Environmental protection risk ai visual early warning terminal

Abstract

The application discloses an environmental protection risk AI visual early warning terminal and belongs to the technical field of environmental protection monitoring. The terminal comprises an adaptive multi-modal data acquisition module, an edge intelligent processing module and a terminal control system. The adaptive multi-modal data acquisition module collects wide-area environmental image data through a wide-area field mode of a dynamic zoom and wide-area field switching lens. After the terminal control system analyzes the wide-area environmental image data to identify a suspected risk source area, the terminal control system controls the lens to switch to a dynamic zoom mode to collect high-definition detail image data. The collected multi-modal data are input into the edge intelligent processing module for parallel processing to generate preliminary results. Then, a lightweight real-time learning model and an encryption algorithm are used to generate encrypted risk early warning data. The terminal control system decrypts and outputs non-sensitive information, transmits sensitive encrypted data according to permissions to complete early warning feedback, and sends parameter adjustment signals to the acquisition module according to the feedback results. The terminal improves the accuracy and real-time performance of environmental protection risk monitoring.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of environmental protection monitoring, and particularly relates to an environmental protection risk AI visual early warning terminal. BACKGROUND

[0002] The existing environmental protection monitoring equipment is mostly fixedly deployed or manually adjusted, and has problems of limited field of view range, single data collection, dependence on remote computing power, weak adaptability to complex environment, etc., and cannot meet the requirements of rapid and accurate environmental protection risk early warning in scenes such as wild fields and remote areas, and the safety of risk analysis results and the pertinence of early warning feedback are insufficient, and the real-time, intelligent and safety requirements of modern environmental protection monitoring cannot be met. SUMMARY

[0003] In view of the deficiencies of the prior art, the environmental protection risk AI visual early warning terminal is provided, which comprises an adaptive multi-modal data acquisition module, an edge intelligent processing module and a terminal control system; the adaptive multi-modal data acquisition module acquires wide-area environmental image data through the wide-area field of view mode of the dynamic zoom and wide-area field of view switching lens, the terminal control system analyzes the wide-area environmental image data to identify the suspected risk source area, and then controls the lens to switch to the dynamic zoom mode to acquire high-definition detail image data, and inputs the synchronous multi-modal data into the edge intelligent processing module for parallel processing to generate preliminary results, and then generates encrypted risk early warning data through a lightweight real-time learning model and an encryption algorithm; the terminal control system decrypts and outputs non-sensitive information, transmits sensitive encrypted data according to the permission to complete the early warning feedback, and sends a parameter adjustment signal to the adaptive multi-modal data acquisition module according to the feedback result; the terminal improves the accuracy and real-time performance of environmental protection risk monitoring.

[0004] To achieve the above object, the application provides the following technical scheme:

[0005] The environmental protection risk AI visual early warning terminal comprises an adaptive multi-modal data acquisition module, an edge intelligent processing module and a terminal control system.

[0006] The wide-area environmental image data is acquired through the wide-area field of view mode of the dynamic zoom and wide-area field of view switching lens in the adaptive multi-modal data acquisition module and is output to the terminal control system, the terminal control system preliminarily analyzes the wide-area environmental image data, identifies the suspected risk source area, and sends a control signal to the dynamic zoom and wide-area field of view switching lens to switch to the dynamic zoom mode and acquire high-definition detail image data, and inputs the multi-modal data of the monitoring area acquired synchronously into the edge intelligent processing module;

[0007] The edge intelligent processing module distributes the received high-definition detail image data and multi-modal data to CPUs, GPUs and NPUs through an internal parallel scheduling unit for parallel processing to generate preliminary results, and then calls an internal lightweight real-time learning model and an internal encryption algorithm to analyze the preliminary results and generate encrypted risk warning data.

[0008] The terminal control system receives the encrypted risk warning data, decrypts and outputs non-sensitive warning information, and transmits sensitive encrypted data to a designated terminal according to a preset permission management rule to complete the warning feedback. The terminal control system sends a parameter adjustment signal to the adaptive multi-modal data acquisition module according to the warning feedback result.

[0009] Specifically, when the wide-field mode of the dynamic zoom and wide-field switching lens in the adaptive multi-modal data acquisition module acquires wide-field environmental image data, it includes:

[0010] The ambient light sensing unit in the adaptive multi-modal data acquisition module detects the light intensity data of the monitoring area at a first resolution and transmits the light intensity data to the terminal control system. The terminal control system determines whether the current lighting conditions meet the image acquisition requirements of the dynamic zoom and wide-field switching lens based on the received light intensity data, and sends an exposure parameter adjustment signal to the adaptive multi-modal data acquisition module.

[0011] The adaptive multi-modal data acquisition module adjusts the exposure time of the dynamic zoom and wide-field switching lens according to the received exposure parameter adjustment signal. The dynamic zoom and wide-field switching lens is an integrated motor-driven optical lens that can be electrically switched between wide-angle fisheye mode and long-focus zoom mode. The wide-angle fisheye mode corresponds to the wide-field mode, and the long-focus zoom mode corresponds to the dynamic zoom mode.

[0012] After the exposure time is adjusted, the adaptive multi-modal data acquisition module controls the dynamic zoom and wide-field switching lens to switch to the wide-field mode and acquires wide-field environmental image data of the monitoring area through the wide-field mode.

[0013] Specifically, after the adaptive multi-modal data acquisition module receives the exposure parameter adjustment signal sent by the terminal control system, it also includes:

[0014] Analyzing the adjustment parameters in the exposure parameter adjustment signal;

[0015] If the exposure parameter adjustment signal also carries a monitoring frequency adjustment parameter, the acquisition interval time of the dynamic zoom and wide-field switching lens and the acquisition cycle of the multi-modal sensor are adjusted according to the monitoring frequency adjustment parameter.

[0016] If the exposure parameter adjustment signal also carries the field of view adjustment parameter, then the wide field of view angle of the dynamic zoom and wide field of view switching lens in wide field of view mode is adjusted according to the field of view adjustment parameter.

[0017] After the parameters are adjusted, the adaptive multimodal data acquisition module sends a confirmation signal to the terminal control system confirming that the parameter adjustment is complete.

[0018] Specifically, when the terminal control system performs preliminary analysis of wide-area environmental image data, it includes:

[0019] The terminal control system receives wide-area environmental image data transmitted by the adaptive multimodal data acquisition module, and performs noise reduction processing on the wide-area environmental image data by calling the built-in image noise reduction algorithm to obtain the noise-reduced wide-area environmental image data.

[0020] The built-in image segmentation algorithm divides the denoised wide-area environmental image data into multiple sub-region image data, with each sub-region image data corresponding to a local area of ​​the monitoring area.

[0021] The terminal control system retrieves a preset stored normal environment image template, compares the image data of each sub-region obtained by segmentation with the corresponding sub-region template in the preset normal environment image template, and calculates the difference value of each sub-region image data and sub-region template in color, texture and contour features; the normal environment image template is the standard image data of the corresponding monitoring area under the state of no environmental risk, and contains the normal visual features of each sub-region.

[0022] The terminal control system compares the difference value of the image data of each sub-region with a preset threshold. If the difference value of the image data of any sub-region exceeds the preset threshold, it determines that there is an anomaly in the corresponding sub-region and marks the corresponding monitoring area as a suspected risk source area. The preset threshold is set according to historical monitoring data and is used to determine whether there is a critical value for anomalies.

[0023] Specifically, sending a control signal to the dynamic zoom and wide field-of-view switching lens to switch it to dynamic zoom mode and acquire high-definition detail image data includes:

[0024] After completing the preliminary analysis of the wide-area environmental image data, the terminal control system acquires the coordinate information of the marked suspected risk source areas in the wide-area environmental image data. It then converts the coordinate information into the lens's focus position parameters based on the optical parameters of the dynamic zoom and wide-field-of-view switching lens. Simultaneously, it combines the size information of the suspected risk source areas in the wide-area environmental image data to determine the lens zoom magnification parameters covering the suspected risk source areas. Finally, it integrates the focus position parameters and the lens zoom magnification parameters into a control signal and sends it to the dynamic zoom and wide-field-of-view switching lens. The optical parameters include the lens focal length and the corresponding field of view.

[0025] After receiving the control signal, the dynamic zoom and wide field of view switching lens analyzes the focus position parameter in the control signal, drives the focus motor in the lens to adjust the lens position to the corresponding focus position, analyzes the zoom magnification parameter, drives the zoom motor to adjust the lens spacing to the specified zoom magnification, and completes the switching from wide field of view mode to dynamic zoom mode.

[0026] The dynamic zoom and wide field-of-view switching lens activates the built-in image sensor to acquire high-definition detailed image data of the suspected risk source area at a second resolution; the second resolution is higher than the first resolution.

[0027] Specifically, the process of generating the preliminary results includes:

[0028] The edge intelligent processing module receives high-definition detailed image data and multimodal data transmitted from the adaptive multimodal data acquisition module, performs preprocessing, and encapsulates them into data transmission packets;

[0029] The parallel scheduling unit parses the data transmission packet, separates the high-definition detail image data after format conversion and the normalized multimodal data, and calls the preset task allocation rules to allocate the feature extraction task in the high-definition detail image data to the GPU, the numerical calculation task in the multimodal data to the CPU, and the model inference task for environmental risk identification to the NPU, and generates task allocation instructions and sends them to the corresponding processors.

