Industrial Internet of Things (IIoT) Full Lifecycle Environmental Impact Situation Identification Method and System

Abstract

This invention discloses a method and system for identifying the environmental impact situation throughout the entire lifecycle of an industrial Internet of Things (IoT) device. Relating to the field of data processing technology, the method includes: acquiring a current image data sequence, acquiring a historical image data sequence, performing target monitoring, generating historical target monitoring bounding boxes, and generating a current target monitoring bounding box; acquiring historical bounding box change trend values, acquiring current bounding box change trend values, and acquiring the absolute difference of the environmental monitoring object; if the absolute difference exceeds a preset threshold, cropping the current target monitoring bounding box according to its coordinate parameters to obtain the region of interest for the environmental monitoring object; performing semantic segmentation on the region of interest to obtain the segmentation result of the environmental monitoring object, and obtaining the monitoring and identification result for the environmental monitoring object based on the segmentation result. This invention has the advantages of reliable target identification, good monitoring effect, and high monitoring efficiency.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, specifically to a method and system for identifying the environmental impact situation throughout the entire lifecycle of the Industrial Internet of Things. Background Technology

[0002] Throughout the entire lifecycle of IoT engineering construction and operation, environmental protection monitoring faces severe challenges. Existing environmental protection monitoring and identification methods mainly rely on manual inspections or static image analysis, making it difficult to dynamically capture real-time changes in diverse environmental objects such as IoT device foundations, roads, construction waste sites, tensioning sites, vegetation, and residential areas.

[0003] Specifically, existing technologies typically only target a single time-series image, ignoring the temporal correlation between historical and current data. This results in insufficient sensitivity to gradual environmental damage (such as vegetation degradation and slow expansion of construction waste sites). Furthermore, changes in bounding box size are easily affected by noise from lighting and angles, and conventional algorithms cannot effectively distinguish between natural fluctuations and anomalous changes, leading to a high false alarm rate. Finally, subsequent analysis of anomalous areas lacks refined processing methods. Directly performing semantic segmentation on the entire image is not only computationally redundant but also susceptible to background interference, making it difficult to accurately identify key environmental issues (such as a significant reduction in vegetation cover, soil erosion, and encroachment on ecological vegetation by engineering projects). Summary of the Invention

[0004] To address the technical problem that existing technologies cannot achieve early identification and accurate positioning of abnormal changes in environmental objects in Internet of Things (IoT) projects, this invention provides a method and system for identifying the environmental impact situation throughout the entire life cycle of industrial IoT.

[0005] A method for identifying the environmental impact situation throughout the entire lifecycle of an industrial Internet of Things (IoT) includes: acquiring the current image data sequence of a monitored area in an IoT project during the current identification period, and acquiring the historical image data sequence of the monitored area in the IoT project during the previous historical identification period, wherein the monitored area includes multiple environmental monitoring objects; performing target monitoring on each historical image data in the historical image data sequence and each current image data in the current image data sequence, generating historical target monitoring bounding boxes for each environmental monitoring object in each historical image data, and generating current target monitoring bounding boxes for each environmental monitoring object in each current image data; and obtaining the historical bounding box change of the j-th environmental monitoring object based on the historical target monitoring bounding boxes of multiple j-th environmental monitoring objects. The system calculates the trend value of the current bounding box of the j-th environmental monitoring object, and obtains the trend value of the current bounding box change of the j-th environmental monitoring object based on the current target bounding box of multiple objects. It also calculates the absolute difference between the historical and current bounding box trend values ​​of the j-th environmental monitoring object. If the absolute difference of the j-th environmental monitoring object exceeds a preset threshold, the system crops the current target bounding box of the j-th environmental monitoring object based on the coordinate parameters of the current image data closest to the current moment, thereby obtaining the region of interest for the j-th environmental monitoring object. Semantic segmentation is then performed on the region of interest of the j-th environmental monitoring object to obtain the segmentation result, and the monitoring and identification result for the j-th environmental monitoring object is obtained based on the segmentation result.

[0006] Optionally, target detection for each historical image data in the historical image data sequence and each current image data in the current image data sequence includes: target detection for each historical image data in the historical image data sequence and each current image data in the current image data sequence based on the Yo1ov7 algorithm, wherein the Yo1ov7 algorithm includes an input layer, a backbone network, a neck network, and an output layer.

[0007] Optionally, obtaining the historical boundary box change trend value of the j-th environmental monitoring object based on the historical target monitoring boundary boxes of multiple j-th environmental monitoring objects includes: obtaining the area within the historical boundary boxes of multiple j-th environmental monitoring objects based on the historical target monitoring boundary boxes of multiple j-th environmental monitoring objects; obtaining the historical area change ratio of the j-th environmental monitoring object according to the sequential order of each historical image data in the historical image data sequence, based on the area within the historical boundary box of the j-th environmental monitoring object in the later historical image data and the area within the historical boundary box of the j-th environmental monitoring object in the previous historical image data; determining the validity of each historical area change ratio and obtaining the valid historical area change ratio, and obtaining the historical boundary box change trend value of the j-th environmental monitoring object based on each valid historical area change ratio of the j-th environmental monitoring object.

[0008] Optionally, obtaining the current boundary box change trend value of the j-th environmental monitoring object based on the current target monitoring boundary boxes of multiple j-th environmental monitoring objects includes: obtaining the area within the current boundary box of the j-th environmental monitoring object in each current image data according to the current target monitoring boundary boxes of multiple j-th environmental monitoring objects; obtaining the current area change ratio of the j-th environmental monitoring object according to the sequential order of each current image data in the current image data sequence, based on the area within the current boundary box of the j-th environmental monitoring object in the next current image data and the area within the current boundary box of the j-th environmental monitoring object in the previous current image data; determining the validity of each current area change ratio and obtaining the valid current area change ratio, and obtaining the current boundary box change trend value of the j-th environmental monitoring object based on each valid current area change ratio of the j-th environmental monitoring object.

[0009] Optionally, obtaining the absolute difference of the j-th environmental monitoring object based on the historical boundary box change trend value and the current boundary box change trend value includes: subtracting the current boundary box change trend value from the historical boundary box change trend value to obtain the relative difference value, and taking the absolute value of the relative difference value as the absolute difference value.

[0010] Optionally, semantic segmentation of the region of interest for the j-th environmental monitoring object includes: using the DeepLab algorithm to perform semantic segmentation of the region of interest for the j-th environmental monitoring object, wherein the DeepLab algorithm includes an encoder and a decoder, and the encoder includes a deep convolutional neural network and spatial pyramid pooling with dilated convolution.

