Fire-fighting hidden danger identification recording method based on artificial intelligence

Abstract

The application discloses a fire-fighting hidden danger identification record method based on artificial intelligence, and relates to the technical field of fire-fighting hidden danger identification.The method comprises the following steps: based on artificial intelligence, uniform dimension feature vectors are extracted from continuously acquired fire-fighting hidden danger identification data and unique identification codes are given, forming a cross-sequence consistent identification vector group; meanwhile, the overlapping relationship of the corresponding space region is continuously mapped, generating a space evolution track group; the cross-sequence consistent identification vector group and the space evolution track group are compared synchronously, the coupling relationship between the feature consistency maintenance degree and the space overlapping change trend is calculated, and it is judged whether the feature expression is consistent and the space overlapping relationship gradually changes or not.The application solves the problem that the same fire-fighting hidden danger is split and recorded under the condition that the features are consistent but the space gradually deviates, realizes the continuous association and dynamic regulation of the hidden danger identification result, and thus improves the completeness and accuracy of the identification record.

Description

Technical Field

[0001] This invention relates to the field of fire hazard identification technology, and more specifically to a fire hazard identification and recording method based on artificial intelligence. Background Technology

[0002] Artificial intelligence-based fire hazard identification and recording refers to an information processing method that utilizes artificial intelligence technology to automatically analyze and process images, videos, or multi-source data in fire-related scenarios, thereby achieving intelligent identification of potential fire safety hazards and structurally recording and managing the identification results. This technology typically uses computer vision and deep learning algorithms as its core, performing preprocessing, feature extraction, and model inference on on-site data collected by monitoring equipment to automatically detect various hazard types such as flames, smoke, blocked passages, and missing or abnormal fire-fighting facilities. In practical applications, the process generally includes data acquisition, data preprocessing, intelligent identification, and result recording and communication: First, real-time or historical data is acquired through cameras or sensors; then, preprocessing operations such as noise reduction, enhancement, and format conversion are performed on the data; next, a trained artificial intelligence model analyzes the data, outputting information such as hazard category, location, and confidence level; finally, the identification results are recorded according to a preset data structure and stored in conjunction with time and spatial information, and can be uploaded to a management platform or trigger subsequent business processing flows through network communication mechanisms, thereby achieving digital and standardized management of fire hazard information.

[0003] The existing technology has the following shortcomings: In the process of identifying and recording fire hazards based on artificial intelligence, when the same fire hazard maintains consistent feature expression in consecutive identification results but the spatial overlapping area gradually shifts, a phenomenon occurs where multiple identification results partially overlap and their positions gradually change. Due to slight changes in perspective or natural expansion of the target shape during image acquisition, the position of the detection box shifts, making it impossible for the spatial position of the same hazard in different identification results to be completely aligned. This results in a situation where the features are consistent but the spatial distribution is inconsistent. Existing technologies cannot adjust the target association strategy in fire hazard identification and recording based on the identification results under the condition of consistent feature expression and gradually changing spatial overlap. This causes the same fire hazard to be judged as multiple different targets during the association process, resulting in multiple independent hazard records during the recording process. Ultimately, this leads to fragmentation of the recording results and affects the continuous analysis and overall judgment of fire hazards.

[0004] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] The purpose of this invention is to provide an artificial intelligence-based method for identifying and recording fire hazards, in order to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a fire hazard identification and recording method based on artificial intelligence, specifically including the following steps: S1. Based on artificial intelligence, extract unified dimension feature vectors from continuously acquired fire hazard identification data and assign them unique identifier codes to form a cross-sequence consistent identifier vector group. At the same time, continuously map the overlapping relationship of corresponding spatial regions to generate a spatial evolution trajectory group. S2. Synchronously compare the cross-sequence consistent identifier vector group with the spatial evolution trajectory group. By calculating the coupling relationship between the degree of feature consistency maintenance and the spatial overlap change trend, determine whether there is a situation where the feature expression is consistent and the spatial overlap relationship gradually changes, and generate a state identifier. S3. When there are consistent feature expressions and the spatial overlap relationship gradually changes, artificial intelligence is invoked to perform unified mapping processing on cross-sequence consistent identifier vector groups to generate a unique hidden danger identifier sequence for determining the identification result. S4. Based on the unique hazard identification sequence and spatial evolution trajectory group, the continuous identification data is collected and processed, and the association judgment constraint relationship is constructed based on the degree of feature consistency maintenance and spatial overlap change trend. The target association strategy in the fire hazard identification record is adjusted according to the identification results. S5. Combining the changing trends of the spatial evolution trajectory group and the cross-sequence consistent identifier vector group, the association judgment constraint relationship is updated, and dynamic control processing is performed through artificial intelligence to complete the continuous control of fire hazard identification records.

