Multi-mode electronic fence forbidden zone intrusion identification method and system based on AI

Abstract

The invention provides a multi-mode electronic fence forbidden zone intrusion identification method and system based on AI. The method comprises the following steps: firstly, acquiring synchronous multi-modal data acquired by a heterogeneous sensor network; secondly, according to the real-time production activity state, the type attribute and the real-time vehicle scheduling information, a differential perception focusing strategy is generated; performing weighted focusing fusion on the synchronous multi-modal data according to a differential sensing focusing strategy to generate dynamic intrusion sensing features; and then performing pattern analysis through an artificial intelligence model based on the dynamic intrusion sensing features, and finally identifying an intrusion behavior which accords with a preset factory safety regulation definition. According to the technical scheme provided by the invention, the identification precision of real threats in a complex industrial environment and the overall adaptability of the system are improved, and high-precision and self-adaptive intelligent identification of forbidden zone intrusion behaviors in the complex industrial environment is also realized.

Description

Technical Field

[0001] This application relates to the field of video analytics technology, and in particular to an AI-based multimodal electronic fence restricted area intrusion identification method and system. Background Technology

[0002] In the field of factory safety production, real-time perimeter monitoring and safety protection of key areas such as production workshops, hazardous material storage areas and factory passages is a core requirement. With the improvement of factory intelligence, traditional physical isolation methods are no longer sufficient to cope with the safety challenges in complex environments. Therefore, there is an urgent need for a restricted area intrusion identification technology that can intelligently sense, accurately identify and proactively warn, so as to dynamically adapt to the complex safety situation formed by the intertwining of multiple factors such as production activities, hazardous material storage and logistics scheduling in the factory.

[0003] Currently, most existing technical solutions rely on high-definition video surveillance systems deployed in factory areas, combined with specific behavior recognition algorithms to achieve intrusion detection. These solutions analyze moving targets in the video stream and determine whether an intrusion event has occurred based on preset rules (such as whether personnel or vehicles have entered the designated virtual electronic fence area).

[0004] However, existing solutions based on a single visual modality have significant drawbacks. First, their recognition effectiveness is highly dependent on environmental conditions such as lighting and weather, and their reliability is significantly reduced at night or in inclement weather. Second, the solutions struggle to effectively distinguish between authorized normal work activities and genuine security threats; for example, they cannot effectively identify unauthorized personnel heat sources and the heat radiation from equipment itself near hazardous materials storage areas. Third, due to a lack of perception and fusion capabilities regarding the factory environment context (such as production status and hazardous material properties), the system cannot achieve dynamic and differentiated monitoring that matches the real-time risk level, resulting in a high false alarm rate or a high risk of missed alarms, making it difficult to meet the precise security needs of complex industrial scenarios. Summary of the Invention

[0005] This application provides an AI-based multimodal electronic fence restricted area intrusion identification method and system to solve the problems of existing identification schemes based on a single visual modality being greatly constrained by environmental conditions, unable to effectively distinguish between authorized activities and security threats, and having high false alarm rates and poor adaptability due to a lack of understanding of the complex operating context of the factory.

[0006] Firstly, this application provides an AI-based multimodal electronic fence restricted area intrusion identification method, including: Acquire synchronous multimodal data in real time from a heterogeneous sensor network pre-deployed in a factory area, wherein the factory area includes factory passages, hazardous materials storage areas and production workshops, and the synchronous multimodal data includes visible light video data, thermal infrared image data and audio feature segment data; Based on the real-time production activity status of the production workshop, the type and attributes of the hazardous materials stored in the hazardous materials storage area, and the real-time vehicle scheduling information of the factory area passage, a differentiated perception focusing strategy is generated. Based on the differentiated perception focusing strategy, the synchronous multimodal data is weighted and focused to generate dynamic intrusion perception features; Based on the dynamic intrusion perception characteristics, pattern analysis is performed using an artificial intelligence model to identify intrusion behaviors that conform to the preset factory safety procedures.

[0007] Optionally, synchronous multimodal data pre-deployed in the factory area is acquired in real time from a heterogeneous sensor network. This synchronous multimodal data includes visible light video data, thermal infrared image data, and audio feature segment data, including: Visible light video data containing moving objects is acquired from cameras pre-positioned in the factory area passageways; thermal infrared image data characterizing the thermal radiation distribution of objects is acquired from infrared sensors facing the hazardous materials storage area; and raw audio streams containing equipment operating sounds are collected from audio sensors outside the production workshop. The original audio stream is parsed to separate the periodic beat signal corresponding to the operating cycle of the production equipment in the production workshop, and abnormal sound event segments that are different from the periodic beat signal are extracted. The abnormal sound event segments are analyzed to determine the start time point, end time point, and sound energy change pattern within the start time point and end time point of each abnormal sound event segment, so as to form audio feature segment data; The visible light video data, the thermal infrared image data, and the audio feature segment data are integrated to form a multimodal data stream.

[0008] Optionally, a differentiated perception focusing strategy is generated based on the real-time production activity status of the production workshop, the type and attributes of the hazardous materials stored in the hazardous materials storage area, and the real-time vehicle scheduling information of the factory area access roads, including: Receive real-time production activity status information from the production workshop; A first perception priority is defined for the audio feature segment data based on the real-time production activity status information. Receive type and attribute information of hazardous materials stored in the hazardous materials storage area, including the volatility and heat sensitivity of the hazardous materials; The monitoring granularity is set for the thermal infrared image data based on the type attribute information; Receive real-time vehicle dispatch information for the factory area access road, the real-time vehicle dispatch information including the location and passage time window of the planned transport vehicles; Based on the real-time vehicle dispatch information, a dynamic sensing area is defined for the visible light video data. The first sensing priority, the monitoring granularity, and the sensing area range are combined to generate a differentiated sensing focusing strategy.