[0011] An industrial IoT full lifecycle environmental impact situation identification system is also provided. The system includes a management platform, a sensor network platform, and an object platform connected sequentially. The management platform includes: a data acquisition module, used to acquire the current image data sequence of the area to be monitored in the IoT project during the current identification period, and to acquire the historical image data sequence of the area to be monitored in the IoT project during the previous historical identification period, wherein the area to be monitored includes multiple environmental monitoring objects; a target monitoring module, used to perform target monitoring on each historical image data in the historical image data sequence and each current image data in the current image data sequence, generating historical target monitoring bounding boxes for each environmental monitoring object in each historical image data, and generating current target monitoring bounding boxes for each environmental monitoring object in each current image data; and a first data processing module, used to process data based on the historical data of multiple j-th environmental monitoring objects. The system acquires the historical bounding box change trend value of the j-th environmental monitoring object, and obtains the current bounding box change trend value of the j-th environmental monitoring object based on multiple current target monitoring bounding boxes of the j-th environmental monitoring object. It also obtains the absolute difference between the historical and current bounding box change trends. A second data processing module, when the absolute difference of the j-th environmental monitoring object exceeds a preset threshold, crops the current target monitoring bounding box of the j-th environmental monitoring object based on the coordinate parameters of the current target monitoring bounding box in the current image data, thereby obtaining the region of interest for the j-th environmental monitoring object. A segmentation and monitoring recognition module performs semantic segmentation on the region of interest of the j-th environmental monitoring object to obtain the segmentation result of the j-th environmental monitoring object, and obtains the monitoring recognition result for the j-th environmental monitoring object based on the segmentation result.

[0012] Optionally, the target detection module is also used to: perform target detection on each historical image data in the historical image data sequence and each current image data in the current image data sequence based on the Yo1ov7 algorithm, wherein the Yo1ov7 algorithm includes an input layer, a backbone network, a neck network and an output layer.

[0013] Optionally, the first data processing module is further configured to: obtain the area within the historical boundary frame of the j-th environmental monitoring object based on the historical target monitoring boundary frames of the j-th environmental monitoring object; obtain the historical area change ratio of the j-th environmental monitoring object according to the sequential order of each historical image data in the historical image data sequence, based on the area within the historical boundary frame of the j-th environmental monitoring object in the next historical image data and the area within the historical boundary frame of the j-th environmental monitoring object in the previous historical image data; determine the validity of each historical area change ratio and obtain the valid historical area change ratio, and obtain the historical boundary frame change trend value of the j-th environmental monitoring object based on each valid historical area change ratio of the j-th environmental monitoring object.

[0014] Optionally, the first data processing module is further configured to: obtain the area within the current bounding box of the j-th environmental monitoring object in each current image data according to the current target monitoring bounding boxes of multiple j-th environmental monitoring objects; obtain the current area change ratio of the j-th environmental monitoring object according to the order of each current image data in the current image data sequence, based on the area within the current bounding box of the j-th environmental monitoring object in the next current image data and the area within the current bounding box of the j-th environmental monitoring object in the previous current image data; determine the validity of each current area change ratio and obtain the valid current area change ratio, and obtain the current bounding box change trend value of the j-th environmental monitoring object based on each valid current area change ratio of the j-th environmental monitoring object.

[0015] The beneficial effects of this invention are reflected in: In the entire industrial IoT lifecycle environmental impact situation identification method, by comparing the complete image sequences of the historical and current identification periods, and utilizing the continuous bounding box data generated by target monitoring, the method calculates the historical bounding box change trend value reflecting the baseline change pattern and the current bounding box change trend value capturing the latest dynamics. Based on the absolute difference, it identifies abnormal objects that deviate from the norm, enhancing the perception sensitivity of gradual environmental risks such as the slow expansion of waste disposal sites and the continuous degradation of vegetation, overcoming the lag of single-point image analysis. Furthermore, the method employs dual filtering of the bounding box area change ratio. On the one hand, by eliminating large fluctuations caused by noise such as illumination and angle between adjacent frames, it effectively distinguishes between real environmental anomalies and imaging interference. On the other hand, it uses temporal trend differences rather than jumps as the judgment criterion, avoiding misjudgments due to occasional noise and significantly improving monitoring reliability. Furthermore, for abnormal candidate objects with absolute differences exceeding a preset threshold, the method accurately crops the region of interest based on its latest bounding box coordinates, focusing computational resources on the micro-environmental details of the object, reducing the computational load of semantic segmentation by several orders of magnitude, and completely avoiding interference from complex backgrounds in full-image analysis, thereby supporting actionable environmental protection decisions. Attached Figure Description

[0016] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0017] Figure 1 This is a schematic diagram illustrating the steps of the industrial Internet of Things (IoT) full lifecycle environmental impact situation identification method of the present invention; Figure 2 This is a schematic diagram of part of step S2 in the Industrial Internet of Things Full Lifecycle Environmental Impact Situation Identification Method of the present invention; Figure 3 This is a schematic diagram of another part of the steps in S2 of the Industrial Internet of Things full life cycle environmental impact situation identification method of the present invention; Figure 4 This is a partial flowchart of steps S1 to S2 in the Industrial Internet of Things Full Lifecycle Environmental Impact Situation Identification Method of the present invention; Figure 5 This is a partial flowchart of steps S2 to S5 in the Industrial Internet of Things Full Lifecycle Environmental Impact Situation Identification Method of the present invention; Figure 6 This is a partial flowchart of S5 in the Industrial Internet of Things Full Lifecycle Environmental Impact Situation Identification Method of the present invention; Figure 7 This is a schematic diagram of the composition of the Industrial Internet of Things (IIoT) full lifecycle environmental impact situation identification system of the present invention; Figure 8 This is a schematic diagram of the optimized industrial Internet of Things (IoT) involved in this invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0019] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0020] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, the terms "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0021] like Figure 1 , Figure 4 , Figure 5 and Figure 6 As shown, a method for identifying the environmental impact situation throughout the entire lifecycle of an industrial Internet of Things (IoT) is provided. In one embodiment, the method includes: S1. Obtain the current image data sequence of the area to be monitored in the IoT project during the current identification period, and obtain the historical image data sequence of the area to be monitored in the IoT project during the previous historical identification period, wherein the area to be monitored includes multiple environmental monitoring objects. S2. Perform target monitoring on each historical image data in the historical image data sequence and each current image data in the current image data sequence, generate historical target monitoring bounding boxes for each environmental monitoring object in each historical image data, and generate current target monitoring bounding boxes for each environmental monitoring object in each current image data. S3. Obtain the historical boundary box change trend value of the j-th environmental monitoring object based on the historical target monitoring boundary boxes of multiple j-th environmental monitoring objects, obtain the current boundary box change trend value of the j-th environmental monitoring object based on the current target monitoring boundary boxes of multiple j-th environmental monitoring objects, and obtain the absolute difference of the j-th environmental monitoring object based on the historical boundary box change trend value and the current boundary box change trend value. S4. If the absolute difference of the j-th environmental monitoring object exceeds the preset threshold, the current target monitoring bounding box of the j-th environmental monitoring object is cropped according to the coordinate parameters of the current target monitoring bounding box of the j-th environmental monitoring object in the current image data, so as to obtain the region of interest of the j-th environmental monitoring object. S5. Perform semantic segmentation on the region of interest of the j-th environmental monitoring object to obtain the segmentation result of the j-th environmental monitoring object, and obtain the monitoring and identification result of the j-th environmental monitoring object based on the segmentation result.