[0007] Preferably, S1 is as follows: Based on artificial intelligence, feature extraction processing is performed on continuously acquired fire hazard identification data. By performing multi-scale feature mapping on image data and performing unified dimension compression processing, a unified dimension feature vector is generated. The unified dimension feature vector is then numbered according to time sequence to assign a unique identifier code. The uniform dimension feature vector and the unique identifier code are arranged sequentially according to time order, and then combined according to the same number order to form a cross-sequence consistent identifier vector group; For each unified dimension feature vector in the cross-sequence consistent identifier vector group, the corresponding spatial region location information is extracted. By aligning the spatial regions in adjacent time series and calculating the area ratio of the overlapping part of the spatial regions, the overlapping range is determined. Then, according to the order of change of the overlapping range, the spatial regions in the continuous time series are matched and connected one by one to establish a continuous correspondence between spatial regions and realize the continuous mapping of the overlapping relationship of spatial regions. Based on the continuous mapping results of the overlapping relationships of spatial regions, the spatial regions are connected in chronological order. By recording the location change paths of the spatial regions in the continuous time series, a set of spatial evolution trajectories is generated.

[0008] Preferably, S2 specifically includes the following steps: S201. Synchronous comparison is performed between the cross-sequence consistent identifier vector group and the spatial evolution trajectory group. The feature vectors of each unified dimension in the cross-sequence consistent identifier vector group are aligned one by one with the corresponding spatial regions in the spatial evolution trajectory group according to the time order. A one-to-one correspondence between feature information and spatial evolution trajectory is established based on the unique identifier code. S202. After completing the synchronization comparison, by performing adjacent difference analysis on the unified dimension feature vectors of continuous time series in the cross-sequence consistent identifier vector group, the continuous time period with difference changes below the preset change range is divided into stable intervals of feature consistency maintenance. At the same time, the spatial region overlap relationship of continuous time series in the spatial evolution trajectory group is extracted, and the continuous time period with continuous change of overlap relationship is divided into continuous change intervals of spatial overlap change trend. The stable intervals and continuous change intervals are matched in chronological order to form a coupling relationship. S203. Based on the coupling relationship, the continuous time series is judged and processed. By comparing the time coverage of the stable interval and the continuous change interval one by one, when the coverage of the stable interval and the coverage of the continuous change interval meet the continuous correspondence in the same time period, it is judged that there is a situation where the feature expression is consistent and the spatial overlap relationship gradually changes. The time series in the corresponding time period is assigned a preset mark value to generate a status identifier.

[0009] Preferably, S202 specifically refers to: After completing the synchronization comparison, the differences of the unified dimension feature vectors of the continuous time series in the cross-sequence consistent identifier vector group are calculated one by one, and the difference values ​​between adjacent unified dimension feature vectors are arranged in chronological order. The time periods in which the continuous difference values ​​are within the preset range of change are segmented and marked to determine the stable interval of the degree of feature consistency maintenance. The spatial region overlap relationship of continuous time series in the spatial evolution trajectory group is extracted frame by frame. By obtaining the change in the spatial region overlap range in adjacent time series and arranging them in chronological order, continuous time periods with unidirectional change are segmented and marked to determine the continuous change interval of the spatial overlap change trend. The stable intervals that maintain the consistency of features in chronological order are aligned with the continuous intervals that show spatial overlap. By comparing the start and end positions of each time period, corresponding matching is performed on the intervals that overlap in time range to form a coupling relationship.

