monitoring of the present invention.

[0047] This embodiment proposes a full-scene feature recognition method for illegal behavior in port intelligent monitoring. The specific process is as follows: Figure 1 shown.

[0048] S1: Preprocessing first operation information and first meteorological information within a first future time interval to obtain first multi-source operation information.

[0049] In order to accurately predict the target violation behavior that will occur with a higher probability in the first time interval in the future, the first operation information and first meteorological information corresponding to the first time interval will be retrieved in this step. Since the violation probability of different types of operations under specified meteorological conditions will also be different, the target operation information and target meteorological information in the first time interval can be used to make a preliminary prediction of the violation behavior with a higher probability of occurring in the first time interval.

[0050] The S1 may specifically include the following sub-steps:

[0051] S11: Acquire first operation information and first weather information of a target port within a first time interval.

[0052] The target port is the port where the violation prediction is to be performed, and various types of operations may occur within the area defined by the target port.

[0053] The first operation information includes various types of operation information. Specific operation information may include: ship berthing and unberthing operations, which refer to operations performed when a ship enters or leaves a port, including navigation, berthing, and unberthing; container operations, which involve the loading, unloading, stacking, unpacking, and repacking of containerized cargo; bulk cargo, general cargo, and piece cargo operations, which involve the loading, unloading, and storage of dry bulk cargo such as coal, ore, and grain, as well as various piece cargoes in bags, bundles, and barrels; and crude oil and liquefied natural gas operations, which specifically focus on the loading, unloading, and storage of liquid and gas cargoes such as oil, chemicals, and liquefied natural gas. The above operation types are provided as examples and are not exhaustive.

[0054] The first meteorological information may include various meteorological factors that affect port operations, including, for example, refined forecast data such as wind speed, wave height, visibility, precipitation probability, and lightning warning.

[0055] The first operation information can obtain corresponding data by calling the API interface of the port operation system, and the first meteorological information can obtain accurate meteorological data by accessing the meteorological forecast system.

[0056] S12: Performing a first preprocessing and a first spatiotemporal alignment process on the first operation information and the first meteorological information to obtain first multi-source operation information.

[0057] For a specified type of port operation information, the probability of violations will change significantly with meteorological information. Therefore, in this step, it is first necessary to perform the first preprocessing operation to improve the outliers of the two types of information, and format them to achieve the unification of data types. Then, the two types of data are aligned in time and space to facilitate the subsequent correlation prediction of the target violation.

[0058] The first preprocessing specifically includes:

[0059] Convert the first operation information and the first weather information into a standardized JSON format with a unified timestamp (UTC+8 time zone);

[0060] Resampling the first meteorological information into a time series that matches the operation time granularity of the first operation information (e.g., every hour / every three hours);

[0061] The sliding window method is used to detect the period of missing operation records (the window size is preferably 1 hour), the KNN algorithm is used to interpolate the missing meteorological data to the neighboring grid, and obvious data anomalies are corrected.

[0062] The first spatiotemporal alignment process refers to matching the spatiotemporal coordinates of the first operation information and the first weather information. This is because the coordinate positions of different first operation information may be different, and the weather information at different locations may also be different. For example, the weather information such as waves and visibility may be significantly different between a location near a shore, a location on the shore, and a location at sea. The first spatiotemporal alignment process specifically includes:

[0063] Establish a grid coordinate system for the port operation area (preferably a 500m×500m grid);

[0064] The meteorological data is spatially associated with the location of the first operation information (e.g., Pier 3 corresponds to meteorological grid G07, Figure 2 shown).

[0065] A two-dimensional grid matrix is ​​generated based on the data after the first preprocessing and the first spatiotemporal alignment processing. In the two-dimensional grid matrix, each grid corresponds to at least one specified first operation information, and each grid also corresponds to a meteorological data information.

[0066] If a grid corresponds to multiple pieces of first operation information and multiple pieces of meteorological data, primary operation information and primary meteorological data need to be selected. The selection rules include: for operation information, selecting the first operation information with the largest area as the primary operation information; for meteorological data, selecting the meteorological factors that have the greatest impact on the primary operation information as the primary meteorological data.