[0022] In this embodiment, it should be noted that in S1, image sequences of the same monitored area are acquired simultaneously for two identification periods (the current identification period and the historical identification period). Specifically, it is necessary to acquire continuous images taken within the area covered by the target IoT project during the current identification and analysis period (referred to as the current identification period) to form the current image data sequence. Simultaneously, it is also necessary to retrospectively acquire continuous images taken within the exact same geographical area (i.e., the monitored area) during the identification period immediately preceding the current identification period to form the historical image data sequence. This ensures that the comparative analysis is conducted at the same spatial location. The key point is that what is acquired is not a single photograph, but a continuous series of images (i.e., a sequence) within each time period, which provides raw data support for subsequently capturing the temporal change patterns of the target. For example, the current image data sequence may come from drone patrols taken once a day within the past week, while the historical image data sequence may be archived images taken daily at the same time and along the same flight path for the corresponding seven days a month ago, covering the same project area.

[0023] Furthermore, the definition of the area to be monitored must clearly include all key environmental monitoring objects that may be involved in the construction and operation of the IoT project. These objects are not randomly selected, but strictly focus on the core elements of the environmental impact of the IoT project. In the example, they are explicitly listed as the foundations of various IoT facilities in the IoT project (such as the foundation area of ​​signal towers), roads (such as construction access roads or maintenance channels), spoil heaps (such as waste disposal sites), tensioning sites (such as the work site for laying conductor tension), vegetation (such as the original woodland, grassland or later restored green belts along the IoT project route), and residential areas (such as affected villages or houses near the IoT project).

[0024] Furthermore, the historical and current image data sequences acquired by S1 must be spatially large and precisely aligned to ensure that, over time, each category of environmental protection object (e.g., the specific waste dump, the construction road, the vegetation around the tower, or nearby farmhouses) exists simultaneously in images from both time periods, and its location is traceable across different images in the sequence. Only by ensuring the consistency of the monitoring area and the existence and comparability of the core environmental protection objects during data acquisition can subsequent steps such as monitoring historical trends, calculating boundary differences, and cropping areas of interest for analysis be truly meaningful. For example, the designated monitoring area should cover the area extending from the tower base to the surrounding potential waste impact zone, and include adjacent vegetated slopes and residential boundaries.

[0025] In S2, target monitoring is performed on the acquired time-series image sequences to locate and define each environmental monitoring object. Specifically, the historical image data sequences and current image data sequences obtained in S1 need to be processed separately. For each historical image and each current image in the sequence, a target monitoring algorithm (such as YOLOv7) can be used. YOLOv7 is a deep learning model whose structure typically includes an input layer receiving images, a backbone network extracting deep features, a neck network performing multi-scale feature fusion, and finally, an output layer predicting the location and category of the object. The output of S2 is to generate a bounding box for each environmental monitoring object (such as a specific tower base, a section of road, or a specific patch of vegetation) in each image data of each identification period sequence. This bounding box accurately marks the location and coverage area of ​​the object in the corresponding image data. Importantly, this full-coverage target monitoring is performed for both the historical and current sequences, resulting in two complete sets of results: one set of historical target monitoring bounding boxes for each object in each historical image, and the other set of current target monitoring bounding boxes for each object in each current image.

[0026] Furthermore, target monitoring is not performed on a single representative image, but rather applied to the entire time-series image sequence (e.g., daily historical images from the past 7 days and daily current images from the last 7 days). This means that for each specific environmental object (e.g., the j-th environmental monitoring object named "slag heap southwest of Tower 3"), a set of bounding boxes that change over time will be generated in the historical sequence (each box representing the area it was located in images at different historical time points), and similarly, a set of bounding boxes that change with the latest time point will be generated in the current sequence (representing its location at different current time points). Generating serialized bounding boxes for each object is the basis for subsequent analysis (S3) of its size change trends. For example, when processing the image sequence of the current week, for the object "slag heap southwest of Tower 3", the algorithm needs to accurately locate its bounding box in the current images taken each day; at the same time, for the historical sequence images of the week a month ago, the algorithm also needs to retrospectively generate the corresponding bounding box for it in the historical images of each day. In this way, the location and size change trajectory of the slag heap over the past month has a quantifiable and comparable data basis. Once the bounding box marks the location of the object, its size and area are accurately recorded, providing crucial input for subsequent judgment on whether it is a natural fluctuation (such as the influence of light on boundary delineation) or a real abnormal expansion (such as the continuous dumping of construction waste leading to an increase in land occupation).

[0027] In S3, the overall change pattern (trend) of the bounding box size of a specific environmental monitoring object is quantified and compared over two different time periods (historical period and current period). For each independent environmental object marked as j in the sequence (such as a specific waste dump A, a specific vegetation area B, a specific road section C, etc.), it is necessary to analyze its bounding box change behavior in the historical and current identification periods. Specifically, firstly, based on the bounding box of the object monitored in each historical image data in the historical image data sequence (the historical target monitoring bounding box generated in S2), its bounding box area (reflecting the coverage of the object on the image) is accurately calculated. Then, the relative change ratio of the bounding box area between two adjacent image data (i.e., the percentage increase or decrease) is analyzed in chronological order.

[0028] Furthermore, not all change ratios are directly adopted; a validity judgment process is required to filter out large area fluctuations that are likely caused by transient noise (such as differences in lighting, slight shifts in shooting angle, short-term occlusion, etc.), as these usually do not represent the true environmental changes of the object. Only change ratios between adjacent image data with a change amplitude below the maximum fluctuation range (indicating that they may be real changes) are considered valid and included in the calculation. Finally, these valid historical area change ratios are aggregated to form a single historical bounding box change trend value. This value represents the overall measure of the rate and direction of the object's bounding box size change over the entire historical period (positive values ​​indicate an expanding trend, and negative values ​​indicate a contracting trend). The same logic and process need to be applied to the current image data sequence, calculating the bounding box area of ​​the object in each current image data in the current sequence, analyzing the current area change ratios of adjacent data, performing validity judgments to filter out reliable changes, and then aggregating them to obtain a current bounding box change trend value.

[0029] Furthermore, after obtaining the historical and current bounding box trend values ​​for the same object j, the next step is to calculate the absolute difference between these two trend values. This absolute difference is calculated by taking the absolute value of the relative difference between the two objects, and then taking the absolute value of this relative difference as the final absolute difference. Essentially, it compares the overall bounding box change rate of the object in the current recognition period with its change rate considered a baseline or normal trend in historical recognition periods. If this absolute difference is large, it means that the object's behavior in the current period (whether it's increased expansion or accelerated contraction) deviates significantly from its pattern in past periods, which usually indicates an unusual or abnormal environmental event. This comparison based on the overall trend of the time series, rather than the abrupt changes in a single image, significantly improves the monitoring sensitivity for slowly evolving, persistent environmental problems and effectively reduces false alarms caused by noise interference.