[0010] Preferably, S3 is as follows: When consistent feature expressions and gradually changing spatial overlap relationships occur, artificial intelligence is invoked to perform unified mapping processing on cross-sequence consistent identifier vector groups. By extracting the unified dimension feature vectors of continuous time series in the cross-sequence consistent identifier vector group according to the time order, the similarity of the unified dimension feature vectors in adjacent time series is calculated one by one, and the similarity calculation results are arranged in time order. Continuous time periods with similarity within a preset similarity range are segmented and grouped, and the unified dimension feature vectors in the same segment are grouped together. Based on the grouping results, a corresponding association set between unified dimension feature vectors is established to form a unified mapping relationship. Based on the unified mapping process, the unified dimension feature vectors in each unified mapping relationship are arranged in chronological order, and a corresponding unique identifier code is assigned to each unified mapping relationship. The unique identifier codes in the corresponding time series are combined sequentially to generate a unique hidden danger identifier sequence. Based on the unique hazard identification sequence, the unified dimension feature vector in the cross-sequence consistent identification vector group is assigned to the same fire hazard. The identification result is output according to the correspondence.

[0011] Preferably, S4 specifically includes the following steps: S401. Based on the unique hazard identification sequence and the spatial evolution trajectory group, the continuous identification data is collected and processed. The continuous identification data is divided into time order according to the unique hazard identification sequence, and the continuous identification data with the same unique hazard identification is matched with the corresponding spatial area in the spatial evolution trajectory group. The matched continuous identification data is combined and collected to form a continuous identification data set with the corresponding unique hazard identification. S402. Based on the continuous identification data set, extract the feature consistency retention degree and spatial overlap change trend corresponding to each continuous identification data set, and arrange the feature consistency retention degree and spatial overlap change trend synchronously in chronological order. By jointly constraining the feature consistency retention degree and spatial overlap change trend, construct the association judgment constraint relationship. S403. Based on the association determination constraint relationship and combined with the identification results, adjust the target association strategy in the fire hazard identification record. By dividing the association range of continuous identification data in the continuous identification data set, perform unified association processing on continuous identification data that meets the association determination constraint relationship, and perform separation processing on continuous identification data that does not meet the association determination constraint relationship, so as to achieve the adjustment of the target association strategy.

[0012] Preferably, S402 is as follows: Based on the continuous recognition dataset, the continuous recognition data is traversed in chronological order to extract the feature consistency retention degree and spatial overlap change trend corresponding to each time position. The extraction results are then serialized and arranged in chronological order to form a feature consistency retention degree sequence and a spatial overlap change trend sequence. The sequence of feature consistency preservation degree and the sequence of spatial overlap change trend are synchronized and aligned at the same time position. The feature consistency preservation degree and spatial overlap change trend at each time position are combined and recorded to form a corresponding combined sequence arranged in chronological order. For continuous time periods in the corresponding combined sequence, interval division processing is performed. The degree of feature consistency and spatial overlap change trend within each interval are jointly constrained and integrated. By uniformly associating and labeling the data within the same interval, and connecting and expressing the association labels of each interval in chronological order, an association judgment constraint relationship covering continuous time series is constructed.

[0013] Preferably, S5 is as follows: By combining the changing trends of the spatial evolution trajectory group and the cross-sequence consistent identifier vector group, the spatial region changing trend in the spatial evolution trajectory group and the uniform dimension feature vector changing trend in the cross-sequence consistent identifier vector group are extracted in chronological order. The spatial region changing trend and the uniform dimension feature vector changing trend are synchronized and aligned. The changing trends at each time position are combined accordingly to form a changing trend combination sequence. The association determination constraint relationship is updated based on the combination sequence of changing trends. By dividing the combination sequence of changing trends within a continuous time period into intervals, the association determination constraint relationship is adjusted according to the changing trend within the interval. At the same time, artificial intelligence is called to continuously adjust the updated association determination constraint relationship, and the target association range is divided and reconstructed based on the recognition results to achieve dynamic control processing.