[0009] Optionally, according to the differentiated perception focusing strategy, the synchronous multimodal data is weighted and focused fused to generate dynamic intrusion perception features, including: Based on the first perception priority, assign a primary attention value to the audio feature segment data; Based on the monitoring granularity, the level of detail required for processing the thermal infrared image data is determined, and the thermal infrared image data is finely adjusted to obtain clear thermal infrared image data. Based on the defined dynamic sensing area, image sequence content that matches the sensing area is selected from the visible light video data; The audio feature segment data assigned the primary concern value, the clear thermal infrared image data, and the selected visible light video image sequence content are integrated to generate dynamic intrusion detection features.

[0010] Optionally, based on the monitoring granularity, the level of detail in processing the thermal infrared image data is determined, and the thermal infrared image data is finely adjusted to obtain clear thermal infrared image data, including: Based on the determined level of detail, the aforementioned thermal infrared image data is divided into multiple continuous temperature ranges; Based on the type and attribute information of the hazardous materials, the key temperature ranges directly related to potential risks are determined from the multiple temperature ranges that have been divided. The thermal infrared image data is processed to enhance the image detail within the critical temperature range, while weakening the image detail within the non-critical temperature range, resulting in enhanced image data. In the enhanced image data, the connected regions formed by all pixels within the critical temperature range are identified to generate clear thermal infrared image data.

[0011] Optionally, based on the dynamic intrusion perception characteristics, pattern analysis is performed using an artificial intelligence model to identify intrusion behaviors that conform to the preset factory safety procedures, including: The dynamic intrusion perception features are input into a pre-trained artificial intelligence model, which parses the dynamic intrusion perception features into a moving target sequence corresponding to visible light video data, an abnormal heat source distribution corresponding to thermal infrared image data, and a sound event label corresponding to audio feature segment data. Analyze the moving target sequence to determine whether the trajectory of the moving target crosses the electronic fence boundary defined by the factory safety regulations, and determine the speed change pattern of the moving target; Analyze the distribution of the abnormal heat sources to determine whether the heat sources are located within the restricted area of ​​the hazardous materials storage area, and distinguish whether the heat sources are stationary or moving. The analysis results of the moving target sequence, the analysis results of the abnormal heat source distribution, and the sound event tags are combined to generate multi-dimensional judgment criteria. The multi-dimensional judgment criteria are matched with multiple intrusion behavior rules predefined in the factory safety procedures to obtain the matching results; When the matching results show that the multi-dimensional judgment criteria satisfy all the conditions of any one of the multiple intrusion behavior rules in terms of time sequence and spatial location, it is judged as an intrusion behavior.

[0012] Optionally, the multi-dimensional judgment criteria are matched with multiple predefined intrusion behavior rules in the factory safety procedures to obtain matching results, including: Establish the correspondence between each intrusion behavior rule in the factory safety regulations and the factory area, wherein the first type of rule corresponds to the production workshop area, the second type of rule corresponds to the hazardous materials storage area, and the third type of rule corresponds to the factory area passageway; Based on the spatial location information contained in the multi-dimensional judgment criteria, the analysis results of the moving target sequence, the analysis results of the abnormal heat source distribution, and the sound event tags are mapped to the corresponding physical regions. A correlation analysis is conducted on the multi-dimensional judgment criteria mapped to the same physical region to determine the chronological order and spatial relative relationship of moving targets, abnormal heat sources, and sound events on the time axis. The multi-dimensional judgment criteria after correlation analysis are compared with the intrusion behavior rules of the corresponding physical area one by one to check whether the trajectory characteristics of the moving target, the state characteristics of the heat source, and the type characteristics of the sound event simultaneously meet the condition combination defined by the rule. When all feature conditions meet the condition combination defined by the satisfying rule within the preset time tolerance and spatial tolerance range, a matching result indicating a successful match is generated.

[0013] Secondly, this application provides an AI-based multimodal electronic fence restricted area intrusion identification system, including: The acquisition module is used to acquire synchronous multimodal data collected in real time by a heterogeneous sensor network pre-deployed in the factory area. The factory area includes factory passages, hazardous materials storage areas, and production workshops. The synchronous multimodal data includes visible light video data, thermal infrared image data, and audio feature segment data. The generation module is used to generate a differentiated perception focusing strategy based on the real-time production activity status of the production workshop, the type and attributes of the hazardous materials stored in the hazardous materials storage area, and the real-time vehicle scheduling information of the factory area passage. The fusion module is used to perform weighted focusing fusion on the synchronous multimodal data according to the differentiated perception focusing strategy to generate dynamic intrusion perception features; The identification module is used to identify intrusion behaviors that conform to the preset factory safety procedures by performing pattern analysis through an artificial intelligence model based on the dynamic intrusion perception characteristics.

[0014] Thirdly, this application provides a computing device, including a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are to be invoked and executed by the processing component to implement an AI-based multimodal electronic fence restricted area intrusion identification method as described in the first aspect above.

[0015] Fourthly, this application provides a computer storage medium storing a computer program, which, when executed by a computer, implements an AI-based multimodal electronic fence restricted area intrusion identification method as described in the first aspect.

[0016] This application constructs a more comprehensive environmental perception capability by introducing data from three modalities: visible light, thermal infrared, and audio, effectively overcoming the dependence of a single visual modality on lighting and weather conditions. More importantly, the solution creatively transforms the real-time operational context of the factory (such as production status, hazardous material attributes, and vehicle scheduling) into an operable "differentiated perception focusing strategy," enabling the system to dynamically adjust monitoring priorities and achieve intelligent matching of perception resources with real-time risks. This adaptive data fusion and analysis mechanism based on scene understanding improves the accuracy of identifying real threats in complex industrial environments and the overall adaptability of the system.

[0017] Furthermore, by establishing a clear correspondence between rules and physical areas, and by performing correlation analysis on multi-dimensional perceived information in the spatiotemporal dimensions, the accuracy and rationality of the judgment are ensured. This method transforms abstract safety procedures into quantifiable, multi-condition combination judgment rules bound to specific areas. By examining the spatiotemporal synergy of moving targets, heat sources, and sound events to make the final decision, it effectively avoids false alarms caused by isolated judgment of a single signal, making the identification results more consistent with the actual operational logic of factory safety.