[0030] After receiving the feature extraction task, the GPU calls the built-in convolutional neural network model to extract features from the high-definition detail image data after format conversion, and obtains the first image feature vector.

[0031] After receiving the numerical calculation task, the CPU performs statistical analysis on the normalized multimodal data, calculates the difference between temperature and humidity and normal range, the value of excessive gas component concentration, and the decibel abnormal value of ambient sound, and obtains the data statistical results.

[0032] After receiving the model inference task, the NPU loads a pre-trained environmental risk identification model and performs real-time target detection and recognition on high-definition detailed image data, generating a second identification result that includes target category and confidence level; the target categories include sewage discharge and garbage accumulation.

[0033] The parallel scheduling unit collects the first image feature vector, data statistics results, and second identification results, and integrates them through a data fusion algorithm to generate preliminary results containing suspected risk features, abnormal environmental parameter values, and risk category confidence.

[0034] Specifically, the step of calling the built-in lightweight real-time learning model and the built-in encryption algorithm to analyze the preliminary results and generate encryption risk warning data includes:

[0035] The edge intelligent processing module reads the preliminary results generated by the parallel scheduling unit and simultaneously retrieves historical environmental risk processing data for the corresponding monitoring area from the local storage unit built into the edge intelligent processing module. The preliminary results and historical environmental risk processing data are then input into the built-in lightweight real-time learning model. The historical environmental risk processing data includes basic information and historical similar risk cases categorized and archived based on this basic information. The basic information includes image features, environmental parameters, and duration of the risk at the time of its occurrence. The first output branch of the lightweight real-time learning model is used to identify the type of environmental risk. The environmental risk type includes one or more of the following: dust, smoke, fire, abnormal water color, and illegal dumping of solid waste. The second output branch of the lightweight real-time learning model is used to determine the risk level based on the quantitative indicators of the risk's scale, concentration, intensity, and duration.

[0036] The lightweight real-time learning model corrects the risk categories and calibrates the risk levels in the preliminary results by comparing and analyzing the processing results and actual impact of similar historical risk cases, thus obtaining the corrected risk categories and calibrated risk levels.

[0037] Specifically, the step of calling the built-in lightweight real-time learning model and the built-in encryption algorithm to analyze the preliminary results and generate encryption risk warning data also includes:

[0038] The edge intelligent processing module integrates the corrected risk category, the calibrated risk level with the corresponding high-definition detailed image data fragments and multimodal data key indicators, and adds analysis timestamps and edge intelligent processing module identification information to form a structured risk analysis result, which is then stored in the local storage unit of the edge intelligent processing module.

[0039] The edge intelligent processing module classifies the risk analysis results, identifies sensitive data in the risk analysis results, and then calls the built-in AES-256 encryption algorithm to generate a random encryption key. The random encryption key is then used to encrypt the sensitive data to obtain an encrypted sensitive data block. The sensitive data includes the geographical coordinates of the monitoring area, the number of the production equipment inside the enterprise, and the concentration data of the sensitive gas components.

[0040] Non-sensitive data and risk overview information in the risk analysis results are kept in plaintext to obtain plaintext data; the non-sensitive data includes risk type names and public area warning information.

[0041] The edge intelligent processing module combines the encrypted sensitive data block with plaintext data to generate encrypted risk warning data.

[0042] Specifically, the transmission of sensitive encrypted data to a designated terminal according to preset permission management rules to complete the early warning feedback includes:

[0043] The terminal control system receives encrypted risk warning data transmitted by the edge intelligent processing module and extracts sensitive encrypted data from it; the encrypted risk warning data includes encrypted sensitive data blocks and plaintext data;

[0044] The terminal control system reads the preset permission management rules stored locally, filters out the list of specified terminals with access rights to the current sensitive encrypted data according to the permission management rules, and determines the permission level corresponding to each specified terminal; the permission management rules include sensitive encrypted data access permission standards for different levels of users and terminals;

[0045] The terminal control system retrieves the encryption key corresponding to the current sensitive encrypted data from the key management unit built into the edge intelligent processing module, and transmits the encryption key to each terminal in the designated terminal list through an encrypted wireless transmission link. At the same time, the sensitive encrypted data is synchronously transmitted to the corresponding designated terminal. The key management unit is used to store the random encryption key generated by the AES-256 encryption algorithm.

[0046] After receiving the encryption key and sensitive encrypted data, each designated terminal uses the encryption key to decrypt the sensitive encrypted data and obtain the complete sensitive information.

[0047] The terminal control system receives early warning information confirmation instructions from the local monitoring center terminal and various designated terminals, summarizes and integrates all the feedback early warning information confirmation instructions to form a structured early warning feedback result, and stores it in the feedback result database of the terminal control system to complete the early warning feedback; the early warning information confirmation instruction includes the reception status of the early warning information and preliminary processing opinions.

[0048] Specifically, the terminal control system sends a parameter adjustment signal to the adaptive multimodal data acquisition module based on the early warning feedback result, including:

[0049] The terminal control system retrieves the stored early warning feedback results from the feedback result database and analyzes the preliminary processing opinions in the early warning feedback results;

[0050] If the preliminary handling opinion indicates that the monitoring frequency of suspected risk source areas should be increased, then the monitoring frequency adjustment parameters should be determined based on the current monitoring interval of the suspected risk source areas.

[0051] If the initial handling opinion indicates that the monitoring range should be expanded, then the field of view adjustment parameters should be determined based on the wide field of view angle range of the dynamic zoom and wide field of view switching lens.

[0052] The terminal control system converts the determined monitoring frequency adjustment parameters or field of view adjustment parameters according to the parameter format recognized by the adaptive multimodal data acquisition module, generates corresponding parameter signals, encapsulates the parameter signals into parameter adjustment signals, and sends them to the adaptive multimodal data acquisition module.

[0053] Compared with the prior art, the beneficial effects of the present invention are:

[0054] 1. This invention proposes an AI-powered visual early warning terminal for environmental risks. It achieves precise data acquisition through an adaptive multimodal data acquisition module. First, it uses a wide-field-of-view mode to cover a large area for monitoring, then switches to a dynamic zoom mode to capture risk details. Multimodal data is combined to supplement environmental information, solving the problems of limited field of view and single data type in traditional equipment. The edge intelligent processing module utilizes parallel processing of CPU, GPU, and NPU, along with lightweight model analysis, balancing data processing efficiency and recognition accuracy. Simultaneously, encryption algorithms ensure data security, improving the comprehensiveness, timeliness, and reliability of environmental risk monitoring.

[0055] 2. This invention proposes an AI visual early warning terminal for environmental risks. The terminal constructs a complete closed loop of data collection, processing, feedback, and optimization. The terminal control system dynamically adjusts the collection parameters based on the early warning feedback results. This not only strengthens the key monitoring of risk areas but also flexibly adapts to the monitoring needs of complex scenarios such as the field and remote areas, reducing the cost of manual intervention. The differentiated processing method of directly outputting non-sensitive information and transmitting sensitive information according to permissions balances the convenience of early warning with the protection of data privacy. Attached Figure Description

[0056] Figure 1 This is a flowchart illustrating the principle of the environmental risk AI visual early warning terminal of the present invention.

[0057] Figure 2 This is a flowchart illustrating the process of encrypting risk warning data in the environmental risk AI visual early warning terminal of this invention. Detailed Implementation

[0058] Example 1:

[0059] Please see Figure 1 One embodiment of the present invention is an environmental risk AI visual early warning terminal, which is adapted to complex scenarios such as the wild and remote areas, including: an adaptive multimodal data acquisition module, an edge intelligent processing module and a terminal control system;

[0060] Wide-area environmental image data is acquired by the dynamic zoom and wide-area field-of-view switching lens in the adaptive multimodal data acquisition module and output to the terminal control system. The terminal control system performs preliminary analysis on the wide-area environmental image data, identifies suspected risk source areas, and sends a control signal to the dynamic zoom and wide-area field-of-view switching lens to switch it to dynamic zoom mode and acquire high-definition detailed image data. Combined with the multimodal data of the monitoring area acquired simultaneously, the data is input into the edge intelligent processing module.

[0061] The edge intelligent processing module distributes the received high-definition detailed image data and multimodal data to the CPU, GPU, and NPU for parallel processing through the built-in parallel scheduling unit to generate preliminary results. Then, it calls the built-in lightweight real-time learning model and the built-in encryption algorithm to analyze the preliminary results and generate encryption risk warning data.

[0062] The terminal control system receives encrypted risk warning data, decrypts and outputs non-sensitive warning information, and transmits sensitive encrypted data to designated terminals according to preset permission management rules to complete the warning feedback. Based on the warning feedback results, the terminal control system sends parameter adjustment signals to the adaptive multimodal data acquisition module.

[0063] When acquiring wide-area environmental image data through the wide-area field-of-view mode of the dynamic zoom and wide-area field-of-view switching lens in the adaptive multimodal data acquisition module, the following are included:

[0064] A1: The ambient light sensing unit built into the adaptive multimodal data acquisition module detects the light intensity data of the monitoring area at the first resolution and transmits the light intensity data to the terminal control system. The terminal control system determines whether the current lighting conditions are suitable for the image acquisition requirements of the dynamic zoom and wide field of view switching lens based on the received light intensity data, and sends an exposure parameter adjustment signal to the adaptive multimodal data acquisition module.