[0014] The technical effects and advantages provided by the present invention in the above technical solution are as follows: 1. This invention introduces a collaborative modeling mechanism of unified dimensional feature vectors and spatial evolution trajectory groups during the fire hazard identification and recording process. This enables the joint expression of fire hazards in both temporal and spatial dimensions, allowing for the simultaneous characterization and analysis of feature consistency and spatial location changes during continuous identification. Furthermore, by constructing a coupling relationship between the degree of feature consistency and the trend of spatial overlap, it can accurately identify the evolutionary process of hazards in complex scenarios that exhibit "consistent features but gradual spatial shifts." Moreover, a unique hazard identification sequence is used to uniformly map the identification data across time, thereby avoiding the problem of repeated identification of the same hazard due to spatial location shifts. This ensures that the hazard identification results remain continuous and consistent over time, improving the accuracy and consistency of fire hazard identification and recording.

[0015] 2. This invention introduces association-based constraint relationships and dynamically updates them in conjunction with changing trends. This allows the target association strategy to no longer rely on fixed spatial locations or static overlap relationships, but rather to adaptively adjust based on the joint constraints of feature changes and spatial changes. This ensures stable association with the same hazard even as it expands, moves, or changes in form. Simultaneously, artificial intelligence performs dynamic control processing, continuously adjusting and reconstructing the target association range. This enables fire hazard identification records to dynamically adapt to the evolution of hazard events, avoiding fragmented records and enhancing the ability to continuously track and analyze the development process of hazard events. Ultimately, this improves the accuracy of hazard evolution trend judgment and decision support capabilities in fire safety management. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0017] Figure 1 This is a schematic diagram of the process of the present invention. Detailed Implementation

[0018] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the description of this disclosure will be more complete and fully convey the concept of the exemplary embodiments to those skilled in the art.

[0019] This invention provides, for example Figure 1 The fire hazard identification and recording method based on artificial intelligence, as shown, specifically includes the following steps: S1. Based on artificial intelligence, extract unified dimension feature vectors from continuously acquired fire hazard identification data and assign them unique identifier codes to form a cross-sequence consistent identifier vector group. At the same time, continuously map the overlapping relationship of corresponding spatial regions to generate a spatial evolution trajectory group. In this embodiment, S1 specifically refers to: Based on artificial intelligence, feature extraction processing is performed on continuously acquired fire hazard identification data. By performing multi-scale feature mapping on image data and performing unified dimension compression processing, a unified dimension feature vector is generated. The unified dimension feature vector is then numbered according to time sequence to assign a unique identifier code. The uniform dimension feature vector and the unique identifier code are arranged sequentially according to time order, and then combined according to the same number order to form a cross-sequence consistent identifier vector group; For each unified dimension feature vector in the cross-sequence consistent identifier vector group, the corresponding spatial region location information is extracted. By aligning the spatial regions in adjacent time series and calculating the area ratio of the overlapping part of the spatial regions, the overlapping range is determined. Then, according to the order of change of the overlapping range, the spatial regions in the continuous time series are matched and connected one by one to establish a continuous correspondence between spatial regions and realize the continuous mapping of the overlapping relationship of spatial regions. Based on the continuous mapping results of the overlapping relationships of spatial regions, the spatial regions are connected in chronological order. By recording the location change paths of the spatial regions in the continuous time series, a set of spatial evolution trajectories is generated.

[0020] When processing continuously acquired fire hazard identification data based on artificial intelligence, image data can be input into a pre-trained deep learning model. The deep learning model extracts edge information, texture distribution, and regional structure in the image step by step through a multi-layer convolutional structure, and outputs feature responses of different scales at different levels. Then, the features of different scales are integrated through channel fusion processing. Subsequently, the feature responses are compressed into vector representations of uniform length through global pooling operations, and the vectors corresponding to each frame of the image are sequentially labeled according to the acquisition order, thereby forming a uniform dimension feature vector and completing the assignment process of unique identifier encoding, so that each feature vector has a distinguishable time series identifier.