[0018] These or other aspects of this application will become more apparent in the following description of the embodiments. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 A flowchart of an AI-based multimodal electronic fence restricted area intrusion identification method provided in this application is shown; Figure 2 This paper presents a schematic diagram of the structure of an AI-based multimodal electronic fence restricted area intrusion identification system provided in this application. Figure 3 A schematic diagram of the structure of a computing device provided in this application is shown. Detailed Implementation

[0021] To enable those skilled in the art to better understand the present application, the technical solution of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0022] In some of the processes described in the specification, claims, and accompanying drawings of this application, multiple operations appearing in a specific order are included. However, it should be clearly understood that these operations may not be executed in the order they appear herein, or may be executed in parallel. The operation numbers, such as 101, 102, etc., are merely used to distinguish different operations and do not themselves represent any execution order. Furthermore, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the descriptions such as "first," "second," etc., in this document are used to distinguish different messages, devices, modules, etc., and do not represent a chronological order, nor do they limit "first" and "second" to different types.

[0023] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0024] Figure 1 This application provides a flowchart of an AI-based multimodal electronic fence restricted area intrusion identification method, such as... Figure 1 As shown, the method includes: Step 101: Acquire synchronous multimodal data collected in real time by a heterogeneous sensor network pre-deployed in the factory area. The factory area includes factory passages, hazardous materials storage areas, and production workshops. The synchronous multimodal data includes visible light video data, thermal infrared image data, and audio feature segment data.

[0025] Optionally, step 101 may specifically include the following steps: Step 1011: Acquire visible light video data containing moving objects from cameras pre-positioned in the factory area passageways; acquire thermal infrared image data characterizing the thermal radiation distribution of objects from infrared sensors facing the hazardous materials storage area; and simultaneously collect raw audio streams containing equipment operating sounds from audio sensors outside the production workshop. Step 1012: Analyze the original audio stream, separate the periodic beat signal corresponding to the operating cycle of the production equipment in the production workshop, and extract abnormal sound event segments that are different from the periodic beat signal. Step 1013: Analyze the abnormal sound event segments to determine the start time point, end time point, and sound energy change pattern within the start time point and end time point of each abnormal sound event segment, so as to form audio feature segment data; Step 1014: Integrate the visible light video data, the thermal infrared image data, and the audio feature segment data to form a multimodal data stream.

[0026] In the above scheme, the factory area refers to the entire industrial site under electronic fence monitoring, including factory access roads (road areas for vehicles and personnel to pass through), hazardous materials storage areas (dedicated warehouses for storing flammable, explosive or toxic materials), and production workshops (core work areas for product manufacturing). Heterogeneous sensor networks are monitoring systems composed of various types of sensors (such as cameras, infrared sensors, and audio sensors) used to acquire synchronous multimodal data (i.e., multiple time-aligned data streams), including visible light video data (a sequence of dynamic images in the visible light band captured by a camera for tracking moving objects), thermal infrared image data (a temperature distribution image generated by infrared radiation from an object's surface perceived by an infrared sensor), and audio feature segment data (segments with time stamps and acoustic features extracted from an audio stream). An infrared sensor is a device that detects infrared radiation and generates a thermal image; its output shows the distribution of thermal radiation from an object, reflecting changes in surface temperature. Audio sensors (such as microphones) capture raw audio streams that are unprocessed, continuous sound signals; Periodic beat signals are repetitive sound patterns that correspond to the regular operation of production equipment; Abnormal sound event segments are abnormal sound intervals that deviate from the repetitive sound pattern; The start and end times define the time range in which the event occurs; The sound energy change pattern describes the characteristics of sound intensity changing over time during an event period; The resulting audio feature segment data refers to data used to characterize abnormal audio events.

[0027] In this solution, firstly, visible light video data containing moving objects (such as vehicles or personnel) is acquired in real time from cameras pre-deployed in the factory's passageways. Simultaneously, thermal infrared image data characterizing the thermal radiation distribution of objects is collected from infrared sensors facing the hazardous materials storage area (temperature distribution maps are generated by detecting surface infrared radiation). Furthermore, raw audio streams containing equipment operating sounds are simultaneously acquired from audio sensors located around the production workshop, ensuring that the multi-source data are time-aligned. Secondly, the raw audio stream is analyzed, and signal processing techniques (such as filtering and pattern matching) are used to separate the periodic beat signals (background sounds of normal operation) corresponding to the operating cycles of the equipment in the production workshop, and these signals are extracted. Unlike the abnormal sound event segments (such as sudden impacts or unusual noises) of the periodic beat signal, these segments are identified by comparing them with normal patterns. Then, each abnormal sound event segment is analyzed in detail. The start and end times are determined by detecting changes in sound energy thresholds, and the sound energy change patterns (such as energy peaks and durations) within that time period are analyzed to form structured audio feature segment data. Finally, the acquired visible light video data, thermal infrared image data, and audio feature segment data are integrated by timestamps, and a unified multimodal data stream is formed through data synchronization techniques (such as interpolation or alignment algorithms) to provide a foundation for subsequent processing.

[0028] For example, in the electronic fence system of Factory A, cameras B are deployed in the factory area passageways to monitor logistics vehicles, infrared sensors C are installed in the hazardous materials storage area to detect abnormal temperatures of hazardous materials, and audio sensors D are arranged around the production workshop to collect the sound of equipment operation. The system first acquires synchronous data in real time from sensors B, C, and D: camera B captures visible light video of vehicle passage, sensor C generates thermal infrared images of the storage area, and sensor D records the raw audio stream of the workshop equipment. Then, the raw audio stream is analyzed to separate the periodic beat signals of normal equipment operation (such as regular noise from machine tools) and extract abnormal sound event segments (such as sudden metallic impact sounds). The start time, end time, and sound energy change pattern of the abnormal sound event segment are analyzed (such as energy increase followed by decay) to form audio feature segment data. Finally, the visible light video, thermal infrared images, and audio feature segments are aligned by time and integrated into a multimodal data stream for subsequent analysis.