[0065] Furthermore, the terminal control system determines whether the current lighting conditions are suitable for the image acquisition requirements of the dynamic zoom and wide-field-of-view switching lens based on the received light intensity data, and sends an exposure parameter adjustment signal to the adaptive multimodal data acquisition module, including:

[0066] (1) The terminal control system receives the light intensity data transmitted by the ambient light sensing unit in the adaptive multimodal data acquisition module; the light intensity data is obtained by the ambient light sensing unit after detecting the monitoring area at a first resolution;

[0067] (2) Call the local storage of the illumination adaptation standard; the illumination adaptation standard is based on the hardware parameters of the dynamic zoom and wide field of view switching lens and the illumination conditions when acquiring high-quality images in the past, and includes the range of illumination intensity suitable for image acquisition under different lens modes, i.e. the suitable range.

[0068] (3) The terminal control system compares the received current light intensity data with the called light adaptation standard. If the current light intensity data is within the appropriate range of the corresponding lens mode, it means that the current light conditions are suitable for image acquisition and there is no need to adjust the exposure parameters. The terminal control system will not send an exposure parameter adjustment signal for the time being. If the current light intensity data is lower than the lower limit of the appropriate range, it means that the light is too weak, which may cause the acquired image to be too dark and lose details. The terminal control system will determine that the exposure parameters need to be increased. If the current light intensity data is higher than the upper limit of the appropriate range, it means that the light is too strong, which may cause the acquired image to be overexposed and reduce the contrast. The terminal control system will determine that the exposure parameters need to be reduced.

[0069] (4) The terminal control system calculates the exposure parameter adjustment value based on the degree of deviation between the light intensity data and the suitable range. For example, when the light intensity data is lower than the lower limit, the exposure time to be extended is determined according to the difference between the two and the preset ratio. The larger the difference, the longer the extension time. When the light intensity data is higher than the upper limit, the exposure time to be shortened is determined according to the difference between the two. The larger the difference, the longer the shortening time.

[0070] (5) The calculated exposure parameter adjustment values ​​are integrated into an exposure parameter adjustment signal and sent to the adaptive multimodal data acquisition module so as to adjust the exposure time of the dynamic zoom and wide field of view switching lens according to the exposure parameter adjustment signal.

[0071] A2: The adaptive multimodal data acquisition module adjusts the exposure time of the dynamic zoom and wide field-of-view switching lens based on the received exposure parameter adjustment signal; the dynamic zoom and wide field-of-view switching lens is an integrated motor-driven optical lens that electronically switches between wide-angle fisheye mode and telephoto zoom mode; the wide-angle fisheye mode corresponds to the wide field-of-view mode; the telephoto zoom mode corresponds to the dynamic zoom mode;

[0072] Furthermore, the specific steps of A2 include:

[0073] (1) The signal parsing unit inside the adaptive multimodal data acquisition module immediately decomposes the received exposure parameter adjustment signal, extracts the adjustment direction and specific adjustment amount of the exposure time, and verifies the integrity and validity of the exposure parameter adjustment signal to ensure that the signal is not lost or tampered with during transmission. If there is an abnormality in the exposure parameter adjustment signal, the re-reception mechanism is triggered. If the exposure parameter adjustment signal is normal, the exposure adjustment instruction is extracted.

[0074] It is important to understand that the adaptive multimodal data acquisition module receives exposure parameter adjustment signals sent by the terminal control system. These signals are generated by the terminal control system based on a comparison of the light intensity data of the monitored area with the lens adaptation standard. They contain complete exposure adjustment instructions, which specify the direction and amount of exposure time adjustment, such as how much to extend or shorten the exposure time. These instructions are the core basis for guiding lens parameter adjustments.

[0075] (2) The signal analysis unit transmits the extracted exposure adjustment command to the lens control unit in the adaptive multimodal data acquisition module. The lens control unit retrieves the current working status data of the lens. The lens control unit is directly associated with the dynamic zoom and wide field of view switching lens and is responsible for controlling various parameters of the lens. The working status data includes the current mode, the current exposure time parameter and the sensitivity setting of the photosensitive element inside the lens.

[0076] (3) The lens control unit calculates based on the adjustment direction and amount in the exposure adjustment command, combined with the current working status data of the lens:

[0077] If the exposure adjustment command is to extend the exposure time, the lens control unit adds the current exposure time to the extension amount in the exposure adjustment command to obtain a first type of preliminary target exposure time. This first type of preliminary target exposure time is then compared with the maximum exposure threshold allowed by the lens hardware. If the first type of preliminary target exposure time does not exceed the maximum exposure threshold, it is determined as the final target exposure time. If the first type of preliminary target exposure time exceeds the maximum exposure threshold, the maximum exposure threshold is set as the final target exposure time. The maximum exposure threshold is set based on the tolerance limit of the lens's image sensor and the image acquisition sharpness requirements.

[0078] If the exposure adjustment command explicitly states that the adjustment direction is to shorten the exposure time, the lens control unit subtracts the shortening amount in the exposure adjustment command from the current exposure time to calculate the second type of preliminary target exposure time. Then, it retrieves the preset minimum exposure threshold from the lens hardware parameters and compares the second type of preliminary target exposure time with the minimum exposure threshold. If the second type of preliminary target exposure time is not lower than the minimum exposure threshold, then the second type of preliminary target exposure time is determined as the final target exposure time; if the second type of preliminary target exposure time is lower than the minimum exposure threshold, then the minimum exposure threshold is set as the final target exposure time, ensuring that the adjusted exposure time can acquire valid image data. The minimum exposure threshold is set based on the minimum exposure requirement that the lens sensor can capture valid image information.

[0079] (4) The lens control unit sends a drive signal to the exposure control component inside the lens. When it is necessary to extend the exposure time, the drive signal controls the shutter motor to slow down the shutter closing speed and at the same time extends the light-sensing time of the lens sensor. When it is necessary to shorten the exposure time, the drive signal speeds up the shutter closing speed and shortens the light-sensing time of the lens sensor. During the adjustment process, the lens control unit monitors the response status of the exposure control component in real time to ensure that its action is consistent with the exposure adjustment command. The exposure control component includes a motor that controls the shutter speed and a circuit that adjusts the light-sensing time of the lens sensor.

[0080] (5) After the adjustment is completed, the lens control unit obtains the actual exposure time parameter of the lens through the feedback circuit, compares it with the calculated final target exposure time, and verifies whether the adjustment has achieved the expected effect. If the actual exposure time is consistent with the final target exposure time, it means that the adjustment is successful. The lens control unit sends the adjustment completion signal to the main control unit of the adaptive multimodal data acquisition module. If there is a deviation, the fine-tuning command is sent again until the actual exposure time meets the target requirements.

[0081] A3: After the exposure time is adjusted, the adaptive multimodal data acquisition module controls the dynamic zoom and wide field of view switching lens to switch to wide field of view mode, and acquires wide-area environmental image data of the monitoring area through the wide field of view mode.

[0082] After receiving the exposure parameter adjustment signal sent by the terminal control system, the adaptive multimodal data acquisition module also includes:

[0083] A2.1: Analyze the adjustment parameters in the exposure parameter adjustment signal;

[0084] If the exposure parameter adjustment signal also carries the monitoring frequency adjustment parameter, then the acquisition interval time of the dynamic zoom and wide field of view switching lens and the acquisition cycle of the multimodal sensor are adjusted according to the monitoring frequency adjustment parameter.

[0085] It should be noted that the monitoring frequency adjustment parameters include two parts of information: one is the acquisition interval adjustment requirement for dynamic zoom and wide field of view switching lenses, and the other is the acquisition cycle adjustment requirement for multimodal sensors. The acquisition interval adjustment requirement and the acquisition cycle adjustment requirement are generated by the terminal control system based on the early warning feedback results. For example, when it is necessary to strengthen the tracking of suspected risk source areas, it will be required to shorten the acquisition interval and cycle.

[0086] Furthermore, the process of adjusting the acquisition interval includes: if the monitoring frequency adjustment parameter requires a shorter acquisition interval, the control unit subtracts the specified shortening amount from the current interval to obtain a first type of preliminary target interval. Then, it retrieves the minimum acquisition interval allowed by the lens hardware and compares the first type of preliminary target interval with the minimum acquisition interval. If the first type of preliminary target interval is not less than the minimum acquisition interval, it is determined as the final target interval. If the first type of preliminary target interval is less than the minimum acquisition interval, the minimum acquisition interval is used as the final target interval to avoid exceeding hardware limits. If the monitoring frequency adjustment parameter requires an extended acquisition interval, the control unit adds the specified extension amount to the current interval to obtain a second type of target interval, while ensuring that the second type of target interval does not exceed a preset maximum interval threshold. The maximum interval threshold is set according to the real-time requirements of environmental monitoring; an excessively long threshold will cause missed critical nodes of risk changes. The minimum acquisition interval is determined by the mechanical response speed of the lens and the processing capability of the image processing unit; an excessively short threshold will cause image data accumulation or acquisition lag.