[0021] After obtaining the unified dimension feature vector and unique identifier encoding, a time series index structure can be constructed. The feature vectors corresponding to each frame are arranged in chronological order, and the feature vectors in different time periods are matched by the identifier encoding. This allows feature vectors with the same identifier position to form a correspondence in different time series. Then, by aggregating and organizing these correspondences, a cross-series consistent identifier vector group is formed, so that feature information at the same position can be uniformly managed and participate in subsequent association processing under different time sources.

[0022] After the cross-sequence consistent identifier vector group is constructed, the spatial location information of the fire hazard area in each frame image can be obtained through the target detection output. The spatial location information can be represented in the form of bounding box coordinates. Then, coordinate mapping processing is performed on the spatial regions in adjacent time series so that the spatial regions collected at different times are under the same reference coordinate system. On this basis, the overlapping area between two spatial regions is extracted, and the degree of occupation of the overlapping area in the overall area is counted to obtain the area ratio information of the overlapping part of the spatial regions. Then, according to the change of area ratio in the continuous time series, the spatial regions are associated and matched according to the order of ratio change to realize the one-to-one connection relationship between the spatial regions.

[0023] After establishing one-to-one connections between spatial regions, path construction processing can be further performed on spatial regions in continuous time series. The center or boundary position of each spatial region at each time point is connected in chronological order to form a continuous trajectory path that reflects the spatial change process. The trajectory paths corresponding to multiple spatial regions are then aggregated and organized to form a spatial evolution trajectory group, so that the positional change process of fire hazards in the time dimension can be completely recorded and used for subsequent analysis and processing.

[0024] Artificial intelligence plays a key role in feature extraction and data representation throughout the process. It uses deep learning models to automatically learn features from image data, transforming visual information in the original image into a unified-dimensional feature vector. Fire hazard identification data refers to information about hazard areas obtained through image acquisition equipment and processed by detection. Feature extraction processing refers to using neural networks to perform multi-layer convolution calculations on images to obtain semantic information. Multi-scale feature mapping refers to extracting feature responses from images at different levels and ranges. Unified-dimensional compression processing refers to integrating multi-scale features into a vector representation with a fixed structure. Unique identifier encoding refers to labeling information used to distinguish different time series data.

[0025] Cross-sequence consistent identifier vector group represents a set of feature vectors that have corresponding relationships in different time series. Spatial region location information represents the location range of the hazard area in the image. Spatial regions in adjacent time series represent the corresponding detection areas in consecutive time frames. Region coordinate alignment processing refers to unifying spatial regions in different time frames to the same coordinate reference system. The area ratio of the overlapping part of spatial regions represents the relative proportion of the overlapping part of two spatial regions. The order of change of the overlapping range represents the trend of the degree of overlap in consecutive time. One-to-one matching connection means associating and connecting spatial regions according to time order. The continuous correspondence between spatial regions represents the mapping relationship of the same hazard in different times. The continuous mapping result of the spatial region overlap relationship represents the set of the mapping relationship. The location change path represents the trajectory information of the spatial region changing over time. The spatial evolution trajectory group represents a set of multiple trajectory paths, which is used to describe the spatial change process of fire hazards in the time series.

[0026] S2. Synchronously compare the cross-sequence consistent identifier vector group with the spatial evolution trajectory group. By calculating the coupling relationship between the degree of feature consistency maintenance and the spatial overlap change trend, determine whether there is a situation where the feature expression is consistent and the spatial overlap relationship gradually changes, and generate a state identifier. In this embodiment, S2 specifically includes the following steps: S201. Synchronous comparison is performed between the cross-sequence consistent identifier vector group and the spatial evolution trajectory group. The feature vectors of each unified dimension in the cross-sequence consistent identifier vector group are aligned one by one with the corresponding spatial regions in the spatial evolution trajectory group according to the time order. A one-to-one correspondence between feature information and spatial evolution trajectory is established based on the unique identifier code. In the specific implementation, the unified-dimensional feature vectors in the cross-sequence consistent identifier vector group can be sorted chronologically according to the unique identifier code, and the spatial regions in the spatial evolution trajectory group can be arranged in the same chronological order. Then, using the unique identifier code as an index, the unified-dimensional feature vectors are aligned one by one with the spatial regions at the corresponding time points. In practice, a dual index structure based on timestamps and identifier codes can be constructed. In the index structure, the feature vectors of each frame of the image are bound to the corresponding spatial regions, and one-to-one matching is achieved by traversing the time series. For example, in the monitoring experimental data of an industrial plant, multiple frames of images are continuously collected. The feature vectors corresponding to a certain fire hazard at different time points maintain a similar distribution in the feature space, while the position of the corresponding spatial region in the image gradually shifts. By aligning the feature vectors and spatial regions under the same identifier code frame by frame, the feature information in each frame can be paired with the corresponding spatial position, thereby establishing a complete time series mapping relationship. This processing can avoid matching errors caused by relying solely on spatial position, providing a stable data foundation for subsequent association judgments.