[0029] This step achieves multi-dimensional synchronous monitoring of the factory area environment through heterogeneous sensor network acquisition and multimodal data integration, providing a rich and consistent perception foundation for intrusion identification, effectively improving the comprehensiveness and timeliness of data sources, and avoiding information loss caused by the limitations of a single sensor.

[0030] Step 102: Generate a differentiated perception focusing strategy based on the real-time production activity status of the production workshop, the type and attributes of the hazardous materials stored in the hazardous materials storage area, and the real-time vehicle scheduling information of the factory area passage.

[0031] Optionally, step 102 may specifically include the following steps: Step 1021: Receive real-time production activity status information from the production workshop; Step 1022: Define a first perception priority for the audio feature segment data based on the real-time production activity status information; Step 1023: Receive the type attribute information of the hazardous materials stored in the hazardous materials storage area, wherein the type attribute information of the hazardous materials includes the volatility and heat sensitivity of the hazardous materials; Step 1024: Set the monitoring granularity for the thermal infrared image data based on the type attribute information; Step 1025: Receive real-time vehicle dispatch information for the factory area access road, the real-time vehicle dispatch information including the location and passage time window of the planned transport vehicles; Step 1026: Delineate the dynamic sensing area range for the visible light video data based on the real-time vehicle dispatch information; Step 1027: Combine the first perception priority, the monitoring granularity, and the perception area range to generate a differentiated perception focusing strategy.

[0032] In the above scheme, real-time production activity status information refers to data obtained from the factory production management system that reflects the current operating status of the production workshop, and is used to indicate whether the equipment is in active production or in a silent standby state. The type attributes of hazardous materials describe the hazardous characteristics of the items stored in a hazardous materials storage area. They mainly include the volatility (the ease with which a substance volatilizes) and thermal sensitivity (the sensitivity of a substance to temperature changes). Real-time vehicle dispatching information is obtained from the logistics system, showing the real-time location of planned transport vehicles within the factory area and the pre-arranged passage time windows. The differentiated perception focus strategy is a data processing guideline developed to address dynamic risks in different areas of the factory. Its core includes defining the first perception priority for audio feature segment data (indicating the level of attention to audio data), setting the monitoring granularity for thermal infrared image data (referring to the level of detail in analyzing thermal image data), and delineating the dynamic perception area range for visible light video data (the key video monitoring area that is dynamically adjusted according to vehicle scheduling).

[0033] In this scheme, firstly, real-time production activity status information from the production workshop is received from the factory production management system. This real-time activity status information directly reflects the current working status of the workshop. Secondly, based on the received real-time production activity status information, a first perception priority is defined for audio feature segment data from that workshop. For example, when the workshop is in a silent standby state, any abnormal sound is more suspicious and therefore given a higher priority. Then, the type and attribute information of hazardous materials stored in the hazardous materials storage area is received from the warehouse management system, especially their volatility and heat sensitivity parameters. Next, based on the received type and attribute information, the monitoring granularity is set for the thermal infrared image data. For example, for hazardous materials that are highly sensitive to heat, a finer monitoring granularity is needed to capture subtle temperature changes. Simultaneously, real-time vehicle dispatch information for factory access routes is received from the logistics dispatch system, including vehicle location and passage time windows. Subsequently, based on this real-time vehicle dispatch information, a dynamic sensing area is defined for the visible light video data covering the factory access routes; for example, areas outside the planned vehicle passage times and routes are designated as key areas of focus. Finally, the three types of parameters—first sensing priority, monitoring granularity, and dynamic sensing area—are combined into a structured, differentiated sensing focusing strategy that can guide subsequent data fusion processing.

[0034] Following the specific implementation of the previous solution, in Factory A, the production management system reports that production workshop E is currently in a "silent standby" state. The warehouse management system indicates that the chemical F stored in the hazardous materials storage area is highly heat-sensitive. The logistics system shows that a planned vehicle G will pass through the factory area in the next half hour. The system first receives the "silent standby" status information of workshop E and defines the audio data collected from this workshop as having a high first-priority perception. At the same time, based on the high heat sensitivity of chemical F, the thermal infrared image data is set to high monitoring granularity (i.e., very detailed analysis of temperature changes is required). Next, based on the passage time window and location of vehicle G, the dynamic perception area range is delineated for the visible light video data, focusing the monitoring on the time and area of ​​non-planned vehicle passage. Finally, these parameters are combined into a complete differentiated perception focusing strategy to guide the system to focus on abnormal sounds in the silent workshop, meticulously scan the temperature of the hazardous materials storage area, and monitor the video dynamics of unplanned passage areas in subsequent processing.

[0035] This step generates a differentiated perception focus strategy closely related to the current actual risk situation by integrating real-time dynamic information from multiple operating systems in the factory. This enables the monitoring system to intelligently adjust the focus of analysis on different modalities and different areas, thereby improving the accuracy of response to real threats and the efficiency of resource utilization, and effectively avoiding misjudgment of normal activities in complex industrial environments.

[0036] Step 103: According to the differentiated perception focusing strategy, the synchronous multimodal data is weighted and focused to generate dynamic intrusion perception features.

[0037] Optionally, step 103 may specifically include the following steps: Step 1031: Assign a primary attention value to the audio feature segment data according to the first perception priority; Step 1032: Based on the monitoring granularity, determine the level of detail required for processing the thermal infrared image data, and perform detailed adjustments on the thermal infrared image data to obtain clear thermal infrared image data; Step 1032 may specifically include the following steps: Based on the determined level of detail, the thermal infrared image data is divided into multiple continuous temperature ranges; based on the type and attribute information of the hazardous material, a critical temperature range directly related to the potential risk is determined from the divided temperature ranges; the thermal infrared image data is processed to enhance the image detail within the critical temperature ranges while weakening the image detail within non-critical temperature ranges, resulting in enhanced image data; in the enhanced image data, all connected regions formed by pixels within the critical temperature ranges are identified to generate clear thermal infrared image data.

[0038] Step 1033: Based on the defined dynamic sensing area range, filter out image sequence content that matches the sensing area range from the visible light video data; Step 1034: The audio feature segment data assigned the primary concern value, the clear thermal infrared image data, and the selected visible light video image sequence content are integrated to generate dynamic intrusion detection features.