[0087] Furthermore, the process of adjusting the acquisition period includes: if the monitoring frequency adjustment parameter requires a shorter acquisition period, the control unit subtracts the specified shortening margin from the current period to obtain a first type of preliminary target period, and then retrieves the sensor's minimum acquisition period. If the first type of preliminary target period is not less than the minimum acquisition period, it is determined as the final target period; if it is less than the minimum acquisition period, the minimum acquisition period is taken as the final target period. If the monitoring frequency adjustment parameter requires an extended acquisition period, the control unit adds the specified extension margin to the current period to obtain a second type of target period, while ensuring that the second type of target period does not exceed the preset maximum period threshold. The minimum acquisition period is determined by the sensor's sampling rate and data storage capacity; too short a period will lead to data redundancy or sensor overheating. The maximum period threshold is set according to the timeliness requirements of multimodal data; too long a period will lead to the omission of environmental parameter changes.

[0088] If the exposure parameter adjustment signal also carries the field of view adjustment parameter, then the wide field of view angle of the dynamic zoom and wide field of view switching lens in wide field of view mode is adjusted according to the field of view adjustment parameter.

[0089] A2.2: After all parameters have been adjusted, the adaptive multimodal data acquisition module sends a parameter adjustment completion confirmation signal to the terminal control system.

[0090] When the terminal control system performs preliminary analysis of wide-area environmental image data, it includes:

[0091] B1: The terminal control system receives wide-area environmental image data transmitted by the adaptive multimodal data acquisition module, and performs noise reduction processing on the wide-area environmental image data by calling the built-in image noise reduction algorithm to obtain the noise-reduced wide-area environmental image data.

[0092] In this invention, the image denoising algorithm adopts a transform domain-based denoising algorithm. However, the transform domain-based denoising algorithm is existing technology in this field and is not an inventive solution of this application, so it will not be described in detail here.

[0093] B2: The built-in image segmentation algorithm divides the denoised wide-area environmental image data into multiple sub-region image data, with each sub-region image data corresponding to a local area of ​​the monitoring area;

[0094] In this invention, the image segmentation algorithm adopts an edge detection-based segmentation algorithm. However, the edge detection-based segmentation algorithm is existing technology in this field and is not an inventive solution of this application, so it will not be described in detail here.

[0095] B3: The terminal control system retrieves the preset stored normal environment image template, compares the image data of each sub-region obtained by segmentation with the corresponding sub-region template in the preset normal environment image template, and calculates the difference value of each sub-region image data and sub-region template in color, texture and contour features; the normal environment image template is the standard image data of the corresponding monitoring area under the state of no environmental risk, and contains the normal visual features of each sub-region.

[0096] Furthermore, the specific steps for B3 include:

[0097] (1) After the terminal control system completes the segmentation of the wide-area environmental image data, it starts the template retrieval process, including: reading the identification information of the current monitoring area, such as the area number and geographic location code; accessing the template database in the local storage unit according to the identification information, and filtering out the preset normal environmental image template that is completely corresponding to the current monitoring area in the template database, so as to ensure that the normal environmental image template and the image data of the currently segmented sub-area are completely consistent in terms of area division and size ratio, and avoid comparison error due to template mismatch;

[0098] Furthermore, the identification information is synchronously recorded and transmitted to the terminal control system when the adaptive multimodal data acquisition module is started, and is used to accurately match the corresponding normal environmental image template.

[0099] Furthermore, the environmental normal image template is generated when there are no environmental risks in the monitored area, by acquiring images in wide field of view mode using the same dynamic zoom and wide field of view switching lens, and then processing them using the same image segmentation algorithm.

[0100] (2) Preprocess the retrieved normal environment image templates, such as performing integrity checks;

[0101] (3) Establish sub-region matching correspondence, including: reading the position coordinate information of each sub-region image data after current segmentation, such as the horizontal and vertical coordinate range in the entire wide-area environment image data, and reading the position coordinate information of each sub-region template in the normal environment image template. By comparing the coordinates, the currently segmented sub-region is matched one by one with the sub-region in the template to form a one-to-one comparison combination. For example, the vegetation sub-region located in the upper left corner of the wide-area environment image with a coordinate range of any interval will be accurately matched with the vegetation sub-region template with the same coordinate range in the template, ensuring that each sub-region to be compared can find a unique corresponding normal template and avoiding cross-region mismatch.

[0102] (4) The terminal control system extracts and compares features in the order of color, texture, and outline, and calculates the difference value.

[0103] In the color feature comparison stage, the terminal control system uses a color space conversion method to extract color features from the current sub-region image data and the corresponding template sub-region. Specifically, this includes: converting the current sub-region image from RGB space to HSV space, then calculating the mean hue, mean saturation, and mean brightness of the two sub-regions in the HSV space, and then calculating the absolute difference between the two in terms of hue, saturation, and brightness. Different weights are assigned to hue, saturation, and brightness according to their importance to environmental risk identification. Finally, a weighted sum is used to obtain the comprehensive difference value of the color features. The color space conversion method is existing technology in this field and is not an inventive solution of this application, so it will not be described in detail here.

[0104] In the texture feature comparison stage, the terminal control system uses the gray-level co-occurrence matrix method to extract texture features. Specifically, this includes: first, converting the current sub-region image data and the corresponding template sub-region into grayscale images; then, selecting multiple directions and multiple distance intervals in the grayscale images, calculating the gray-level co-occurrence matrix for each direction and interval, and extracting four texture parameters—contrast, correlation, energy, and homogeneity—from the gray-level co-occurrence matrix; calculating the relative differences between the current sub-region and the template sub-region on these four parameters, and then averaging the four relative differences to obtain the comprehensive difference value of the texture features. The gray-level co-occurrence matrix method is prior art in this field and is not an inventive solution of this application, so it will not be described in detail here.

[0105] In the contour feature comparison stage, the terminal control system uses the Canny edge detection algorithm to perform edge detection on the current sub-region image data and the corresponding template sub-region, respectively, to obtain the edge contour maps of the two sub-regions. Subsequently, contour features are extracted from the edge contour maps, including calculating the perimeter, area, circularity, rectangularity of the contour, and the coordinates of feature points on the contour. Next, the absolute difference in perimeter, area, circularity, and rectangularity of the edge contour maps of the two sub-regions is calculated, and the average deviation of the feature point positions is calculated by comparing the Euclidean distance of the feature point coordinates. The absolute difference and the average deviation are weighted and summed to obtain the comprehensive difference value of the contour features, so as to identify contour additions caused by garbage accumulation and contour deformation anomalies caused by terrain changes. The Canny edge detection algorithm is the prior art in this field and is not an inventive solution of this application, so it will not be described in detail here.

[0106] (5) The terminal control system summarizes the comprehensive difference values ​​of the three features of color, texture and contour, and assigns a total weight to each feature according to the importance of each feature under different environmental scenarios. The weighted calculation is used to obtain the overall feature difference value between the image data of each sub-region and the corresponding template sub-region.

[0107] B4: The terminal control system compares the difference value of the image data of each sub-region with a preset threshold. If the difference value of the image data of any sub-region exceeds the preset threshold, it is determined that there is an anomaly in the corresponding sub-region and the monitoring area corresponding to it is marked as a suspected risk source area. The preset threshold is set according to historical monitoring data and is used to determine whether there is a critical value for anomalies.

[0108] Sending a control signal to the dynamic zoom and wide field-of-view switching lens to switch it to dynamic zoom mode and acquire high-definition detail image data includes:

[0109] C1: After completing the preliminary analysis of the wide-area environmental image data, the terminal control system acquires the coordinate information of the marked suspected risk source areas in the wide-area environmental image data, and converts the coordinate information into the lens focus position parameters based on the optical parameters of the dynamic zoom and wide field of view switching lens. At the same time, combined with the size information of the suspected risk source areas in the wide-area environmental image data, the system determines the lens zoom ratio parameter covering the suspected risk source areas, and integrates the focus position parameter and the lens zoom ratio parameter into a control signal and sends it to the dynamic zoom and wide field of view switching lens; the optical parameters include the lens focal length and the corresponding relationship of the field of view angle;

[0110] Furthermore, the coordinate information is converted into the lens's focus position parameters based on the optical parameters of the dynamic zoom and wide field-of-view switching lens, including:

[0111] (1) After the terminal control system obtains the coordinate information of the marked suspected risk source area in the wide-area environmental image data, it starts the optical parameter calling process and calls the optical parameter database of the dynamic zoom and wide-area field of view switching lens; the coordinate information is in pixel units and includes the pixel coordinates of the upper left and lower right corners of the area, which comes from the preliminary analysis results after the wide-area environmental image segmentation; the optical parameter database stores the complete optical characteristic parameters of the lens, including the focal length range in the wide-area field of view mode and the dynamic zoom mode, the field of view angle value of the corresponding mode, the corresponding relationship curve of the field of view angle changing with the focal length, and the conversion ratio of the pixel size and the actual physical size of the monitoring area when the lens is imaging. These parameters provide the core basis for coordinate transformation.

[0112] (2) Using the optical parameters called as input, first extract the lens focal length and corresponding field of view values ​​in the wide field of view mode, and calculate the actual physical angle corresponding to each pixel in the wide field of view image based on these two parameters: divide the horizontal field of view in the wide field of view mode by the total number of horizontal pixels in the wide field of view image to obtain the horizontal physical angle covered by a single pixel in the monitoring area; divide the vertical field of view by the total number of vertical pixels in the wide field of view image to obtain the vertical physical angle covered by a single pixel, thereby determining the correspondence between the pixel and the actual physical angle.