[0027] The cross-sequence consistent identifier vector group represents a set of unified-dimensional feature vectors organized by unified identifier encoding in different time series, used to express the feature information of the same fire hazard at different times; the spatial evolution trajectory group represents a set of trajectories formed by the positional changes of spatial regions in a continuous time series, used to describe the spatial change process of the hazard in the image; synchronous comparison represents the process of matching feature information with spatial evolution trajectories in a unified time sequence; the unified-dimensional feature vector represents fixed-structure feature data extracted and compressed by an artificial intelligence model; the unique identifier encoding represents the labeling information used to distinguish feature vectors of different time series; the feature information represents the semantic expression carried by the unified-dimensional feature vector; the spatial evolution trajectory represents the path information of spatial regions changing over time; the one-to-one correspondence represents the unique matching relationship established between feature information and spatial location in the same time series, which ensures that each feature vector corresponds to a spatial region and provides consistent data support for subsequent state determination and association processing.

[0028] S202. After completing the synchronization comparison, by performing adjacent difference analysis on the unified dimension feature vectors of continuous time series in the cross-sequence consistent identifier vector group, the continuous time period with difference changes below the preset change range is divided into stable intervals of feature consistency maintenance. At the same time, the spatial region overlap relationship of continuous time series in the spatial evolution trajectory group is extracted, and the continuous time period with continuous change of overlap relationship is divided into continuous change intervals of spatial overlap change trend. The stable intervals and continuous change intervals are matched in chronological order to form a coupling relationship. S203. Based on the coupling relationship, the continuous time series is judged and processed. By comparing the time coverage of the stable interval and the continuous change interval one by one, when the coverage of the stable interval and the coverage of the continuous change interval meet the continuous correspondence in the same time period, it is judged that there is a situation where the feature expression is consistent and the spatial overlap relationship gradually changes. The time series in the corresponding time period is assigned a preset mark value to generate a status identifier.

[0029] Based on the coupling relationship, segment-by-segment judgment processing is performed on continuous time series. Specifically, the coverage of stable intervals and continuously changing intervals on the time axis is compared one by one. The start and end positions of each time segment are aligned and scanned, and the overlapping time ranges are extracted. When stable intervals and continuously changing intervals form continuous overlap within the same time segment without interruption, it is determined that the time segment simultaneously satisfies the conditions of stable features and continuous spatial change, thus identifying it as a situation where the feature expression is consistent and the spatial overlap relationship gradually changes. In the specific implementation, a list of time intervals can be constructed and traversed sequentially. Cross-detection is performed on each stable interval and each continuously changing interval. When consecutive cross-intervals are detected, the time segment is recorded as a valid judgment interval, and a preset label value is assigned to the time series within the interval. For example, in actual monitoring experimental data, the feature changes of a certain hidden danger area remain flat in continuous image frames, while the corresponding spatial position gradually shifts. Through interval comparison, it can be identified that the time segment simultaneously satisfies both conditions, thus generating a unified label for the time segment for subsequent association processing.

[0030] The coupling relationship represents the correspondence between the degree of feature consistency and the spatial overlap trend in the time dimension. The continuous time series represents the set of image frames arranged in the acquisition order. The time coverage of the stable interval and the continuously changing interval represents the start and end positions of each interval on the time axis. Satisfying continuous correspondence means that the two intervals have continuous overlap on the time axis without any interruption. The consistent feature expression and the gradual change in spatial overlap relationship represent the state features that remain stable at the feature level while undergo continuous changes at the spatial level. The preset label value represents the identification information used to distinguish different states. The state label represents the result information after the time series that meets the conditions is labeled. It is used to characterize the change characteristics of the hidden danger state within the time period and to provide a unified basis for subsequent identification records and target association.