[0039] In the above scheme, dynamic intrusion detection features are a comprehensive feature representation formed by applying differential focusing processing to multimodal data; The audio feature segment data with the primary attention value refers to the audio data segment that has been assigned a weight value based on its first perceptual priority. This primary attention value directly affects the influence of the audio data segment in subsequent fusion analysis. The level of detail is determined by the granularity of the monitoring and the level of detail in processing the thermal infrared image data. Clear thermal infrared image data is thermal image data obtained through meticulous adjustments; Multiple consecutive temperature ranges refer to several small, continuous temperature zones into which the entire temperature range is divided; Potential risks refer to situations that may lead to safety accidents; Critical temperature range refers to specific temperature zones that require close monitoring. Image detail enhancement refers to improving image sharpness and contrast; Non-critical temperature ranges refer to temperature zones of lower importance in terms of image detail. Enhanced image data refers to images that have undergone contrast optimization processing; The connected region formed by pixels refers to spatially adjacent pixel blocks with similar temperatures, which ultimately generate thermal image data; The dynamic sensing area is a key monitoring geographical area that changes over time and is defined based on real-time vehicle dispatch information for visible light video data. Visible light video image sequence content refers to the image sequence content selected from visible light video data that matches the range of the dynamic sensing area.

[0040] In this scheme, firstly, based on the first perception priority defined in the differentiated perception focusing strategy for audio feature segment data, a specific primary attention value is assigned to each audio feature segment data; for example, higher priority corresponds to a higher value. Secondly, based on the monitoring granularity set in the differentiated perception focusing strategy, the level of detail for processing the thermal infrared image data is determined. Based on the determined level of detail, the entire temperature range represented by the original thermal infrared image data is divided into multiple continuous temperature intervals. Next, based on the hazardous material type attribute information obtained from the warehouse management system, key temperature intervals directly related to potential risks (such as the ignition point or leakage indication temperature of a specific hazardous material) are selected from these temperature intervals. Then, the original thermal infrared image data is processed to significantly enhance the image detail performance, such as contrast or brightness, of pixels in these key temperature intervals, while reducing the detail performance of pixels in non-key temperature intervals. An enhanced image is obtained. Then, within this enhanced image, regions (i.e., connected regions) where all pixel values ​​fall within the critical temperature range and are adjacent to each other are identified using methods such as region growing or contour finding, generating the final clear thermal infrared image data. Simultaneously, based on the dynamic sensing region defined in the differentiated sensing focusing strategy, the original visible light video data is spatially filtered, retaining only image frames whose content falls within this dynamic sensing region, forming the filtered visible light video image sequence. Finally, the obtained audio feature segment data with primary attention values, the clear thermal infrared image data, and the visible light video image sequence are aligned and integrated according to their timestamps. For example, the audio weights, the enhanced region information of the thermal image, and the specific region of the video are merged into a unified data structure, thereby generating dynamic intrusion sensing features for subsequent pattern analysis.

[0041] Following the specific implementation of the previous solution, based on the differentiated sensing focus strategy already generated at Factory A, the system begins processing the acquired multimodal data stream. For audio feature segment data (such as a sudden knocking sound segment) from production workshop E in a "silent standby" state, a higher primary concern value (e.g., 0.9) is assigned based on its high first sensing priority. For the thermal infrared image of the hazardous materials storage area, based on the high monitoring granularity set for the highly heat-sensitive chemical F, the image temperature range is finely divided into multiple narrow temperature intervals, and the area near the flash point of chemical F is identified as the critical temperature zone. The images are then processed to enhance details within the critical temperature range (making object outlines clearer within this temperature range) and weaken details in other ranges. Finally, all connected image regions within the critical temperature range are identified to obtain clear thermal infrared image data. For visible light videos of factory passageways, image sequences displaying the key monitoring area are selected from the original videos based on the dynamic sensing area range (excluding areas along the planned vehicle G's passage path). Finally, the weighted audio data, the enhanced thermal images of specific temperature areas, and the selected video content are integrated to generate a unified dynamic intrusion sensing feature.

[0042] This step employs a differentiated perception focusing strategy to perform targeted weighting and focusing on the raw multimodal data, generating a dynamic intrusion perception feature that highlights current high-risk information. It effectively integrates key clues from different sensors, providing a high-quality, highly relevant analytical foundation for accurately identifying intrusion behavior and improving the system's perception intelligence and analysis efficiency in complex industrial environments.

[0043] Step 104: Based on the dynamic intrusion perception features, perform pattern analysis using an artificial intelligence model to identify intrusion behaviors that conform to the preset factory safety procedures.

[0044] Optionally, step 104 may specifically include the following steps: Step 1041: Input the dynamic intrusion perception features into a pre-trained artificial intelligence model, and parse the dynamic intrusion perception features into a moving target sequence corresponding to visible light video data, an abnormal heat source distribution corresponding to thermal infrared image data, and a sound event label corresponding to audio feature segment data. Step 1042: Analyze the moving target sequence, determine whether the trajectory of the moving target crosses the electronic fence boundary defined by the factory safety regulations, and determine the speed change pattern of the moving target; Step 1043: Analyze the distribution of the abnormal heat source, determine whether the heat source is located within the restricted area of ​​the hazardous materials storage area, and distinguish whether the heat source is stationary or moving. Step 1044: Combine the analysis results of the moving target sequence, the analysis results of the abnormal heat source distribution, and the sound event tags to generate multi-dimensional judgment criteria; Step 1045: Match the multi-dimensional judgment criteria with multiple intrusion behavior rules predefined in the factory safety procedures to obtain the matching results; Step 1045 may specifically include the following steps: Establish a correspondence between each intrusion behavior rule in the factory safety procedures and the factory area. The first type of rule corresponds to the production workshop area, the second type to the hazardous materials storage area, and the third type to the factory access road. Based on the spatial location information included in the multi-dimensional judgment criteria, map the analysis results of the moving target sequence, the analysis results of the abnormal heat source distribution, and the sound event tags to the corresponding physical areas. Perform correlation analysis on the multi-dimensional judgment criteria mapped to the same physical area to determine the chronological order and spatial relative relationship of the moving target, abnormal heat source, and sound event. Compare the multi-dimensional judgment criteria after correlation analysis with the intrusion behavior rules of the corresponding physical areas one by one, checking whether the trajectory characteristics of the moving target, the state characteristics of the heat source, and the type characteristics of the sound event simultaneously satisfy the condition combination defined by the rule. When all feature conditions meet the condition combination defined by the rule within the preset time and spatial tolerance ranges, a matching result indicating a successful match is generated.