[0030] In S4, the absolute difference between the historical bounding box trend value and the current bounding box trend value for each environmental object j is calculated and compared with a preset threshold. The preset threshold is determined by calculating the standard deviation of a finite number of historical statistical trend values ​​for all objects, and then taking 2 to 3 times that standard deviation as the preset threshold. Only when this absolute difference exceeds the preset threshold does it mean that the object's overall change behavior (such as the speed and extent of expansion or contraction) within the current identification period significantly deviates from its baseline pattern in the historical period, suggesting a potentially noteworthy abnormal environmental event (such as accelerated engineering encroachment, rapid expansion of waste disposal sites, or sudden vegetation degradation). At this point, a subsequent refined analysis process will be initiated for object j. The key prerequisite for refined analysis is the accurate extraction in S4 of a candidate object j containing only this abnormality. This extraction process is not arbitrary but strictly relies on the results of the target monitoring stage, using the current target monitoring bounding box located in the latest image data (usually the image data closest to the current moment) of the current image data sequence, as determined in S2. This bounding box provides the precise coordinate range of object j in the image. Based on these coordinate parameters, the corresponding region of the image data is directly cropped, and the resulting cropped area is the region of interest for the object. This is a highly focused small image fragment, roughly equivalent in size to the bounding box of the most recently detected object.

[0031] Furthermore, by focusing on the region of interest, the computational burden of blindly performing semantic segmentation on a massive original panoramic image is avoided, enabling a precise strategy of focusing analysis on areas of anomaly. Through cropping, S4 strictly limits the region of interest for subsequent computation (S5 semantic segmentation) to the specific object j suspected of having a problem and its immediate vicinity. This not only significantly reduces the total number of pixels to be processed (improving computational efficiency), but more importantly, it significantly reduces the interference of background noise. In panoramic images, the boundaries of objects are often mixed with a large amount of complex background information, including a large amount of irrelevant terrain, vegetation, or other engineering facilities. Direct segmentation on the entire image can easily cause serious confusion for the analysis target (such as confirming how much vegetation has been encroached upon by the waste disposal site, or whether there is bare ground within the vegetated area). However, within the cropped region of interest focused on object j, the semantic segmentation algorithm (S5) can devote all its attention and computational power to analyzing the internal details of that specific object. This focused analysis is the guarantee for accurate monitoring.

[0032] In S5, refined semantic segmentation is performed on the highly focused region of interest image generated in S4. Based on the segmentation results, the specific environmental elements and their changing characteristics within a particular environmental object j are quantitatively analyzed, ultimately leading to accurate conclusions regarding environmental problem monitoring and identification. For each object j identified by S3-S4 as having a significant abnormal trend deviation, computational resources are concentrated on its cropped region of interest image. A semantic segmentation algorithm (such as DeepLabv3+) is used, which can classify and label each pixel within the region, accurately distinguishing the categories of key environmental elements within the object (for example, in the region of interest image of a vegetated area, the algorithm will accurately label pixels such as "healthy vegetation," "bare soil," "soil erosion gullies," "artificially hardened surface," and "dead vegetation"; in a construction waste site, it will label "areas covered with green netting," "compacted construction waste," "loose construction waste," and "areas with exposed soil and water"). The DeepLab algorithm typically consists of a powerful encoder (often using a deep convolutional neural network such as ResNet, and integrating a spatial pyramid pooling module with dilated convolutions to effectively capture multi-scale features) responsible for extracting rich contextual information from the image; and a sophisticated decoder responsible for restoring spatial details and generating the final pixel-level classification label map, ensuring good segmentation accuracy for details such as boundaries and small objects.

[0033] Furthermore, after obtaining pixel-level fine segmentation results, the area proportion, shape changes, or spatial distribution patterns of specific subcategories can be quantitatively calculated, thereby achieving accurate identification of environmental issues. This process completely solves the problems of low efficiency and background interference caused by direct segmentation on the entire image. Based on these quantitative indicators and spatial distribution information, the specific key environmental issues currently existing in object j can be accurately determined. By comparing and analyzing the segmentation results with preset environmental requirements or historical states, S5 finally outputs quantifiable monitoring and identification results for object j that are specific to environmental issues, providing a direct basis for taking targeted environmental intervention measures.

[0034] In summary, the method for identifying the environmental impact situation throughout the entire lifecycle of the Industrial Internet of Things (IIoT) enhances the sensitivity to progressive environmental risks such as the slow expansion of construction waste dumps and the continuous degradation of vegetation by comparing complete image sequences from historical and current identification periods and utilizing continuous bounding box data generated by target monitoring. It calculates historical bounding box trend values ​​reflecting baseline change patterns and current bounding box trend values ​​capturing the latest dynamics, and identifies abnormal objects deviating from the norm based on absolute differences. This overcomes the lag in single-point image analysis. Furthermore, the dual filtering of the bounding box area change ratio, on the one hand, removes large fluctuations caused by noise such as illumination and angle between adjacent frames. Fluctuations effectively distinguish between real-world environmental anomalies and imaging interference. On the other hand, using temporal trend differences rather than jumps as the judgment criterion avoids misjudgments caused by occasional noise, significantly improving monitoring reliability. Furthermore, for abnormal candidate objects with absolute differences exceeding a preset threshold, the region of interest is precisely cropped based on its latest bounding box coordinates, focusing computational resources on the micro-environmental details of the object, reducing the computational load of semantic segmentation by several orders of magnitude, and completely avoiding interference from complex backgrounds in full-image analysis. This allows key environmental issues such as soil erosion gullies, vegetation degradation patches, and engineering encroachment boundaries to be quantitatively identified through pixel-level analysis, thereby supporting actionable environmental decisions.

[0035] In one implementation, target monitoring in S2 for each historical image data in the historical image data sequence and each current image data in the current image data sequence includes: The Yo1ov7 algorithm is used to perform target detection on each historical image data in the historical image data sequence and each current image data in the current image data sequence. The Yo1ov7 algorithm includes an input layer, a backbone network, a neck network, and an output layer.

[0036] In this embodiment, it should be noted that the main function of the input layer is to perform image data preprocessing. The purpose of using existing data augmentation methods is to increase the amount of data and improve the generalization performance of the model by mixing multiple images together to generate new images. In addition, data augmentation can also effectively avoid overfitting, thereby reducing the error of the model on the test set.

[0037] Furthermore, after preprocessing, the image data enters the backbone network to extract image features. The backbone network of Yo1ov7 is mainly divided into two parts: one is the CBS module, which consists of convolution, batch normalization, and the sliu activation function, with a total of 4 layers; the other is the ELAN module and the MP module, with a total of 7 layers.

[0038] The ELAN module is an efficient network architecture that enables the network to learn more features and has stronger robustness by controlling the shortest and longest gradient paths. ELAN has two branches. The first branch processes image features through a 1x1 convolution to increase the number of feature channels. The second branch first processes the features through a 1x1 convolution module, then through four 3x3 convolution modules, and finally combines the four features to obtain the final feature set.

[0039] The MP module has two branches, both responsible for downsampling features. The first branch performs max pooling followed by a 1x1 convolution to adjust the number of channels. The second branch first performs a 1x1 convolution to adjust the number of channels, then a 3x3 convolutional block with a stride of 2, which also performs downsampling. Finally, the results from the first and second branches are combined to obtain the final result.

[0040] Furthermore, among the image features extracted by the backbone network, high-level features have richer semantic information, while shallow features contain more location and detail information but less semantic information. In order to balance the two types of feature information, a top-up structure and a feature pyramid structure, i.e., a neck network, are used to complete the feature fusion structure processing, enrich the feature information, and improve the accuracy of monitoring and recognition.