[0113] (3) Using the pixel coordinates of the suspected risk source area as input, calculate the pixel coordinates of the center point of the suspected risk source area: by adding the horizontal pixel coordinates of the upper left corner and the lower right corner of the area and taking the average value, the horizontal pixel coordinates of the center point are obtained; similarly, by adding the vertical pixel coordinates of the upper left corner and the lower right corner of the area and taking the average value, the vertical pixel coordinates of the center point are obtained, thereby determining the specific pixel position of the center point of the suspected risk source area in the wide-area environment image.

[0114] (4) Using the center point pixel coordinates and pixel-physical angle correspondence as input, the center point pixel coordinates are converted into the offset angle relative to the lens optical axis: First, calculate the difference between the horizontal pixel coordinates of the center point and the center point of the horizontal pixel of the wide-area environment image, multiply it by the horizontal physical angle of a single pixel to obtain the horizontal offset angle of the center point relative to the lens optical axis; then calculate the difference between the vertical pixel coordinates of the center point and the center point of the vertical pixel of the wide-area environment image, multiply it by the vertical physical angle of a single pixel to obtain the vertical offset angle, thereby clarifying the target offset angle that the lens optical axis needs to be aligned with.

[0115] (5) The terminal control system takes the target offset angle and the motor control parameters in the lens optical parameters as input to determine the amount of motor rotation required to align the lens optical axis with the target offset angle: according to the horizontal offset angle, the motor rotation amount in the horizontal direction is obtained by looking up the motor horizontal rotation amount lookup table; according to the vertical offset angle, the motor rotation amount in the vertical direction is obtained by looking up the motor vertical rotation amount lookup table. The motor rotation amounts in these two directions together constitute the focus position parameters, which directly correspond to the rotation angle of the focus motor inside the lens, thus completing the conversion of coordinate information to focus position parameters; the motor control parameters are pre-stored in the optical parameter database to record the correspondence between the offset angle and the motor rotation amount.

[0116] Furthermore, by combining the size information of the suspected risk source area in the wide-area environmental image data, the zoom ratio parameters of the lens covering the suspected risk source area are determined, including:

[0117] (1) The terminal control system extracts the size information of the marked suspected risk source area in the wide-area environmental image data from the marked suspected risk source area information, namely the pixel width and pixel height of the area; the pixel width and pixel height of the area are obtained by calculating the horizontal difference and vertical difference of the pixel coordinates of the upper left corner and the lower right corner of the area, reflecting the size of the pixel range of the area in the wide-area environmental image;

[0118] (2) Call the actual physical angle corresponding to a single pixel in the wide field of view mode that has been acquired. With the pixel width and horizontal physical angle as input, multiply the pixel width by the horizontal physical angle of a single pixel to obtain the horizontal angle of the suspected risk source area in the actual space. At the same time, with the pixel height and vertical physical angle as input, multiply the pixel height by the vertical physical angle of a single pixel to obtain the vertical angle in the actual space.

[0119] (3) Retrieve the curve of the relationship between the focal length and the field of view of the lens in dynamic zoom mode from the lens optical parameter database. Take the horizontal and vertical angles in the actual space as input, and query the minimum field of view that can cover both angles at the same time in the curve. That is, the minimum field of view combination that is not less than the actual horizontal angle and not less than the actual vertical angle. Then match the corresponding focal length value in the curve according to the field of view combination, that is, the target focal length.

[0120] (4) Retrieve the current focal length value used in the wide field of view mode. With the target focal length and the current wide field of view mode focal length as input, calculate the ratio between the two. Divide the target focal length by the current wide field of view mode focal length, and the resulting value is the initial zoom factor parameter.

[0121] (5) The terminal control system extracts the maximum zoom magnification and minimum zoom magnification of the lens from the lens optical parameter database, and compares the initially calculated zoom magnification parameter with the maximum zoom magnification and minimum zoom magnification: if the initial zoom magnification parameter is between the maximum zoom magnification and the minimum zoom magnification, it is directly determined as the final zoom magnification parameter; if the initial zoom magnification parameter is greater than the maximum zoom magnification, the maximum zoom magnification is used as the final zoom magnification parameter; if the initial zoom magnification parameter is less than the minimum zoom magnification, the minimum zoom magnification is used as the final zoom magnification parameter, ensuring that the determined zoom magnification parameter is within the range allowed by the lens hardware and can be normally executed by the dynamic zoom and wide field of view switching lens.

[0122] C2: After receiving the control signal, the dynamic zoom and wide field of view switching lens analyzes the focus position parameter in the control signal, drives the focus motor in the lens to adjust the lens position to the corresponding focus position, analyzes the zoom magnification parameter, drives the zoom motor to adjust the lens spacing to the specified zoom magnification, and completes the switching from wide field of view mode to dynamic zoom mode.

[0123] C3: The dynamic zoom and wide field of view switching lens activates the built-in image sensor to acquire high-definition detailed image data of the suspected risk source area at a second resolution; the second resolution is higher than the first resolution.

[0124] Example 2:

[0125] Please see Figure 2 The process of generating the preliminary results in this embodiment includes:

[0126] D1: The edge intelligent processing module receives high-definition detailed image data and multimodal data transmitted by the adaptive multimodal data acquisition module, performs preprocessing, and encapsulates them into data transmission packets;

[0127] D2: The parallel scheduling unit parses the data transmission packet, separates the high-definition detail image data after format conversion and the normalized multimodal data, and calls the preset task allocation rules to allocate the feature extraction task in the high-definition detail image data to the GPU, the numerical calculation task in the multimodal data to the CPU, and the model inference task for environmental risk identification to the NPU, and generates task allocation instructions and sends them to the corresponding processors.

[0128] After receiving the feature extraction task, the GPU calls the built-in convolutional neural network model to extract features from the high-definition detail image data after format conversion, and obtains the first image feature vector. The convolutional neural network model is existing technology in this field and is not an inventive solution of this application, so it will not be described in detail here.

[0129] After receiving the numerical calculation task, the CPU performs statistical analysis on the normalized multimodal data, calculates the difference between temperature and humidity and normal range, the value of excessive gas component concentration, and the decibel abnormal value of ambient sound, and obtains the data statistical results.

[0130] After receiving the model inference task, the NPU loads a pre-trained environmental risk identification model and performs real-time target detection and recognition on high-definition detailed image data, generating a second identification result that includes target category and confidence level; the target category includes, but is not limited to, sewage discharge and garbage accumulation.

[0131] D3: The parallel scheduling unit collects the first image feature vector, data statistics results, and second identification results, and integrates them through a data fusion algorithm to generate preliminary results containing suspected risk features, abnormal environmental parameter values, and risk category confidence.

[0132] In this invention, the data fusion algorithm adopts a weighted fusion algorithm, which is the prior art in this field and is not an inventive solution of this application, so it will not be described in detail here.

[0133] The process involves analyzing the preliminary results using the built-in lightweight real-time learning model and the built-in encryption algorithm to generate encryption risk warning data, including:

[0134] E1: The edge intelligent processing module reads the preliminary results generated by the parallel scheduling unit, and synchronously retrieves the historical environmental risk processing data of the corresponding monitoring area from the local storage unit built into the edge intelligent processing module. The preliminary results and the historical environmental risk processing data are then input into the built-in lightweight real-time learning model.

[0135] The historical environmental risk processing data includes basic information and historical similar risk cases classified and archived based on the basic information; the basic information includes image features, environmental parameters, and risk duration at the time of the historical risk occurrence.

[0136] The first output branch of the lightweight real-time learning model is used to identify environmental risk types; the environmental risk types include one or more of dust, smoke, fire, abnormal water color, and illegal dumping of solid waste; the second output branch of the lightweight real-time learning model is used to determine the risk level based on the quantitative indicators of the risk's scale, concentration, intensity, and duration.

[0137] E2: The lightweight real-time learning model corrects the risk categories in the preliminary results and calibrates the risk levels by comparing and analyzing the processing results and actual impact of similar historical risk cases, thus obtaining the corrected risk categories and calibrated risk levels.

[0138] Furthermore, the lightweight real-time learning model is a multi-task learning model, which receives the preliminary results and outputs classification results for risk event types and rating results for the severity of risk events by analyzing the correlation between the features of different modal data, which together constitute the risk analysis results. The multi-task learning model is existing technology in this field and is not an inventive solution of this application, so it will not be described in detail here.

[0139] E3: The edge intelligent processing module integrates the corrected risk category, the calibrated risk level with the corresponding high-definition detailed image data fragments and multimodal data key indicators, and adds analysis timestamps and edge intelligent processing module identification information to form a structured risk analysis result, which is then stored in the local storage unit of the edge intelligent processing module.

[0140] E4: The edge intelligent processing module classifies the risk analysis results, identifies sensitive data in the risk analysis results, and then calls the built-in AES-256 encryption algorithm to generate a random encryption key. The random encryption key is then used to encrypt the sensitive data to obtain an encrypted sensitive data block. The sensitive data includes the specific geographical coordinates of the monitoring area, the number of the production equipment inside the enterprise, and the precise concentration data of the sensitive gas components. The AES-256 encryption algorithm is existing technology in this field and is not an inventive solution of this application, so it will not be described in detail here.