[0045] Step 1046: When the matching result shows that the multi-dimensional judgment criteria satisfy all the conditions of any one of the multiple intrusion behavior rules in terms of time sequence and spatial location, it is judged as an intrusion behavior.

[0046] In the above plan, the pre-established factory safety regulations are documents that are formulated in advance and clearly define various safety standards and violations. Intrusion behavior is defined as a specific intrusion action or event that violates procedures; Artificial intelligence models are trained computer programs that analyze input data and output recognition results; Moving target sequence is the position and trajectory information of moving objects at consecutive points in time, identified from video data; Abnormal heat source distribution refers to the location and shape of temperature concentration areas in thermal infrared images that differ from the normal state. Sound event tags are category identifiers obtained after classifying audio feature segments; The electronic fence boundary is a virtual warning line drawn on the factory area map in the regulations; The speed change pattern describes the pattern of how the speed of a moving target changes; The restricted area refers to the specific areas within the hazardous materials storage area where unauthorized entry is strictly prohibited; The heat source can be in a stationary state (position unchanged) or a moving state (position changed); The multi-dimensional judgment criteria are based on the overall judgment formed by comprehensively analyzing the moving target, heat source, and sound. The multiple rules for intrusion behavior are a series of conditions that must be met to determine different intrusion behaviors, which are detailed in the regulations. The matching result is the conclusion after the rule matching operation; The rules are divided into three categories according to the area: the first category of rules corresponds to the production workshop area, the second category of rules corresponds to the hazardous materials storage area, and the third category of rules corresponds to the factory area passageway. The corresponding physical area is the actual geographical location within the factory; Sequence refers to the order in which events occur on a timeline; Spatial relative relationship refers to the orientation and distance relationship between different targets or events; The multi-dimensional judgment criteria after correlation analysis are comprehensive information processed through spatiotemporal correlation analysis; Simultaneously satisfying the combination of conditions defined in the rule requires that all specified conditions be true; The preset time tolerance and spatial tolerance ranges are the allowable time error and spatial distance error.

[0047] In this scheme, firstly, the generated dynamic intrusion perception features are input into a pre-trained artificial intelligence model. Based on the patterns it has learned, the AI ​​model decomposes and restores this comprehensive feature into more interpretable intermediate results: namely, parsing the moving target sequence (including target location and trajectory) from the video components, parsing the abnormal heat source distribution (including heat zone location and shape) from the thermal infrared components, and parsing sound event tags (such as "impact sound" and "alarm sound") from the audio components. Secondly, the moving target sequence is specifically analyzed. A trajectory tracking algorithm is used to determine whether the movement path of any moving target crosses the electronic fence boundary defined in the factory safety regulations, and the speed change pattern of the moving target throughout the sequence is analyzed (e.g., whether it is accelerated). (Speed, deceleration, or constant speed); then, specifically analyze the distribution of abnormal heat sources, using image analysis to determine if a heat source appears within the restricted area of ​​the hazardous materials storage area, and further distinguish whether the abnormal heat source is stationary or moving based on the changes in the position of the heat source between consecutive frames; next, combine the three types of information—the moving target sequence analysis results (e.g., "Target A crosses the boundary and accelerates"), the abnormal heat source distribution analysis results (e.g., "Heat source B appears in the restricted area and moves"), and the sound event tags (e.g., "Tag: Glass breaking sound")—to form a unified multi-dimensional judgment criterion; subsequently, match this multi-dimensional judgment criterion with multiple predefined intrusion behavior rules in the factory safety procedures; first, establish each rule in the procedures... The rules are mapped to specific factory areas (production workshops, hazardous materials storage areas, factory passageways), clarifying which areas each of the first, second, and third categories of rules governs. Then, based on the spatial location information inherent in the multi-dimensional judgment criteria, moving targets, abnormal heat sources, and sound events are mapped to their actual physical locations. Next, a correlation analysis is performed on this information mapped to the same physical area to determine their temporal order (e.g., whether the sound occurred first or the target appeared first) and their spatial relationship (e.g., whether the heat source and the moving target are close). Finally, the multi-dimensional judgment criteria obtained after this correlation analysis and bound to a specific area are matched with the intrusion behavior rule corresponding to that specific area. Each condition is compared to the trajectory characteristics of the moving target, the state characteristics of the heat source, and the type characteristics of the sound event to check whether they simultaneously meet all the conditions defined by the intrusion behavior rule. During the comparison, matching is allowed within a preset time tolerance (e.g., events within a few seconds are considered simultaneous) and spatial tolerance range (e.g., the target within a few meters of the boundary is considered crossing). When all feature conditions meet the rule within the tolerance range, a matching result indicating a successful match is generated. Finally, the matching result is checked. When the matching result shows that the multi-dimensional judgment criteria jointly meet all the conditions specified by one of the multiple intrusion behavior rules in terms of time sequence and spatial location, the system finally determines that an intrusion behavior conforming to the factory safety procedure definition has occurred.