[0041] Furthermore, after feature fusion processing, the image features are divided into three scales to monitor targets at large, medium, and small scales respectively. To accelerate inference, the output layer's processing structure mainly includes a reparameterized convolution (RepConv) module. The RepConv module consists of two phases: training and inference. The training phase includes three branches: the top branch is a 1x1 convolution for feature smoothing; the middle branch is a 3x3 convolutional block for feature extraction; and the bottom branch performs no operation and is directly added to the results of the previous two branches. The inference phase includes a 3x3 convolutional operation with a length of 1, whose weights are derived from reparameterization.

[0042] It should also be noted that, considering the large image size of UAV and satellite images, this study uses the high-resolution models Yolov7W6 and Yolov7E6 from the Yolov7 series to train models for environmental monitoring objects such as tower bases, roads, vegetation, spoil heaps, tensioning fields, and residential areas.

[0043] like Figure 2 As shown, in one embodiment, obtaining the historical boundary box change trend value of the j-th environmental monitoring object in S2 based on the historical target monitoring boundary boxes of multiple j-th environmental monitoring objects includes: S21. Obtain the area within the historical boundary frame of multiple j-th environmental monitoring objects based on the historical target monitoring boundary frames of multiple j-th environmental monitoring objects; S22. According to the sequential order of each historical image data in the historical image data sequence, obtain the historical area change ratio of the j-th environmental monitoring object based on the historical boundary area of ​​the j-th environmental monitoring object in the next historical image data and the historical boundary area of ​​the j-th environmental monitoring object in the previous historical image data. S23. Determine the validity of each historical area change ratio and obtain the valid historical area change ratio. Based on the valid historical area change ratios of each environmental monitoring object, obtain the historical boundary box change trend value of the j-th environmental monitoring object.

[0044] In this embodiment, it should be noted that in S21, the coordinate information of the historical target monitoring bounding box of each specific environmental protection object in each historical image data is precisely quantified, and the actual pixel area contained inside its bounding box (bounding box area) is accurately calculated. This is equivalent to obtaining the physical size of the object identified in the image at each historical time point (such as each day within the historical monitoring period). For example, for the "Southeast Stockpile of Tower No. 3" of the waste soil dump, the area inside its bounding box in the images of each day in the historical image sequence of the past 7 days will be calculated, thereby obtaining a set of historical size data sequences that change over time.

[0045] In S22, using the historical bounding box areas obtained in S21 arranged chronologically, the area changes between adjacent historical image data (i.e., two consecutive time points) are quantitatively analyzed. Following the chronological order of the historical image data sequence, the area within the bounding box of the j-th object is subtracted from the area within the bounding box of the object preceding it. This area difference is then divided by the area within the bounding box of the object preceding it, thus calculating the relative area change ratio between two adjacent time points (historical area change ratio). This ratio is a dimensionless indicator; a positive value indicates an increase in area (e.g., expansion of a spoil heap), and a negative value indicates a decrease in area (e.g., vegetation degradation and shrinkage). Its absolute value reflects the degree of drastic change. This step provides a crucial series of continuously changing values ​​for understanding how object dimensions fluctuate over time.

[0046] In S23, the validity of the historical area change ratios calculated in S22 at each adjacent time point is determined. The core purpose is to filter out accidental large-scale area fluctuations caused by transient noise (such as sudden changes in daily illumination, minor adjustments in shooting angle, and short-term cloud shadow occlusion). These fluctuations do not represent actual changes in environmental conditions (for example, a vegetation boundary box shrinking drastically by 50% due to strong shadows on a certain day). Only those ratios with changes less than the area change ratio threshold (i.e., insignificant changes) are considered valid historical area change ratios and retained. The area change ratio threshold can be obtained through statistical analysis based on historical data. Specifically, the absolute value of the area change ratio of all adjacent images in a limited number of normal historical time series data is first taken, and then its 99th percentile is taken as the area change ratio threshold.

[0047] Secondly, all historical area change ratios deemed valid are aggregated and calculated to generate a historical bounding box change trend value. This trend value comprehensively reflects the overall size change pattern of the object throughout the entire historical identification period: a stable positive value indicates that the object has historically had a tendency to continuously expand (such as the slow accumulation of spoil heaps), a stable negative value indicates that there has historically had continuous contraction pressure (such as the gradual thinning of slope vegetation), and a value close to zero represents that the size has been relatively stable historically.

[0048] It should also be noted that the expression for obtaining the historical bounding box change trend value can be expressed as: ;in, Let be the historical boundary value of the j-th environmental monitoring object. This represents the number of historical image data in the historical image data sequence. Let be the area within the historical bounding box of the j-th environmental monitoring object in the (i+1)-th historical image data. Let be the area within the historical bounding box of the j-th environmental monitoring object in the i-th historical image data. This is the threshold for the proportion of area change.

[0049] It should also be noted that throughout the entire expression, To calculate the relative change rate of the bounding box area of ​​the j-th environmental monitoring object for two adjacent historical image data (the i-th and i+1-th images) in the historical sequence, the numerator is used as the percentage change in area. The denominator represents the area difference (positive for expansion, negative for contraction), and the denominator represents the reference benchmark for the relative rate of change. It is the area within the historical bounding box of the j-th environmental monitoring object in the previous historical image data, and can also eliminate the influence of object size differences.

[0050] Furthermore, This reflects the process of effectively judging the proportion of historical area changes. Specifically, To calculate the absolute value of the change ratio, then compare it with the area change ratio threshold. Comparison. If , Output 0, otherwise output 1.

[0051] In other words, when the absolute value of the historical area change ratio is greater than or equal to , Output -1 or 0. After maxing out the output, output 0. This indicates that the historical area change ratio is noise (e.g., a sudden change in daily illumination causing an abnormal 30% expansion of the bounding box) and is not included in the statistics. When the absolute value of the historical area change ratio is less than... , Output 1. After maxing out the output, it is determined to be a reliable signal (such as normal daily vegetation shrinkage of 0.8%), and the original value is retained. In summary, transient noise (such as illumination and angular shift) can cause large-scale area fluctuations. Therefore, the expression is specifically designed to filter out large fluctuations (rather than small fluctuations).

[0052] Furthermore, This is a preset maximum natural fluctuation threshold for different types of environmental protection objects (vegetation / construction sites, etc.). It can be obtained not only through statistical analysis of historical data but also by setting it based on the differences in object type. When setting it based on object type, it should not be set too low to avoid failing to filter abnormal noise. For example, vegetated areas have higher solar radiation stability... (i.e., a daily change exceeding 20% ​​is considered noise); the spoil heap is affected by mechanical operations. (Tolerating higher volatility).

[0053] Furthermore, the final result is generated by accumulating the effective historical area change ratios. By eliminating instantaneous interference and quantifying gradual trends, the system particularly highlights slow environmental damage. In summary, the entire expression, through its architecture of zeroing out large fluctuations and accumulating effective signals, transforms the goal of filtering out large fluctuations caused by instantaneous noise into calculable logical steps, laying the foundation for scientific and accurate identification. By directly addressing the essence of noise in the environmental impact situational awareness of the entire lifecycle of the Industrial Internet of Things (IIoT) as sudden and drastic jumps, it ultimately solves the problems of high false alarm rates and insensitivity to gradual changes.