[0067] The first multi-source operation information is determined according to the two-dimensional grid matrix. That is, the first multi-source operation information includes a correspondence between each piece of the first operation information and corresponding meteorological data.

[0068] Specifically, the first multi-source operation information includes the following four-dimensional data (time × location × operation type × weather information). For example, during the time interval of 8:00 AM to 12:00 PM on the 10th, within the (2, 2)th grid, the operation type is container unloading, and the weather information within this grid (such as wind speed, rainfall, temperature changes, wave height, and visibility) is as follows.

[0069] S2: Inputting the first multi-source operation information into a first violation information determination model to obtain first violation information.

[0070] The first violation information determination model is used to determine the probability of occurrence of various types of violation behaviors within the first time interval. The first violation information records several types of violation behavior information with a high probability of occurrence within the first time interval, as well as the corresponding occurrence probability of each type of violation behavior information.

[0071] The first violation information determination model is obtained by the following steps:

[0072] S21: Acquire a plurality of first historical sample data of the target port.

[0073] Each piece of the first historical sample data is the operation data of the target port in the past preset time period, and the operation data includes operation data generated within a specified time interval and a specified coordinate range and composed of operation type, weather information and violation information.

[0074] S22: Performing standardization processing on each piece of the first historical sample data to obtain multiple first historical sample vectors.

[0075] Each of the first historical sample vectors includes data in five dimensions, including time, coordinates, operation type, weather information, and violation information.

[0076] S23: Performing a first clustering process on the plurality of first historical sample vectors to obtain a plurality of first cluster centers.

[0077] Because each first historical sample vector is a five-dimensional vector, it captures violation information at the target port under different time, coordinate locations, operational information, and weather data. When the sample size is sufficiently large, this data is sufficient to characterize the probability of violations occurring at the target port under different operational scenarios. For example, in the container unloading scenario, multiple first historical sample vectors may capture violation information generated under different coordinate and weather conditions at the target port.

[0078] In order to further explore the occurrence patterns of illegal behaviors, in this step, it is necessary to perform a first clustering process on the plurality of the first historical sample vectors, for example, a K-MEANS clustering algorithm may be used to obtain a plurality of cluster centers.

[0079] S24: Obtain the first violation information determination model based on the plurality of first cluster centers.

[0080] That is, the first violation information determination model is mainly composed of a plurality of the first cluster centers.

[0081] Preferably, the first violation information determination model is further configured to receive input first multi-source operation information, calculate the Euclidean distance between the first multi-source operation information and a plurality of the first cluster centers, and select the first N first cluster centers with the shortest Euclidean distance as target cluster centers. N can be selected based on the specific circumstances. For example, when 3 is selected, three target cluster centers can be determined. Furthermore, the violations corresponding to these three target cluster centers can be used as the first violation information corresponding to the first multi-source operation information. The probability of each type of violation occurring is negatively correlated with the length of the corresponding Euclidean distance, i.e., the shorter the Euclidean distance, the greater the probability of occurrence.

[0082] Preferably, the first violation information also includes all or part of the first multi-source operation information corresponding to the high-probability violation, specifically including time, location coordinates, etc. This information in the first violation information can be used to subsequently determine the corresponding sensor data.

[0083] S3: Determine a plurality of first simulated violation rules based on the first violation information.

[0084] Based on the first violation information, the boundary conditions corresponding to various types of violations can be determined. Specifically, stacking violations can be restricted through three-dimensional geographic fences, and operation process violations can be restricted through preset timing rules.

[0085] Through this step, multiple corresponding first simulated violation rules can be determined in advance based on the first violation information that has a high degree of consistency with the first operation information and the first meteorological information in the first time period in the future. Therefore, when a certain type of violation is subsequently determined to have occurred, the corresponding simulated violation rules can be quickly retrieved to quickly verify the predicted violation, thereby ensuring the real-time and accuracy of the early warning.