[0048] Following the specific implementation of the previous solution, the system inputs the generated dynamic intrusion perception features into the AI ​​model. The AI ​​model interprets these features as follows: identifying a moving target sequence in the factory passage video (showing an unmanned vehicle H moving towards the hazardous materials storage area); identifying an abnormal heat source distribution in the thermal infrared image of the hazardous materials storage area (a new heat source near the chemical storage tank F); and parsing sound event tags ("metal scraping sound") from the audio of the production workshop E. Analyzing the moving target sequence, it is determined that the trajectory of vehicle H crosses the "electronic fence boundary" and its speed suddenly increases. Analyzing the abnormal heat source distribution confirms that the heat source is within the "restricted area" and is in a "moving state." These analysis results are then combined with the sound tag group. The system synthesizes multi-dimensional judgment criteria; then performs intrusion behavior rule matching: establishing the relationship between rules and areas (e.g., rule K corresponds to the hazardous materials storage area); mapping moving targets, heat sources, and sound events to corresponding areas based on their locations; performing correlation analysis on the information mapped to the hazardous materials storage area, finding that the moving target (vehicle H) and the moving heat source appear sequentially in time and are close in space; comparing the multi-dimensional judgment criteria after this correlation analysis with the conditions of rule K (rule K requires: there is a moving heat source in the restricted area and an unauthorized moving target); since all features meet the conditions within the allowable time and space tolerances, a successful matching result is generated; the system finally determines that an intrusion behavior conforming to rule K has occurred.

[0049] This step uses an artificial intelligence model to perform in-depth analysis and multi-dimensional correlation analysis on the fused dynamic intrusion perception features, and performs precise matching based on preset security rules. Ultimately, it achieves accurate and automatic identification of intrusion behavior in complex industrial environments, improves the intelligence level and decision reliability of the security system, and effectively ensures safe production in the factory.

[0050] Figure 2 This application provides a structural schematic diagram of an AI-based multimodal electronic fence restricted area intrusion detection system, as shown below. Figure 2 As shown, the system includes: The acquisition module 21 is used to acquire synchronous multimodal data collected in real time by a heterogeneous sensor network pre-deployed in the factory area. The factory area includes factory passages, hazardous materials storage areas and production workshops. The synchronous multimodal data includes visible light video data, thermal infrared image data and audio feature segment data. The generation module 22 is used to generate a differentiated perception focusing strategy based on the real-time production activity status of the production workshop, the type and attributes of the hazardous materials stored in the hazardous materials storage area, and the real-time vehicle scheduling information of the factory area passage. The fusion module 23 is used to perform weighted focusing fusion on the synchronous multimodal data according to the differentiated perception focusing strategy to generate dynamic intrusion perception features; The identification module 24 is used to identify intrusion behaviors that conform to the preset factory safety procedures by performing pattern analysis through an artificial intelligence model based on the dynamic intrusion perception characteristics.

[0051] Figure 2 The aforementioned AI-based multimodal electronic fence restricted area intrusion detection system can perform... Figure 1 The implementation principle and technical effects of the AI-based multimodal electronic fence restricted area intrusion identification method described in the illustrated embodiment will not be repeated here. The specific methods by which each module and unit of the AI-based multimodal electronic fence restricted area intrusion identification system in the above embodiments are performed have been described in detail in the embodiments related to this method, and will not be elaborated upon here.

[0052] In one possible design, Figure 2 The AI-based multimodal electronic fence restricted area intrusion detection system shown in the embodiment can be implemented as a computing device, such as... Figure 3 As shown, the computing device may include a storage component 31 and a processing component 32; The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are invoked and executed by the processing component 32.

[0053] The processing component 32 is used for the above Figure 1 The embodiment describes an AI-based multimodal electronic fence restricted area intrusion identification method.

[0054] The processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above-described method. Alternatively, the processing component may be implemented as one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above-described method.

[0055] Storage component 31 is configured to store various types of data to support operations at the terminal. The storage component can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0056] Of course, computing devices may also include other components, such as input / output interfaces, display components, communication components, etc.

[0057] Input / output interfaces provide interfaces between processing components and peripheral interface modules, which can be output devices, input devices, etc.

[0058] The communication components are configured to facilitate wired or wireless communication between computing devices and other devices.

[0059] The computing device can be a physical device or an elastic computing host provided by a cloud computing platform. In this case, the computing device can refer to a cloud server, and the aforementioned processing components, storage components, etc., can be basic server resources rented or purchased from the cloud computing platform.

[0060] This application also provides a computer storage medium storing a computer program, which, when executed by a computer, can perform the above-described functions. Figure 1 The embodiment shown illustrates an AI-based multimodal electronic fence restricted area intrusion identification method.

[0061] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0062] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0063] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0064] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A multimodal electronic fence restricted area intrusion identification method based on AI, characterized in that, include: Acquire synchronous multimodal data in real time from a heterogeneous sensor network pre-deployed in a factory area, wherein the factory area includes factory passages, hazardous materials storage areas and production workshops, and the synchronous multimodal data includes visible light video data, thermal infrared image data and audio feature segment data; Based on the real-time production activity status of the production workshop, the type and attributes of the hazardous materials stored in the hazardous materials storage area, and the real-time vehicle scheduling information of the factory area passage, a differentiated perception focusing strategy is generated. Based on the differentiated perception focusing strategy, the synchronous multimodal data is weighted and focused to generate dynamic intrusion perception features; Based on the dynamic intrusion perception characteristics, pattern analysis is performed using an artificial intelligence model to identify intrusion behaviors that conform to the preset factory safety procedures.

2. The method according to claim 1, characterized in that, Acquire synchronous multimodal data in real time from a heterogeneous sensor network pre-deployed in the factory area. This synchronous multimodal data includes visible light video data, thermal infrared image data, and audio feature segment data, including: Visible light video data containing moving objects is acquired from cameras pre-positioned in the factory area passageways; thermal infrared image data characterizing the thermal radiation distribution of objects is acquired from infrared sensors facing the hazardous materials storage area; and raw audio streams containing equipment operating sounds are collected from audio sensors outside the production workshop. The original audio stream is parsed to separate the periodic beat signal corresponding to the operating cycle of the production equipment in the production workshop, and abnormal sound event segments that are different from the periodic beat signal are extracted. The abnormal sound event segments are analyzed to determine the start time point, end time point, and sound energy change pattern within the start and end time points of each abnormal sound event segment, so as to form audio feature segment data. The visible light video data, the thermal infrared image data, and the audio feature segment data are integrated to form a multimodal data stream.