[0141] Furthermore, the specific steps of E4 include:

[0142] (1) The main control unit of the edge intelligent processing module retrieves the generated structured risk analysis results and starts the data classification engine; the data classification engine pre-stores the feature recognition rules of sensitive data; the feature recognition rules are formulated according to the privacy protection requirements and industry standards of environmental monitoring data, and clarify the specific types of sensitive data and their corresponding recognition features;

[0143] (2) The data classification engine identifies each component of the risk analysis results in a preset order. When identifying the specific geographical coordinates of the monitoring area, the data classification engine filters out such sensitive data by matching the latitude and longitude format, regional code identifier and other features contained in the data. When identifying the production equipment number inside the enterprise, the data classification engine locates the relevant information from the data according to the preset equipment numbering rules. When identifying the precise concentration data of sensitive gas components, the data classification engine determines such sensitive data by judging whether the data is associated with the gas type in the list of toxic and harmful gases and is accompanied by specific values ​​and detection units. At the same time, the data classification engine will exclude non-sensitive data, complete the classification of sensitive data and non-sensitive data, and generate a list of sensitive data, recording the storage location and data length of each type of sensitive data.

[0144] (3) The main control unit calls the built-in AES-256 encryption algorithm to start the key generation process, including: collecting physical noise signals, such as random signals generated by voltage fluctuations and temperature changes in the circuit, through a hardware random number generator, and using them as the seed for key generation; performing hash operation and key expansion on the random seed according to the requirements of the AES-256 encryption standard to generate a random encryption key with a length of 256 bits, and performing integrity verification on the generated random encryption key to ensure the key is valid by verifying whether the number of bits and byte distribution of the key conform to the AES-256 encryption standard; after the verification is passed, the main control unit stores the random encryption key in the key management unit in the module and adds a unique identifier to the key, associating it with the corresponding sensitive data list;

[0145] (4) The main control unit extracts all sensitive data from the risk analysis results according to the sensitive data list, integrates them into continuous sensitive data blocks according to data type, and preprocesses the sensitive data blocks, including: first converting sensitive data of different formats into binary data streams, and then dividing the binary data streams into fixed-length data packets according to the grouping requirements of the AES-256 encryption algorithm;

[0146] (5) The main control unit retrieves the generated random encryption key from the key management unit and inputs it into the AES-256 encryption algorithm. At the same time, it inputs the pre-processed sensitive data packets one by one into the AES-256 encryption algorithm to generate encrypted data packets.

[0147] (6) All encrypted data packets are spliced ​​together in the original order to form a complete encrypted sensitive data block and generate an encryption verification code. The main control unit receives the encrypted sensitive data block and the encryption verification code, stores them together, and marks the corresponding key identifier.

[0148] E5: Keep the non-sensitive data and risk overview information in the risk analysis results in plaintext to obtain plaintext data; the non-sensitive data includes, but is not limited to, risk type names and public area warning information.

[0149] Furthermore, the risk overview information is a general description of the environmental risk. It is necessary to verify whether the content only includes publicly available information such as the general area where the risk occurs, the risk level, and preliminary response suggestions, and does not involve private information such as the specific geographical location of the monitoring area, the name of the company involved, or the serial number of sensitive equipment.

[0150] Furthermore, the specific steps for E5 include:

[0151] (1) After completing the identification and extraction of sensitive data, the main control unit of the edge intelligent processing module retrieves the classified risk analysis results, including the list of sensitive data and the remaining unlabeled data, and starts the non-sensitive data screening engine; the non-sensitive data screening engine pre-stores the judgment rules for non-sensitive data; the judgment rules are formulated based on the openness requirements and privacy protection standards of environmental monitoring data, and clarify the specific types and characteristics of non-sensitive data;

[0152] (2) The non-sensitive data screening engine verifies each data item according to the composition structure of the risk analysis results to see if it belongs to the non-sensitive category. When screening non-sensitive data, the non-sensitive data screening engine first checks whether the data is in the sensitive data list. If the data is not included in the sensitive data list, it further verifies the data characteristics: For risk category data, it determines whether it is a publicly disclosed general category and excludes related descriptions containing specific locations and enterprise information; For key indicators of multimodal data, it screens basic parameters without sensitive attributes, such as ordinary temperature and humidity values ​​and the normal concentration range of non-sensitive gases, and excludes the precise concentration of sensitive gases and special equipment operating parameters; For analysis timestamp and module identification information, it confirms that it only contains general information such as time and module number, without additional sensitive related data, and meets the non-sensitive judgment criteria.

[0153] (3) The non-sensitive data filtering engine performs a special verification of the risk overview information. If there are suspected sensitive statements in the risk overview information, the non-sensitive data filtering engine will automatically mark them and send them back to the main control unit. The main control unit will then initiate a manual review process to confirm that there is no sensitive information before including it in the non-sensitive category.

[0154] (4) The non-sensitive data filtering engine integrates all non-sensitive data and risk overview information that have passed the verification according to the original data format, and maintains the original structure and encoding format of the data. For example, risk category data is still stored in text format, multimodal basic parameters are kept in numerical format, and timestamps are kept in standard time encoding to avoid data distortion or information loss due to format conversion. At the same time, the non-sensitive data filtering engine performs integrity checks on the integrated non-sensitive data, and verifies whether the number of data items and the length of data are completely consistent with the non-sensitive parts in the risk analysis results, to ensure that no data is omitted or deleted by mistake.

[0155] (5) The main control unit performs format standardization processing on the integrated non-sensitive data and risk overview information, and adjusts different types of non-sensitive data according to the preset plaintext data format specifications. For example, it unifies the character encoding of text data, the unit expression of numerical data, and the format of timestamps to ensure that plaintext data has uniform readability and compatibility.

[0156] (6) The main control unit performs a final verification of the standardized non-sensitive data and risk overview information. By comparing the original risk analysis results, it confirms that all non-sensitive data does not contain sensitive information and that the data content and format are correct. After the verification is passed, it is marked as plaintext data and stored in the ordinary data storage area of ​​the edge intelligent processing module, which is physically isolated from the sensitive data encryption storage area. At the same time, a plaintext data list is generated to record the data type, storage location and verification results.

[0157] E6: The edge intelligent processing module combines the encrypted sensitive data block with plaintext data to generate encrypted risk warning data.

[0158] The transmission of sensitive encrypted data to a designated terminal according to preset permission management rules, and the completion of early warning feedback, include:

[0159] F1: The terminal control system receives encrypted risk warning data transmitted by the edge intelligent processing module and extracts sensitive encrypted data from it; the encrypted risk warning data includes encrypted sensitive data blocks and plaintext data;

[0160] F2: The terminal control system reads the preset permission management rules stored locally, filters out the list of specified terminals with current access rights to sensitive encrypted data according to the permission management rules, and determines the permission level corresponding to each specified terminal; the permission management rules include sensitive encrypted data access permission standards for different levels of users and terminals;

[0161] F3: The terminal control system retrieves the encryption key corresponding to the current sensitive encrypted data from the key management unit built into the edge intelligent processing module, and transmits the encryption key to each terminal in the designated terminal list through the encrypted wireless transmission link, while simultaneously transmitting the sensitive encrypted data to the corresponding designated terminal; the key management unit is used to store the random encryption key generated by the AES-256 encryption algorithm;

[0162] F4: After receiving the encryption key and sensitive encrypted data, each designated terminal uses the encryption key to decrypt the sensitive encrypted data and obtain the complete sensitive information;

[0163] Furthermore, the specific steps of F4 include:

[0164] (1) The communication module of the designated terminal receives the encryption key and sensitive encrypted data transmitted by the terminal control system through the secure channel, and at the same time receives the attached encryption verification code and key identifier. The communication module first performs integrity detection on the received data, and confirms that the encryption key and sensitive encrypted data have not been lost, tampered with or damaged during the transmission by comparing the data packet length and verification code before and after the transmission. If the detection finds that the data is abnormal, it immediately sends a retransmission request to the terminal control system until complete and error-free data is received.

[0165] (2) The communication module transmits the verified encryption key to the hardware security unit built into the terminal. The hardware security unit performs format verification on the encryption key, checks whether the key length conforms to the AES-256 standard and whether the key encoding is a valid binary format, and verifies whether the key identifier matches the identifier of the sensitive encrypted data to avoid decryption failure due to key and data mismatch. If the verification fails, the hardware security unit refuses to store the key and triggers an error message, waiting to receive the correct key again.

[0166] (3) The main control unit of the designated terminal retrieves sensitive encrypted data, starts the AES-256 decryption algorithm module, and sends a key call request to the hardware security unit. The key call request contains the identifier of the sensitive encrypted data and decryption authorization information. After the hardware security unit verifies the validity of the decryption authorization information, it transmits the encryption key to the decryption algorithm module in encrypted form to ensure that the encryption key is not leaked during use.

[0167] (4) After receiving the encryption key, the decryption algorithm module of the designated terminal first preprocesses the sensitive encrypted data: extracts the encryption check code in the data, verifies the integrity of the sensitive encrypted data through hash operation, if they are inconsistent, it is determined that the data has been tampered with, stops decryption and reports an error; if they are consistent, the sensitive encrypted data is divided into fixed-length encrypted data packets according to the grouping rules of AES-256 encryption, and the initial vector used during encryption is extracted at the same time.