[0086] The S3 specifically includes the following sub-steps:

[0087] S31: Determine a plurality of first violation events based on the first violation information.

[0088] In S2, multiple violations have been identified and the first violation information has been obtained. In this step, multiple first violation events can be determined from the first violation information. The first violation event refers to a violation with a high probability of occurring within a first time interval in the future.

[0089] S32: Determine a plurality of first simulated violation rules according to the plurality of first violation events and the first safety operation specification.

[0090] Each first violation event is a textual description of each violation. The first safety operation specification is the regulation that must be followed to ensure safe operation. For example, when the wind speed is ≥14m / s, bridge crane operation is prohibited; when visibility is less than 500m, tugboat dispatching frequency is reduced by 50%.

[0091] When this step is specifically performed, the first violation event and the first safety operation specification can be semantically analyzed and similarity matched, and the rule with the highest matching degree can be used as the first simulated violation rule, so that corresponding multiple first simulated violation rules can be obtained for multiple first violation events.

[0092] S4: Determine first sensor weight information according to the first violation information, and determine a first sensor data set based on the first sensor weight information.

[0093] Each type of first violation information in S2 corresponds to a different sensor data weight. For example, for a violation involving a hazardous materials leak, the sensor data output by the temperature and hazardous gas sensors is more important, while for a violation involving improper helmet wearing, the sensor data output by the image sensor is generally more important. Furthermore, when determining the sensor data weights, it is also necessary to extract the time and location of the high-probability occurrence of the corresponding violation from the first violation information, thereby facilitating the retrieval of sensor data at the corresponding time and location.

[0094] Therefore, through this step, weight information corresponding to various types of sensor data can be determined based on the first violation information, and the measured sensor data can be weighted based on the obtained first sensor weight information to obtain the first sensor set.

[0095] The S4 may specifically include the following sub-steps:

[0096] S41: Determine a plurality of first high-probability violation information based on the first violation information.

[0097] In step S2, multiple high-probability violations corresponding to the first multi-source operation information have been determined based on the first violation information determination model. Therefore, in this step, multiple high-probability violations and corresponding multiple operation condition information can be extracted from the first violation information.

[0098] The first high-probability violation information includes first high-probability violation behaviors and first high-probability operation condition information, and the first high-probability operation condition information includes violation time information and violation coordinate position information corresponding to each first high-probability violation behavior.

[0099] S42: For each piece of the first high-probability violation information, determine a plurality of first sensing data type information.

[0100] Since different violations have different characteristics, the sensor data types used to detect the violations will also vary greatly. Therefore, in this step, for each of the first high-probability violation information, the sensor data type that will be given priority consideration when detecting the violation information is determined as multiple first sensor data type information.

[0101] S43: For each of the first high-probability operating condition information, determine a plurality of second sensing data type information according to the first sensor distribution information of the target port.

[0102] In order to conduct all-round and multi-angle monitoring of the target port while taking into account the principles of economy and efficiency, various types of sensors are usually arranged in the target port according to certain layout specifications.

[0103] In this step, for each piece of the first high-probability operating condition information, it is necessary to determine a plurality of sensor nodes that will have a monitoring relationship with it in the sensor layout data of the target port, thereby obtaining a plurality of second sensing data type information.

[0104] S44: Determine first sensor weight information according to the plurality of first sensing data type information and the plurality of second sensing data type information.

[0105] In this step, the number of sensors covered by the multiple second sensor data type information is greater than the number of first sensor data type information. Therefore, among the multiple second sensor data type information, higher weights can be set for the multiple first sensor data type information, and lower weights can be set for the remaining multiple second sensor data type information.

[0106] Preferably, among the plurality of first sensor data type information, weights may be differentiatedly set according to human experience using a step rule or other rules, so as to make the determined first sensor weight information more consistent with actual conditions.

[0107] S45: Determine a first sensor data set based on the first sensor weight information and multiple second sensor data type information.

[0108] In this step, the first sensor weight information is used to perform standardization and weighting processing on a plurality of second sensor data type information one by one to obtain the first sensor data set.