[0031] In this embodiment, S202 specifically refers to: After completing the synchronization comparison, the differences of the unified dimension feature vectors of the continuous time series in the cross-sequence consistent identifier vector group are calculated one by one, and the difference values ​​between adjacent unified dimension feature vectors are arranged in chronological order. The time periods in which the continuous difference values ​​are within the preset range of change are segmented and marked to determine the stable interval of the degree of feature consistency maintenance. The spatial region overlap relationship of continuous time series in the spatial evolution trajectory group is extracted frame by frame. By obtaining the change in the spatial region overlap range in adjacent time series and arranging them in chronological order, continuous time periods with unidirectional change are segmented and marked to determine the continuous change interval of the spatial overlap change trend. The stable intervals that maintain the consistency of features in chronological order are aligned with the continuous intervals that show spatial overlap. By comparing the start and end positions of each time period, corresponding matching is performed on the intervals that overlap in time range to form a coupling relationship.

[0032] When processing the uniform-dimensional feature vectors of continuous time series, we can first calculate the dimension-wise differences of the feature vectors of the same potential hazard in consecutive image frames. Specifically, we compare each feature component and summarize the changes in each dimension to form an overall difference representation. Then, we construct a difference sequence according to the time order of the difference results between consecutive time points and perform sliding scan processing on the difference sequence. When the difference changes within a continuous time period are always within a preset range, the time period is divided into intervals and assigned continuous labels, thus forming a stable interval of feature consistency. For example, in actual surveillance video analysis, the morphological changes of smoke areas are small in a short period of time, and the changes in each dimension of the corresponding feature vector are relatively gentle. By continuously detecting such changes, a stable interval can be formed.

[0033] When processing spatial evolution trajectory groups, the spatial region boundary information of each frame in a continuous time series can be used to extract the overlapping range of spatial regions between adjacent time points frame by frame. Change trend information is obtained by comparing the boundaries of the overlapping regions. The overlapping change results in the continuous time series are constructed into a change sequence. Then, directional consistency detection is performed on the change sequence. When the change maintains the same direction of change within a continuous time period, that time period is divided into intervals and labeled, thus forming continuous change intervals of spatial overlapping change trends. For example, in a real-world scenario, the coverage area of ​​a potential hazard area gradually changes as it spreads with airflow. Continuous detection of the direction of change in the overlapping range can identify continuous change intervals.

[0034] After obtaining stable intervals that maintain feature consistency and continuous intervals that show spatial overlap trends, time alignment processing can be performed on the two interval sets. Specifically, this involves extracting the start and end times of each interval and detecting interval overlap in chronological order. When two intervals overlap on the time axis, the overlapping portion is extracted as a matching interval, and an association label is applied to the matching interval, thus forming a coupling relationship. This process can be achieved through interval scanning and interval merging, establishing a unified association basis for feature changes and spatial changes in the time dimension. For example, when both feature stability and spatial change exist within the same time period, that time period is identified as a coupled interval.

[0035] The unified-dimensional feature vector of a continuous time series represents the set of feature data extracted from the same hidden danger area at different time points. The dimension-wise difference calculation represents the process of comparing the corresponding components of each dimension in the feature vector. The difference sequence represents the feature change results arranged in chronological order. The preset change range represents the judgment interval used to limit the degree of feature change. The stable interval represents the time period in which the difference change is continuously within the limited range. The spatial region overlap relationship represents the overlap between spatial regions at different time points. The frame-by-frame change extraction represents the successive analysis of the overlap relationship in continuous time frames. The change sequence represents the time series expression of the change in the overlap range. The continuous change interval represents the time period in which the change direction remains consistent.