[0054] like Figure 3 As shown, in one embodiment, obtaining the current boundary box change trend value of the j-th environmental monitoring object in S2 based on the current target monitoring boundary boxes of multiple j-th environmental monitoring objects includes: S24. Based on the current target monitoring bounding boxes of multiple j-th environmental monitoring objects, obtain the area within the current bounding box of the j-th environmental monitoring object in each current image data; S25. According to the order of each current image data in the current image data sequence, obtain the current area change ratio of the j-th environmental monitoring object based on the current bounding box area of ​​the j-th environmental monitoring object in the next current image data and the current bounding box area of ​​the j-th environmental monitoring object in the previous current image data. S26. Determine the validity of each current area change ratio and obtain the valid current area change ratio. Based on the valid current area change ratios of the j-th environmental monitoring object, obtain the current boundary box change trend value of the j-th environmental monitoring object.

[0055] In this embodiment, it should be noted that S24 is similar to S21. In S24, the calculation of the area within the current bounding box is the basis of the current cycle trend analysis. The core task is to establish a size benchmark for each environmental protection object in each current image data. That is, for the j-th object (such as a vegetation area or a slag dump with a specific number), it is necessary to accurately extract the actual coverage of its bounding box in each current image data in the current image data sequence.

[0056] Similar to S22, S25 reveals the dynamic evolution of environmental objects within the latest monitoring cycle. It strictly calculates the area change within the current bounding box of adjacent current image data according to time sequence (increasing date) to ensure trend continuity. Using the formula (later area - previous area) / previous area, it transforms the absolute size difference into a dimensionless rate of change. Positive values ​​indicate expansion trends (such as soil accumulation), while negative values ​​indicate shrinkage signals (such as vegetation degradation). This approach captures subtle, continuous changes (such as a road construction area expanding by an average of 0.4% per day) while avoiding the bottleneck of comparing objects of different sizes (such as placing large soil dumps and small vegetated areas within a unified quantitative system).

[0057] Similar to S23, S26 aggregates the current bounding box change trend values. First, it determines validity, identifying noise events and valid signals. The change proportions of valid signals are retained and accumulated to form a quantifiable current trend value.

[0058] It should also be noted that the expression for obtaining the current bounding box change trend value can be expressed as: ;in, This represents the current bounding box change trend value of the j-th environmental monitoring object. This represents the number of current image data points in the current image data sequence. Let be the area within the current bounding box of the j-th environmental monitoring object in the (i+1)-th current image data. Let be the area within the current bounding box of the j-th environmental monitoring object in the i-th current image data. This is the threshold for the proportion of area change.

[0059] It should also be noted that the expression for obtaining the current bounding box change trend value is similar to the expression for obtaining the historical bounding box change trend value. The only difference is the data substituted. Therefore, the calculation logic, the technical problem solved, and the technical means to solve the technical problem in the expression in this embodiment have been fully explained in the foregoing embodiments and will not be elaborated on here.

[0060] In one implementation, obtaining the absolute difference between the historical boundary box change trend value and the current boundary box change trend value in S3 includes: Subtract the current bounding box change trend value from the historical bounding box change trend value to obtain the relative difference value, and take the absolute value of the relative difference value as the absolute difference value.

[0061] In this embodiment, it should be noted that in S3, the problem of delayed response to gradual damage in existing technologies is solved by quantifying the degree of deviation of the overall change pattern of the same environmental object in the historical and current two-cycle periods. The specific operation is divided into two steps: subtract the current boundary box change trend value from the historical boundary box change trend value of the object; take the absolute value of the relative difference to eliminate the influence of positive and negative directionality on the judgment, thereby focusing on the deviation magnitude itself (assuming that in a certain vegetation case, the relative difference between the historical trend value and the current trend value is -1.6%, then the absolute value is 1.6%).

[0062] In one implementation, semantic segmentation of the region of interest for the j-th environmental monitoring object in S5 includes: The DeepLab algorithm is used to perform semantic segmentation on the region of interest of the j-th environmental monitoring object. The DeepLab algorithm includes an encoder and a decoder. The encoder includes a deep convolutional neural network and spatial pyramid pooling with dilated convolution.

[0063] In this embodiment, it should be noted that DeepLabV3+ can be used in the DeepLab algorithm series. Specifically, the encoder of DeepLabV3+ mainly consists of two parts: a deep convolutional neural network and spatial pyramid pooling with dilated convolutions, wherein the deep convolutional neural network is an improvement based on the Xception network.

[0064] The Xception network is a deep separable convolutional network. It is characterized by low computational cost and good performance in image classification and object detection. Its main features are as follows: (1) adding more layers; (2) replacing all max pooling layers with separable convolutions, thereby applying separable dilated convolutions to extract feature maps of arbitrary resolution; (3) adding batch normalization after 3x3 separable convolutions.

[0065] After processing by the Xception network, the ASPP structure is used to introduce multi-scale feature information, improving the model's segmentation performance. The main feature of the ASPP structure is that it uses dilated convolutions instead of regular convolutions, thereby expanding the receptive field and helping to extract multi-scale information. Dilated convolutions can expand the receptive field without reducing the feature map resolution, preserving more spatial detail information.

[0066] The ASPP module mainly includes the following parts: (1) a convolution with a kernel size of 1×1 and three dilated convolutions with a kernel size of 3x3. When output_stride=16, its rate is (6, 12, 18). If output_stride=8, the rate is (12, 24, 36); (2) a global average pooling to obtain multi-scale features of the image, a convolution with a kernel size of 1x1, and then upsampling using bilinear interpolation to obtain the same size as other branches; (3) merging the outputs of the 5 parallel branches together, and then passing them through a convolution with a kernel size of 1x1 to obtain new features with 256 channels.

[0067] The decoder part of DeepLabV3+ mainly concatenates the low-level features extracted by DCNN with the high-level features extracted by ASPP structure, and then uses a 3x3 convolution kernel and upsampling.

[0068] Because the DeepLabv3+ algorithm uses supervised learning for model training, it requires the use of Lab tools to annotate the data. This involves adding bounding boxes and classifying objects such as tower bases, roads, construction waste dumps, tension fields, and residential areas in the image. Once the data annotation is complete, it is saved as an image annotation file for model training.