[0109] The first sensor data set includes all sensor data types in the second sensor data type information, but the weight of each sensor data type may vary according to the first sensor weight information.

[0110] S5: Inputting the first sensor data set into a first violation behavior determination model to output a plurality of first violation behaviors.

[0111] Preferably, the first violation determination model is established based on a multi-granularity feature pyramid network (MGFP-Net), and can be analyzed from three dimensions: macro, meso and micro based on the current first sensor data set, so as to accurately determine multiple first violations, where each first violation refers to a violation whose probability of occurrence is greater than a preset value.

[0112] The first violation behavior determination model can perform a comprehensive analysis and processing on the first sensor data set from three dimensions, so that the multiple first violation behaviors obtained are more accurate and in line with the actual situation.

[0113] In another embodiment, the first violation determination model can also be trained based on a common convolutional neural network model CNN, and the training method can adopt well-known means, which will not be described here.

[0114] The multiple first violation behaviors determined all include violation type information, violation time information, and violation location information.

[0115] S6: Based on the plurality of first violations, determine a plurality of second simulation violation rules from a plurality of the first simulation violation rules, and verify the plurality of first violations using the plurality of the second simulation violation rules to obtain a plurality of target violations.

[0116] There may be misjudgments among the multiple first violations determined in step S5. Therefore, it is necessary to find multiple first simulation violation rules corresponding to the multiple first violations from the multiple first simulation violation rules determined in advance, and then verify each first violation through the corresponding first simulation violation rule to determine whether each first violation is correctly judged, and determine the multiple first violation behaviors that are correctly judged as multiple target violations.

[0117] The S6 comprises the following sub-steps:

[0118] S61: For each of the first violation behaviors, determine whether there is a corresponding first simulated violation rule. If so, proceed to S62; otherwise, proceed to S63.

[0119] In S3, a plurality of the first simulated violation rules have been established in advance for violation information with a high probability of occurrence. Therefore, in this step, it is necessary to first determine whether a corresponding first simulated violation rule exists for each of the first violation behaviors.

[0120] S62: Use the first simulated violation rule to verify the first violation behavior. When the verification fails, determine that the first violation behavior is a target violation behavior; otherwise, output a misjudgment result and go to S64.

[0121] Since the first simulated violation rule defines correct and required operating conditions, if the verification fails, that is, if the first violation does not comply with the first simulated violation rule, it indicates that the first violation is indeed a violation and is therefore determined as the target violation. Otherwise, it indicates that the first violation is a misjudgment.

[0122] S63: Perform a first violation determination on the first violation. If the determination is successful, determine that the first violation is a target violation. Otherwise, output a misjudgment result and proceed to S64.

[0123] Preferably, the first violation determination can be performed through expert experience, or by setting determination rules.

[0124] S64: Get multiple target violations and output them.

[0125] This application also proposes a full-scene feature recognition system for illegal behaviors in port intelligent monitoring, which is used to execute the above-mentioned full-scene feature recognition method for illegal behaviors in port intelligent monitoring.

[0126] The full-scene feature recognition method and system for illegal behaviors in the intelligent monitoring of ports proposed in this application relate to the field of intelligent monitoring technology. First, the target port is grid-divided, the operation information and meteorological data in the first time interval in the future are processed, and the illegal behaviors with a high probability of occurring in the first time interval are predicted based on the machine learning model. Then, based on the illegal behaviors with a high probability of occurring, simulated violation rules are determined in advance, and a first sensor data set is determined. Different weight information is set for sensor data of different importance in the first sensor data set. Finally, multiple first violation information is predicted based on the first sensor data, and verified through the simulated violation rules determined in advance to determine the target violation information. The technical solution of the present invention analyzes and applies multi-source data to make fine-grained predictions of illegal behaviors in advance.

[0127] The above description is only a preferred embodiment of the present invention. Therefore, any equivalent changes or modifications made according to the structure, characteristics and principles described in the scope of the patent application of the present invention are included in the scope of the patent application of the present invention.