[0168] (5) The decryption algorithm module performs inverse round transformation operations on each encrypted data packet according to the AES-256 decryption standard, with the encryption key and initial vector as input. This includes: performing inverse round key addition, inverse column mixing, inverse row shifting, and inverse byte substitution operations in sequence to gradually restore the original binary data of the encrypted data packet; performing inverse padding operation on the last group of encrypted data packets to remove the padding bytes added during encryption and restore the original data length; and verifying the results after each round of decryption to check whether the data format and byte encoding conform to the original data characteristics and eliminate information distortion caused by calculation errors.

[0169] (6) The decryption algorithm module splices all the decrypted data packets in their original order to form complete binary sensitive data. Then, it restores the format according to the original format of the sensitive data and converts the binary sensitive data into readable information such as geographical location coordinates, equipment number, and gas concentration value to obtain complete sensitive information.

[0170] F5: The terminal control system receives the early warning information confirmation instructions from the local monitoring center terminal and each designated terminal, summarizes and integrates all the feedback early warning information confirmation instructions to form a structured early warning feedback result, and stores it in the feedback result database of the terminal control system to complete the early warning feedback; the early warning information confirmation instruction includes the reception status of the early warning information and preliminary processing opinions.

[0171] The terminal control system sends parameter adjustment signals to the adaptive multimodal data acquisition module based on the early warning feedback results, including:

[0172] G1: The terminal control system retrieves the stored early warning feedback results from its own feedback result database and analyzes the preliminary processing opinions in the early warning feedback results;

[0173] If the preliminary handling opinion indicates that the monitoring frequency of suspected risk source areas needs to be increased in order to track risk changes in real time, then the monitoring frequency adjustment parameters should be determined based on the current monitoring interval of the suspected risk source area.

[0174] If the preliminary handling opinion indicates that the monitoring scope needs to be expanded to investigate whether there are related risks around the suspected risk source area, then the field of view adjustment parameters shall be determined according to the wide field of view angle range of the dynamic zoom and wide field of view switching lens.

[0175] G2: The terminal control system converts the determined monitoring frequency adjustment parameters or field of view adjustment parameters into a parameter format that can be recognized by the adaptive multimodal data acquisition module, generates the corresponding parameter signal, encapsulates the parameter signal into a parameter adjustment signal, and sends it to the adaptive multimodal data acquisition module.

[0176] Furthermore, the specific steps of G2 include:

[0177] (1) The parameter processing unit of the terminal control system retrieves the determined adjustment parameters. If the monitoring frequency needs to be optimized, the monitoring frequency adjustment parameters are retrieved. If the monitoring range needs to be expanded / reduced, the field of view adjustment parameters are retrieved. At the same time, the parameter format specification of the adaptive multimodal data acquisition module is extracted from the locally stored device configuration database. The parameter format specification is predefined when the module is manufactured, which clarifies the parameter encoding type, data length, unit identifier and check bit rules. Among them, the encoding type is such as binary or hexadecimal, the data length is such as single byte or double byte, and the unit identifier is such as the encoding of the acquisition interval in seconds and the encoding of the field of view in degrees.

[0178] (2) The parameter processing unit converts the adjustment parameters according to the parameter format specification. If the parameter being processed is the monitoring frequency adjustment parameter, the adjusted acquisition interval time is first converted into the numerical type required by the specification, and then the unit identifier is added according to the specification. If the parameter length does not reach the minimum length required by the specification, zeros are added to ensure that the data length is compliant. If the parameter being processed is the field of view adjustment parameter, the adjusted angle value is first converted into the encoding format specified by the specification, such as converting the angle value into a hexadecimal value, and then the unit identifier representing degrees is added. At the same time, the value is checked to ensure that the value and format of the parameter are in compliance with the module requirements.

[0179] (3) Generate check bits for the converted parameters. According to the parity check rules in the parameter format specification, perform calculations on the converted parameter data to generate corresponding check bits. For example, count the number of 1s in the parameter data. If it is odd, set the check bit to 1; if it is even, set it to 0. Append the check bit to the end of the parameter data to form a complete parameter signal. After generation, the accuracy of the check bit needs to be verified again. By recalculating and checking the verification results, ensure that the parameter signal can be verified as complete by the module during transmission.

[0180] (4) The complete parameter signal is transmitted to the signal encapsulation unit of the terminal control system. The encapsulation unit retrieves the preset signal frame structure specification. The signal frame structure specification defines the frame format of the parameter adjustment signal, including the frame header, parameter type identifier, parameter signal, and frame tail.

[0181] (5) The signal encapsulation unit assembles the parameter adjustment signal according to the signal frame structure specification: First, add the frame header code at the beginning of the parameter adjustment signal, then add the corresponding parameter type identifier according to the type of adjustment parameter. For example, if the monitoring frequency is adjusted, add the corresponding identifier, and if the field of view is adjusted, add another identifier. Then embed the generated parameter signal, and finally add the frame tail code to form a complete parameter adjustment signal frame.

[0182] (6) After assembly, the signal encapsulation unit performs an integrity check on the parameter adjustment signal frame, and verifies whether the order of the frame header, parameter type identifier, parameter signal, and frame tail is correct, and whether the length of each part conforms to the specification, so as to avoid the module being unable to recognize due to structural errors.

[0183] (7) The communication unit of the terminal control system starts the dedicated communication link with the adaptive multimodal data acquisition module and sends the encapsulated parameter adjustment signal frame to the adaptive multimodal data acquisition module. During the transmission process, the communication unit monitors the transmission status in real time. If the transmission is interrupted, such as signal loss or link disconnection, the retransmission mechanism is immediately started to retransmit the signal until the signal reception success response is received from the adaptive multimodal data acquisition module. If the retransmission fails multiple times, a fault alarm is triggered to prompt the staff to check the communication link or module status to ensure that the parameter adjustment signal can be transmitted accurately and stably to the adaptive multimodal data acquisition module.

[0184] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments under the guidance of the present invention without departing from the spirit and scope of the present invention. All of these variations are within the protection scope of the present invention.

Claims

1. An environmental risk AI visual early warning terminal, characterized in that, include: Adaptive multimodal data acquisition module, edge intelligent processing module, and terminal control system; Wide-area environmental image data is acquired by the dynamic zoom and wide-area field-of-view switching lens in the adaptive multimodal data acquisition module and output to the terminal control system. The terminal control system performs preliminary analysis on the wide-area environmental image data, identifies suspected risk source areas, and sends a control signal to the dynamic zoom and wide-area field-of-view switching lens to switch it to dynamic zoom mode and acquire high-definition detailed image data. Combined with the multimodal data of the monitoring area acquired simultaneously, the data is input into the edge intelligent processing module. The edge intelligent processing module distributes the received high-definition detailed image data and multimodal data to the CPU, GPU, and NPU for parallel processing through the built-in parallel scheduling unit to generate preliminary results. Then, it calls the built-in lightweight real-time learning model and the built-in encryption algorithm to analyze the preliminary results and generate encryption risk warning data. The terminal control system receives encrypted risk warning data, decrypts and outputs non-sensitive warning information, transmits sensitive encrypted data to designated terminals according to preset permission management rules, completes warning feedback, and sends parameter adjustment signals to the adaptive multimodal data acquisition module based on the warning feedback results. The process involves analyzing the preliminary results using the built-in lightweight real-time learning model and the built-in encryption algorithm to generate encryption risk warning data, including: The edge intelligent processing module reads the preliminary results generated by the parallel scheduling unit and simultaneously retrieves historical environmental risk processing data for the corresponding monitoring area from the local storage unit built into the edge intelligent processing module. The preliminary results and historical environmental risk processing data are then input into the built-in lightweight real-time learning model. The historical environmental risk processing data includes basic information and historical similar risk cases categorized and archived based on this basic information. The basic information includes image features, environmental parameters, and duration of the risk at the time of its occurrence. The first output branch of the lightweight real-time learning model is used to identify the type of environmental risk. The environmental risk type includes one or more of the following: dust, smoke, fire, abnormal water color, and illegal dumping of solid waste. The second output branch of the lightweight real-time learning model is used to determine the risk level based on the quantitative indicators of the risk's scale, concentration, intensity, and duration. The lightweight real-time learning model corrects the risk categories and calibrates the risk levels in the preliminary results by comparing and analyzing the processing results and actual impact of similar historical risk cases, thus obtaining the corrected risk categories and calibrated risk levels.

2. The environmental risk AI visual early warning terminal as described in claim 1, characterized in that, When acquiring wide-area environmental image data through the wide-area field-of-view mode of the dynamic zoom and wide-area field-of-view switching lens in the adaptive multimodal data acquisition module, the following are included: The ambient light sensor built into the adaptive multimodal data acquisition module detects the light intensity data of the monitoring area at the first resolution and transmits the light intensity data to the terminal control system. The terminal control system determines whether the current lighting conditions are suitable for the image acquisition requirements of the dynamic zoom and wide field of view switching lens based on the received light intensity data, and sends an exposure parameter adjustment signal to the adaptive multimodal data acquisition module. The adaptive multimodal data acquisition module adjusts the exposure time of the dynamic zoom and wide field-of-view switching lens based on the received exposure parameter adjustment signal; the dynamic zoom and wide field-of-view switching lens is an integrated motor-driven optical lens that electronically switches between wide-angle fisheye mode and telephoto zoom mode; the wide-angle fisheye mode corresponds to the wide field-of-view mode; the telephoto zoom mode corresponds to the dynamic zoom mode; After the exposure time is adjusted, the adaptive multimodal data acquisition module controls the dynamic zoom and wide field of view switching lens to switch to wide field of view mode, and acquires wide-area environmental image data of the monitoring area through the wide field of view mode.