3. The method according to claim 1, characterized in that, Based on the real-time production activity status of the production workshop, the type and attributes of the hazardous materials stored in the hazardous materials storage area, and the real-time vehicle scheduling information of the factory area access roads, a differentiated perception focusing strategy is generated, including: Receive real-time production activity status information from the production workshop; A first perception priority is defined for the audio feature segment data based on the real-time production activity status information. Receive type and attribute information of hazardous materials stored in the hazardous materials storage area, including the volatility and heat sensitivity of the hazardous materials; The monitoring granularity is set for the thermal infrared image data based on the type attribute information; Receive real-time vehicle dispatch information for the factory area access road, the real-time vehicle dispatch information including the location and passage time window of the planned transport vehicles; Based on the real-time vehicle dispatch information, a dynamic sensing area is defined for the visible light video data. The first sensing priority, the monitoring granularity, and the sensing area range are combined to generate a differentiated sensing focusing strategy.

4. The method according to claim 1, characterized in that, Based on the differentiated perception focusing strategy, the synchronous multimodal data is weighted and focused fused to generate dynamic intrusion perception features, including: Based on the first perception priority, assign a primary attention value to the audio feature segment data; Based on the monitoring granularity, the level of detail required for processing the thermal infrared image data is determined, and the thermal infrared image data is finely adjusted to obtain clear thermal infrared image data. Based on the defined dynamic sensing area, image sequence content that matches the sensing area is selected from the visible light video data; The audio feature segment data assigned the primary concern value, the clear thermal infrared image data, and the selected visible light video image sequence content are integrated to generate dynamic intrusion detection features.

5. The method according to claim 4, characterized in that, Based on the monitoring granularity, the level of detail required for processing the thermal infrared image data is determined, and the thermal infrared image data is finely adjusted to obtain clear thermal infrared image data, including: Based on the determined level of detail, the aforementioned thermal infrared image data is divided into multiple continuous temperature ranges; Based on the type and attribute information of the hazardous materials, the key temperature ranges directly related to potential risks are determined from the multiple temperature ranges that have been divided. The thermal infrared image data is processed to enhance the image detail within the critical temperature range, while weakening the image detail within the non-critical temperature range, resulting in enhanced image data. In the enhanced image data, the connected regions formed by all pixels within the critical temperature range are identified to generate clear thermal infrared image data.

6. The method according to claim 1, characterized in that, Based on the dynamic intrusion detection characteristics, pattern analysis is performed using an artificial intelligence model to identify intrusion behaviors that conform to the preset factory safety procedures, including: The dynamic intrusion perception features are input into a pre-trained artificial intelligence model, which parses the dynamic intrusion perception features into a moving target sequence corresponding to visible light video data, an abnormal heat source distribution corresponding to thermal infrared image data, and a sound event label corresponding to audio feature segment data. Analyze the moving target sequence to determine whether the trajectory of the moving target crosses the electronic fence boundary defined by the factory safety regulations, and determine the speed change pattern of the moving target; Analyze the distribution of the abnormal heat sources to determine whether the heat sources are located within the restricted area of ​​the hazardous materials storage area, and distinguish whether the heat sources are stationary or moving. The analysis results of the moving target sequence, the analysis results of the abnormal heat source distribution, and the sound event tags are combined to generate multi-dimensional judgment criteria. The multi-dimensional judgment criteria are matched with multiple intrusion behavior rules predefined in the factory safety procedures to obtain the matching results; When the matching results show that the multi-dimensional judgment criteria satisfy all the conditions of any one of the multiple intrusion behavior rules in terms of time sequence and spatial location, it is judged as an intrusion behavior.

7. The method according to claim 6, characterized in that, The multi-dimensional judgment criteria are matched with multiple predefined intrusion behavior rules in the factory safety procedures to obtain matching results, including: Establish the correspondence between each intrusion behavior rule in the factory safety regulations and the factory area, wherein the first type of rule corresponds to the production workshop area, the second type of rule corresponds to the hazardous materials storage area, and the third type of rule corresponds to the factory area passageway; Based on the spatial location information contained in the multi-dimensional judgment criteria, the analysis results of the moving target sequence, the analysis results of the abnormal heat source distribution, and the sound event tags are mapped to the corresponding physical regions. A correlation analysis is conducted on the multi-dimensional judgment criteria mapped to the same physical region to determine the chronological order and spatial relative relationship of moving targets, abnormal heat sources, and sound events on the time axis. The multi-dimensional judgment criteria after correlation analysis are compared with the intrusion behavior rules of the corresponding physical area one by one to check whether the trajectory characteristics of the moving target, the state characteristics of the heat source, and the type characteristics of the sound event simultaneously meet the condition combination defined by the rule. When all feature conditions meet the condition combination defined by the satisfying rule within the preset time tolerance and spatial tolerance range, a matching result indicating a successful match is generated.

8. An AI-based multimodal electronic fence restricted area intrusion identification system, characterized in that, include: The acquisition module is used to acquire synchronous multimodal data collected in real time by a heterogeneous sensor network pre-deployed in the factory area. The factory area includes factory passages, hazardous materials storage areas, and production workshops. The synchronous multimodal data includes visible light video data, thermal infrared image data, and audio feature segment data. The generation module is used to generate a differentiated perception focusing strategy based on the real-time production activity status of the production workshop, the type and attributes of the hazardous materials stored in the hazardous materials storage area, and the real-time vehicle scheduling information of the factory area passage. The fusion module is used to perform weighted focusing fusion on the synchronous multimodal data according to the differentiated perception focusing strategy to generate dynamic intrusion perception features; The identification module is used to identify intrusion behaviors that conform to the preset factory safety procedures by performing pattern analysis through an artificial intelligence model based on the dynamic intrusion perception characteristics.

9. A computing device, characterized in that, It includes a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are invoked and executed by the processing component to implement the AI-based multimodal electronic fence restricted area intrusion identification method as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that, The device contains a computer program that, when executed by a computer, implements an AI-based multimodal electronic fence restricted area intrusion identification method as described in any one of claims 1 to 7.

Images