Claims

1. A full-scene feature recognition method for illegal behaviors in port intelligent monitoring, characterized by: The method includes: S1: Preprocessing first operation information and first meteorological information within a first future time interval to obtain first multi-source operation information; The S1 includes the following sub-steps: S11: Acquire first operation information and first weather information of a target port within a first time interval; S12: performing a first preprocessing and a first spatiotemporal alignment process on the first operation information and the first meteorological information to obtain first multi-source operation information; The first preprocessing specifically includes: Converting the first operation information and the first meteorological information into a standardized JSON format with a unified timestamp; resampling the first meteorological information into a time series that matches an operation time granularity of the first operation information; S2: inputting the first multi-source operation information into a first violation information determination model to obtain first violation information; S3: Determine a plurality of first simulated violation rules based on the first violation information; S4: determining first sensor weight information according to the first violation information, and determining a first sensor data set based on the first sensor weight information; The S4 includes the following sub-steps: S41: Determine multiple first high-probability violation information based on the first violation information; the first high-probability violation information includes first high-probability violation behavior and first high-probability operating condition information; S42: determining a plurality of first sensing data type information for each of the first high-probability violation information; S43: for each piece of the first high-probability operating condition information, determine a plurality of second sensor data type information according to the first sensor distribution information of the target port; S44: determining first sensor weight information according to the plurality of first sensing data type information and the plurality of second sensing data type information; S45: Determine a first sensor data set based on the first sensor weight information and multiple second sensor data type information; S5: inputting the first sensor data set into a first violation behavior determination model to output a plurality of first violation behaviors; S6: Based on the plurality of first violations, determine a plurality of second simulation violation rules from a plurality of the first simulation violation rules, and verify the plurality of first violations using the plurality of the second simulation violation rules to obtain a plurality of target violations.

2. The method for identifying full-scene features of illegal behaviors in port intelligent monitoring according to claim 1 is characterized in that: The first meteorological information includes: wind speed, wave height, visibility, precipitation probability, and lightning warning data.

3. The method for identifying full-scene features of illegal behaviors in port intelligent monitoring according to claim 2 is characterized in that: The S12 further includes: A two-dimensional grid matrix is ​​generated based on the data after the first preprocessing and the first spatiotemporal alignment processing. In the two-dimensional grid matrix, each grid corresponds to at least one specified first operation information, and each grid also corresponds to a meteorological data information.

4. The method for identifying full-scene features of illegal behaviors in port intelligent monitoring according to claim 3 is characterized in that: The first violation information determination model is obtained by: S21: Acquire multiple pieces of first historical sample data of the target port; S22: performing normalization processing on each piece of the first historical sample data to obtain multiple first historical sample vectors; S23: performing a first clustering process on the plurality of first historical sample vectors to obtain a plurality of first cluster centers; S24: Obtain the first violation information determination model based on the plurality of first cluster centers.

5. The method for identifying full-scene features of illegal behaviors in port intelligent monitoring according to claim 4 is characterized in that: The S3 includes the following sub-steps: S31: Determine multiple first violation events based on the first violation information; S32: Determine a plurality of first simulated violation rules according to the plurality of first violation events and the first safety operation specification.

6. The method for identifying full-scene features of illegal behaviors in port intelligent monitoring according to claim 5 is characterized in that: The S6 comprises the following sub-steps: S61: for each first violation, determining a plurality of second simulated violation rules from a plurality of first simulated violation rules. If the determination is successful, proceed to S62; otherwise, proceed to S63. S62: Verify the first violation using the second simulated violation rule. If the verification fails, determine that the first violation is a target violation. Otherwise, output a misjudgment result and proceed to S64. S63: Perform a first violation determination on the first violation. If the determination is successful, the first violation is determined to be a target violation. Otherwise, a misjudgment result is output and the process proceeds to S64. S64: Get multiple target violations and output them.

7. A system for identifying features of violations in all scenarios during intelligent port monitoring, for implementing the method for identifying features of violations in all scenarios during intelligent port monitoring as described in any one of claims 1 to 6.

Images