[0069] like Figure 7 As shown, an industrial IoT full lifecycle environmental impact situation identification system is also provided. The system includes a management platform, a sensor network platform, and an object platform that are connected in sequence via communication. The management platform includes: The data acquisition module is used to acquire the current image data sequence of the area to be monitored in the IoT project during the current identification period, and to acquire the historical image data sequence of the area to be monitored in the IoT project during the previous historical identification period, wherein the area to be monitored includes multiple environmental monitoring objects. The target monitoring module is used to perform target monitoring on each historical image data in the historical image data sequence and each current image data in the current image data sequence, generate historical target monitoring bounding boxes for each environmental monitoring object in each historical image data, and generate current target monitoring bounding boxes for each environmental monitoring object in each current image data. The first data processing module is used to obtain the historical boundary box change trend value of the j-th environmental monitoring object based on the historical target monitoring boundary boxes of multiple j-th environmental monitoring objects, obtain the current boundary box change trend value of the j-th environmental monitoring object based on the current target monitoring boundary boxes of multiple j-th environmental monitoring objects, and obtain the absolute difference of the j-th environmental monitoring object based on the historical boundary box change trend value and the current boundary box change trend value. The second data processing module is used to crop the current target monitoring bounding box of the j-th environmental monitoring object according to the coordinate parameters of the current target monitoring bounding box of the j-th environmental monitoring object in the current image data when the absolute difference of the j-th environmental monitoring object exceeds the preset threshold, so as to obtain the area of ​​interest of the j-th environmental monitoring object. The segmentation and monitoring identification module is used to perform semantic segmentation on the region of interest of the j-th environmental monitoring object, obtain the segmentation result of the j-th environmental monitoring object, and obtain the monitoring identification result for the j-th environmental monitoring object based on the segmentation result.

[0070] In one implementation, the target monitoring module is further configured to: perform target monitoring on each historical image data in the historical image data sequence and each current image data in the current image data sequence based on the Yo1ov7 algorithm, wherein the Yo1ov7 algorithm includes an input layer, a backbone network, a neck network, and an output layer.

[0071] In one embodiment, the first data processing module is further configured to: obtain the area within the historical boundary frame of the j-th environmental monitoring object based on the historical target monitoring boundary frames of the j-th environmental monitoring object; obtain the historical area change ratio of the j-th environmental monitoring object according to the sequential order of each historical image data in the historical image data sequence, based on the area within the historical boundary frame of the j-th environmental monitoring object in the next historical image data and the area within the historical boundary frame of the j-th environmental monitoring object in the previous historical image data; determine the validity of each historical area change ratio and obtain the valid historical area change ratio, and obtain the historical boundary frame change trend value of the j-th environmental monitoring object based on each valid historical area change ratio of the j-th environmental monitoring object.

[0072] In one embodiment, the first data processing module is further configured to: obtain the area within the current bounding box of the j-th environmental monitoring object in each current image data according to the current target monitoring bounding boxes of multiple j-th environmental monitoring objects; obtain the current area change ratio of the j-th environmental monitoring object according to the sequential order of each current image data in the current image data sequence, based on the area within the current bounding box of the j-th environmental monitoring object in the next current image data and the area within the current bounding box of the j-th environmental monitoring object in the previous current image data; determine the validity of each current area change ratio and obtain the valid current area change ratio, and obtain the current bounding box change trend value of the j-th environmental monitoring object based on each valid current area change ratio of the j-th environmental monitoring object.

[0073] In this embodiment, it should be noted that the specific methods of performing the above-mentioned industrial IoT full life cycle environmental impact situation identification system have been described in detail in the embodiments of the relevant industrial IoT full life cycle environmental impact situation identification method, and will not be elaborated here.

[0074] It should also be noted that the entire industrial IoT lifecycle environmental impact situation identification system can be applied to the optimized industrial IoT. Figure 7 This is a schematic diagram of the composition of the Industrial Internet of Things (IIoT) full lifecycle environmental impact situation identification system of the present invention. Figure 8 This is a schematic diagram illustrating the optimized industrial Internet of Things (IIoT) involved in this invention. (See diagram below.) Figure 7 and Figure 8 As shown, the optimized Industrial Internet of Things (IIoT) includes a user platform, a service platform, a management platform, a sensor network platform, and an object platform that establish communication in sequence. The user platform is configured to provide front-end services to users; users obtain the necessary perception service information through the user platform, process the perception service information, and transform it into user perception information; users analyze the user perception information and make corresponding decisions based on their own wishes, and transform the user perception information into user control information through the corresponding information system and send it to the service platform, thereby demonstrating the user's corresponding service needs and wishes.

[0075] The physical entities of the user platform include various user terminals, such as mobile phones, computers, and dedicated terminals, which provide user services through integration with user information system software.

[0076] The service platform is configured as an API server or other server used to establish communication between the management platform and the user platform to achieve corresponding functions; the physical entity of the service platform includes various servers.

[0077] The management platform is configured to perform at least one of the following: device operation status monitoring and management, data monitoring and management, device parameter management, and lifecycle management; the management platform is the overall operation platform for the Internet of Things, which may include various management sub-platforms, with different management sub-platforms performing different management tasks; the physical entities of the management platform include various servers.

[0078] The sensor network platform is configured to perform at least one of the following functions: network management, command management, device status management, data protocol management, data parsing, data classification, data transmission monitoring, and data transmission security management. The sensor network platform provides functions such as data communication, transmission, parsing, identification, and classification, avoiding the direct aggregation of data from various object platforms onto the management platform, which would otherwise result in data redundancy and low data processing efficiency. The physical entities of the object platforms include various gateways, edge computing devices, etc.

[0079] The object platform is configured to perform specific production control, detection, measurement and other production tasks; the physical entities of the production objects include various production equipment, sensors and so on.

[0080] The preferred embodiments of this disclosure have been described in detail above with reference to the accompanying drawings. However, this disclosure is not limited to the specific details of the above embodiments. Within the scope of the technical concept of this disclosure, various simple modifications can be made to the technical solutions of this disclosure, and these simple modifications all fall within the protection scope of this disclosure.

[0081] It should also be noted that the various specific technical features described in the above embodiments can be combined in any suitable manner without contradiction. To avoid unnecessary repetition, this disclosure will not describe the various possible combinations separately.

[0082] Furthermore, various different embodiments of this disclosure can be combined in any way, as long as they do not violate the spirit of this disclosure, they should also be regarded as the content disclosed in this disclosure.

[0083] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. A method for identifying the environmental impact situation throughout the entire lifecycle of an industrial Internet of Things (IoT), characterized in that, include: The current image data sequence of the area to be monitored in the IoT project during the current identification period is obtained, and the historical image data sequence of the area to be monitored in the IoT project during the previous historical identification period is obtained, wherein the area to be monitored includes multiple environmental monitoring objects. Target monitoring is performed on each historical image data in the historical image data sequence and each current image data in the current image data sequence. Historical target monitoring bounding boxes are generated for each environmental monitoring object in each historical image data, and current target monitoring bounding boxes are generated for each environmental monitoring object in each current image data. The historical boundary box change trend value of the j-th environmental monitoring object is obtained based on the historical target monitoring boundary boxes of multiple j-th environmental monitoring objects, and the current boundary box change trend value of the j-th environmental monitoring object is obtained based on the current target monitoring boundary boxes of multiple j-th environmental monitoring objects. The absolute difference is obtained based on the historical boundary box change trend value and the current boundary box change trend value of the j-th environmental monitoring object. If the absolute difference of the j-th environmental monitoring object exceeds the preset threshold, the current target monitoring bounding box of the j-th environmental monitoring object is cropped based on the coordinate parameters of the current target monitoring bounding box of the j-th environmental monitoring object in the current image data closest to the current time, so as to obtain the region of interest of the j-th environmental monitoring object. Semantic segmentation is performed on the region of interest of the j-th environmental monitoring object to obtain the segmentation result of the j-th environmental monitoring object, and the monitoring and identification result of the j-th environmental monitoring object is obtained based on the segmentation result.