3. The environmental risk AI visual early warning terminal as described in claim 2, characterized in that, After receiving the exposure parameter adjustment signal sent by the terminal control system, the adaptive multimodal data acquisition module also includes: Analyze the adjustment parameters in the exposure parameter adjustment signal; If the exposure parameter adjustment signal also carries the monitoring frequency adjustment parameter, then the acquisition interval time of the dynamic zoom and wide field of view switching lens and the acquisition cycle of the multimodal sensor are adjusted according to the monitoring frequency adjustment parameter. If the exposure parameter adjustment signal also carries the field of view adjustment parameter, then the wide field of view angle of the dynamic zoom and wide field of view switching lens in wide field of view mode is adjusted according to the field of view adjustment parameter. After the parameters are adjusted, the adaptive multimodal data acquisition module sends a confirmation signal that the parameter adjustment is complete to the terminal control system.

4. The environmental risk AI visual early warning terminal as described in claim 3, characterized in that, When the terminal control system performs preliminary analysis of wide-area environmental image data, it includes: The terminal control system receives wide-area environmental image data transmitted by the adaptive multimodal data acquisition module, and performs noise reduction processing on the wide-area environmental image data by calling the built-in image noise reduction algorithm to obtain the noise-reduced wide-area environmental image data. The built-in image segmentation algorithm divides the denoised wide-area environmental image data into multiple sub-region image data, with each sub-region image data corresponding to a local area of ​​the monitoring area. The terminal control system retrieves a preset stored normal environment image template, compares the image data of each sub-region obtained by segmentation with the corresponding sub-region template in the preset normal environment image template, and calculates the difference value of each sub-region image data and sub-region template in color, texture and contour features; the normal environment image template is the standard image data of the corresponding monitoring area under the state of no environmental risk, and contains the normal visual features of each sub-region. The terminal control system compares the difference value of the image data of each sub-region with a preset threshold. If the difference value of the image data of any sub-region exceeds the preset threshold, it determines that there is an anomaly in the corresponding sub-region and marks the corresponding monitoring area as a suspected risk source area. The preset threshold is set according to historical monitoring data and is used to determine whether there is a critical value for anomalies.

5. The environmental risk AI visual early warning terminal as described in claim 4, characterized in that, Sending a control signal to the dynamic zoom and wide field-of-view switching lens to switch it to dynamic zoom mode and acquire high-definition detail image data includes: After completing the preliminary analysis of the wide-area environmental image data, the terminal control system acquires the coordinate information of the marked suspected risk source areas in the wide-area environmental image data. It then converts the coordinate information into the lens's focus position parameters based on the optical parameters of the dynamic zoom and wide-field-of-view switching lens. Simultaneously, it combines the size information of the suspected risk source areas in the wide-area environmental image data to determine the lens zoom magnification parameters covering the suspected risk source areas. Finally, it integrates the focus position parameters and the lens zoom magnification parameters into a control signal and sends it to the dynamic zoom and wide-field-of-view switching lens. The optical parameters include the lens focal length and the corresponding field of view. After receiving the control signal, the dynamic zoom and wide field of view switching lens analyzes the focus position parameter in the control signal, drives the focus motor in the lens to adjust the lens position to the corresponding focus position, analyzes the zoom magnification parameter, drives the zoom motor to adjust the lens spacing to the specified zoom magnification, and completes the switching from wide field of view mode to dynamic zoom mode. The dynamic zoom and wide field-of-view switching lens activates the built-in image sensor to acquire high-definition detailed image data of the suspected risk source area at a second resolution; the second resolution is higher than the first resolution.

6. The environmental risk AI visual early warning terminal as described in claim 5, characterized in that, The process of generating the preliminary results includes: The edge intelligent processing module receives high-definition detailed image data and multimodal data transmitted from the adaptive multimodal data acquisition module, performs preprocessing, and encapsulates them into data transmission packets; The parallel scheduling unit parses the data transmission packet, separates the high-definition detail image data after format conversion and the normalized multimodal data, and calls the preset task allocation rules to allocate the feature extraction task in the high-definition detail image data to the GPU, the numerical calculation task in the multimodal data to the CPU, and the model inference task for environmental risk identification to the NPU, and generates task allocation instructions and sends them to the corresponding processors. After receiving the feature extraction task, the GPU calls the built-in convolutional neural network model to extract features from the high-definition detail image data after format conversion, and obtains the first image feature vector. After receiving the numerical calculation task, the CPU performs statistical analysis on the normalized multimodal data, calculates the difference between temperature and humidity and normal range, the value of excessive gas component concentration, and the decibel abnormal value of ambient sound, and obtains the data statistical results. After receiving the model inference task, the NPU loads a pre-trained environmental risk identification model and performs real-time target detection and recognition on high-definition detailed image data, generating a second identification result that includes target category and confidence level; the target categories include sewage discharge and garbage accumulation. The parallel scheduling unit collects the first image feature vector, data statistics results, and second identification results, and integrates them through a data fusion algorithm to generate preliminary results containing suspected risk features, abnormal environmental parameter values, and risk category confidence.

7. The environmental risk AI visual early warning terminal as described in claim 6, characterized in that, The step of calling the built-in lightweight real-time learning model and built-in encryption algorithm to analyze the preliminary results and generate encryption risk warning data also includes: The edge intelligent processing module integrates the corrected risk category, the calibrated risk level with the corresponding high-definition detailed image data fragments and multimodal data key indicators, and adds analysis timestamps and edge intelligent processing module identification information to form a structured risk analysis result, which is then stored in the local storage unit of the edge intelligent processing module. The edge intelligent processing module classifies the risk analysis results, identifies sensitive data in the risk analysis results, and then calls the built-in AES-256 encryption algorithm to generate a random encryption key. The random encryption key is then used to encrypt the sensitive data to obtain an encrypted sensitive data block. The sensitive data includes the geographical coordinates of the monitoring area, the number of the production equipment inside the enterprise, and the concentration data of the sensitive gas components. Non-sensitive data and risk overview information in the risk analysis results are kept in plaintext to obtain plaintext data; the non-sensitive data includes risk type names and public area warning information. The edge intelligent processing module combines the encrypted sensitive data block with plaintext data to generate encrypted risk warning data.

8. The environmental risk AI visual early warning terminal as described in claim 7, characterized in that, The transmission of sensitive encrypted data to a designated terminal according to preset permission management rules, and the completion of early warning feedback, include: The terminal control system receives encrypted risk warning data transmitted by the edge intelligent processing module and extracts sensitive encrypted data from it; the encrypted risk warning data includes encrypted sensitive data blocks and plaintext data; The terminal control system reads the preset permission management rules stored locally, filters out the list of specified terminals with access rights to the current sensitive encrypted data according to the permission management rules, and determines the permission level corresponding to each specified terminal; the permission management rules include sensitive encrypted data access permission standards for different levels of users and terminals; The terminal control system retrieves the encryption key corresponding to the current sensitive encrypted data from the key management unit built into the edge intelligent processing module, and transmits the encryption key to each terminal in the designated terminal list through an encrypted wireless transmission link. At the same time, the sensitive encrypted data is synchronously transmitted to the corresponding designated terminal. The key management unit is used to store the random encryption key generated by the AES-256 encryption algorithm. After receiving the encryption key and sensitive encrypted data, each designated terminal uses the encryption key to decrypt the sensitive encrypted data and obtain the complete sensitive information. The terminal control system receives early warning information confirmation instructions from the local monitoring center terminal and various designated terminals, summarizes and integrates all the feedback early warning information confirmation instructions to form a structured early warning feedback result, and stores it in the feedback result database of the terminal control system to complete the early warning feedback; the early warning information confirmation instruction includes the reception status of the early warning information and preliminary processing opinions.

9. The environmental risk AI visual early warning terminal as described in claim 8, characterized in that, The terminal control system sends parameter adjustment signals to the adaptive multimodal data acquisition module based on the early warning feedback results, including: The terminal control system retrieves the stored early warning feedback results from the feedback result database and analyzes the preliminary processing opinions in the early warning feedback results; If the preliminary handling opinion indicates that the monitoring frequency of suspected risk source areas should be increased, then the monitoring frequency adjustment parameters should be determined based on the current monitoring interval of the suspected risk source areas. If the initial handling opinion indicates that the monitoring range should be expanded, then the field of view adjustment parameters should be determined based on the wide field of view angle range of the dynamic zoom and wide field of view switching lens. The terminal control system converts the determined monitoring frequency adjustment parameters or field of view adjustment parameters according to the parameter format recognized by the adaptive multimodal data acquisition module, generates corresponding parameter signals, encapsulates the parameter signals into parameter adjustment signals, and sends them to the adaptive multimodal data acquisition module.

Images