2. The method for identifying the environmental impact situation of the entire lifecycle of the Industrial Internet of Things according to claim 1, characterized in that, The target monitoring of each historical image data in the historical image data sequence and each current image data in the current image data sequence includes: The Yo1ov7 algorithm is used to perform target detection on each historical image data in the historical image data sequence and each current image data in the current image data sequence. The Yo1ov7 algorithm includes an input layer, a backbone network, a neck network, and an output layer.

3. The method for identifying the environmental impact situation of the entire lifecycle of the Industrial Internet of Things according to claim 1, characterized in that, The step of obtaining the historical boundary box change trend value of the j-th environmental monitoring object based on the historical target monitoring boundary boxes of multiple j-th environmental monitoring objects includes: The area within the historical boundary frame of multiple j-th environmental monitoring objects is obtained based on the historical target monitoring boundary frames of multiple j-th environmental monitoring objects; According to the sequential order of each historical image data in the historical image data sequence, the historical area change ratio of the j-th environmental monitoring object is obtained by taking into account the area within the historical boundary box of the j-th environmental monitoring object in the next historical image data and the area within the historical boundary box of the j-th environmental monitoring object in the previous historical image data. The validity of each historical area change ratio is determined and the valid historical area change ratio is obtained. Based on the valid historical area change ratios of each environmental monitoring object, the historical boundary box change trend value of the j-th environmental monitoring object is obtained.

4. The method for identifying the environmental impact situation of the entire life cycle of the Industrial Internet of Things according to claim 1, characterized in that, The step of obtaining the current boundary box change trend value of the j-th environmental monitoring object based on the current target monitoring boundary boxes of multiple j-th environmental monitoring objects includes: Based on the current target monitoring bounding boxes of multiple j-th environmental monitoring objects, obtain the area within the current bounding box of the j-th environmental monitoring object in each current image data; According to the sequential order of each current image data in the current image data sequence, the change ratio of the current area of ​​the j-th environmental monitoring object is obtained by taking into account the current bounding box area of ​​the j-th environmental monitoring object in the next current image data and the current bounding box area of ​​the j-th environmental monitoring object in the previous current image data. The validity of each current area change ratio is determined and the valid current area change ratio is obtained. Based on the valid current area change ratios of the j-th environmental monitoring object, the current boundary box change trend value of the j-th environmental monitoring object is obtained.

5. The method for identifying the environmental impact situation of the entire lifecycle of the Industrial Internet of Things according to claim 1, characterized in that, The process of obtaining the absolute difference between the historical boundary box change trend value and the current boundary box change trend value for the j-th environmental monitoring object includes: Subtract the current bounding box change trend value from the historical bounding box change trend value to obtain the relative difference value, and take the absolute value of the relative difference value as the absolute difference value.

6. The method for identifying the environmental impact situation of the entire life cycle of the Industrial Internet of Things according to claim 1, characterized in that, The semantic segmentation of the region of interest for the j-th environmental monitoring object includes: The DeepLab algorithm is used to perform semantic segmentation on the region of interest of the j-th environmental monitoring object. The DeepLab algorithm includes an encoder and a decoder. The encoder includes a deep convolutional neural network and spatial pyramid pooling with dilated convolution.

7. An industrial Internet of Things (IoT) full lifecycle environmental impact situation identification system, characterized in that, The system includes a management platform, a sensor network platform, and an object platform that are sequentially connected in communication. The management platform includes: The data acquisition module is used to acquire the current image data sequence of the area to be monitored in the IoT project during the current identification period, and to acquire the historical image data sequence of the area to be monitored in the IoT project during the previous historical identification period, wherein the area to be monitored includes multiple environmental monitoring objects. The target monitoring module is used to perform target monitoring on each historical image data in the historical image data sequence and each current image data in the current image data sequence, generate historical target monitoring bounding boxes for each environmental monitoring object in each historical image data, and generate current target monitoring bounding boxes for each environmental monitoring object in each current image data. The first data processing module is used to obtain the historical boundary box change trend value of the j-th environmental monitoring object based on the historical target monitoring boundary boxes of multiple j-th environmental monitoring objects, obtain the current boundary box change trend value of the j-th environmental monitoring object based on the current target monitoring boundary boxes of multiple j-th environmental monitoring objects, and obtain the absolute difference of the j-th environmental monitoring object based on the historical boundary box change trend value and the current boundary box change trend value. The second data processing module is used to crop the current target monitoring bounding box of the j-th environmental monitoring object according to the coordinate parameters of the current target monitoring bounding box of the j-th environmental monitoring object in the current image data when the absolute difference of the j-th environmental monitoring object exceeds the preset threshold, so as to obtain the area of ​​interest of the j-th environmental monitoring object. The segmentation and monitoring identification module is used to perform semantic segmentation on the region of interest of the j-th environmental monitoring object, obtain the segmentation result of the j-th environmental monitoring object, and obtain the monitoring identification result for the j-th environmental monitoring object based on the segmentation result.

8. The industrial IoT full lifecycle environmental impact situation identification system according to claim 7, characterized in that, The target monitoring module is also used for: The Yo1ov7 algorithm is used to perform target detection on each historical image data in the historical image data sequence and each current image data in the current image data sequence. The Yo1ov7 algorithm includes an input layer, a backbone network, a neck network, and an output layer.

9. The industrial IoT full lifecycle environmental impact situation identification system according to claim 7, characterized in that, The first data processing module is also used for: The area within the historical boundary frame of multiple j-th environmental monitoring objects is obtained based on the historical target monitoring boundary frames of multiple j-th environmental monitoring objects; According to the sequential order of each historical image data in the historical image data sequence, the historical area change ratio of the j-th environmental monitoring object is obtained by taking into account the area within the historical boundary box of the j-th environmental monitoring object in the next historical image data and the area within the historical boundary box of the j-th environmental monitoring object in the previous historical image data. The validity of each historical area change ratio is determined and the valid historical area change ratio is obtained. Based on the valid historical area change ratios of each environmental monitoring object, the historical boundary box change trend value of the j-th environmental monitoring object is obtained.

10. The industrial IoT full lifecycle environmental impact situation identification system according to claim 7, characterized in that, The first data processing module is also used for: Based on the current target monitoring bounding boxes of multiple j-th environmental monitoring objects, obtain the area within the current bounding box of the j-th environmental monitoring object in each current image data; According to the sequential order of each current image data in the current image data sequence, the change ratio of the current area of ​​the j-th environmental monitoring object is obtained by taking into account the current bounding box area of ​​the j-th environmental monitoring object in the next current image data and the current bounding box area of ​​the j-th environmental monitoring object in the previous current image data. The validity of each current area change ratio is determined and the valid current area change ratio is obtained. Based on the valid current area change ratios of the j-th environmental monitoring object, the current boundary box change trend value of the j-th environmental monitoring object is obtained.

Images