[0036] Interval division represents the boundary delineation and marking of continuous time periods; interval alignment represents the analysis of the correspondence between different types of intervals on the time axis; time start and time end represent the boundary positions of intervals on the time axis; interval overlap detection represents the determination of whether different intervals have overlapping areas on the time axis; matching interval represents the time period that simultaneously meets the conditions of feature stability and spatial change; association marking represents the unified identification processing of matching intervals; coupling relationship represents the correspondence between the degree of feature consistency and the trend of spatial overlap within the same time range. This relationship is used to support the status determination and target association processing in subsequent fire hazard identification records.

[0037] S3. When there are consistent feature expressions and the spatial overlap relationship gradually changes, artificial intelligence is invoked to perform unified mapping processing on cross-sequence consistent identifier vector groups to generate a unique hidden danger identifier sequence for determining the identification result. In this embodiment, S3 specifically refers to: When consistent feature expressions and gradually changing spatial overlap relationships occur, artificial intelligence is invoked to perform unified mapping processing on cross-sequence consistent identifier vector groups. By extracting the unified dimension feature vectors of continuous time series in the cross-sequence consistent identifier vector group according to the time order, the similarity of the unified dimension feature vectors in adjacent time series is calculated one by one, and the similarity calculation results are arranged in time order. Continuous time periods with similarity within a preset similarity range are segmented and grouped, and the unified dimension feature vectors in the same segment are grouped together. Based on the grouping results, a corresponding association set between unified dimension feature vectors is established to form a unified mapping relationship. Based on the unified mapping process, the unified dimension feature vectors in each unified mapping relationship are arranged in chronological order, and a corresponding unique identifier code is assigned to each unified mapping relationship. The unique identifier codes in the corresponding time series are combined sequentially to generate a unique hidden danger identifier sequence. Based on the unique hazard identification sequence, the unified dimension feature vector in the cross-sequence consistent identification vector group is assigned to the same fire hazard. The identification result is output according to the correspondence.

[0038] When consistent feature expressions and gradually changing spatial overlap occur, artificial intelligence can be invoked to perform unified mapping processing on cross-sequence consistent identifier vector groups. Specifically, this involves extracting consistent dimension feature vectors from continuous time series in chronological order, performing a dimensional comparison of consistent dimension feature vectors at adjacent time positions, summarizing the differences in each dimension to form an overall similarity calculation result, constructing a continuous sequence from the similarity calculation results in chronological order, and performing a continuous scan on this sequence. When the similarity calculation results remain within a preset similarity range within a continuous time period, this continuous time period is defined as a candidate interval, and the candidate intervals are segmented and aggregated. The consistent dimension feature vectors within the same candidate interval are grouped together, and a corresponding index set is established for each group to record the correlation between feature vectors, thereby forming a unified mapping relationship.

[0039] During the establishment of a unified mapping relationship, the unified dimension feature vectors in each group can be rearranged in chronological order, and an independent identifier code can be assigned to each group. The identifier code can be implemented by marking the group index set. At the same time, the identifier codes of the same group in the continuous time series are sequentially concatenated to construct an encoding sequence, and the continuity of the encoding sequence is detected. When the codes remain continuous and consistent in the time series, the continuous encoding segment is organized into a unique hidden danger identifier sequence to express the continuous existence relationship of the same hidden danger in the time dimension.

[0040] After obtaining a unique hazard identification sequence, the attribution determination process can be performed on the uniform dimension feature vectors in the cross-sequence consistent identification vector group based on the encoding sequence. Specifically, by traversing the uniform dimension feature vectors and reading the corresponding identification codes, uniform dimension feature vectors with the same identification codes are merged into the same set, and a uniform index structure is established for the feature vectors in the set. The set is then assigned to the same fire hazard object. At the same time, the identification result is output based on the set division result, so that each fire hazard maintains a uniform identification in the continuous time series.

[0041] Artificial intelligence is used in this process to perform similarity analysis and grouping of feature vectors with a unified dimension. Deep learning models are used to model the feature distribution, enabling the similarity between feature vectors to reflect the semantic stability of potential problems. Cross-sequence consistent identifier vector groups represent feature sets with a unified identifier relationship across different time series. Unified mapping processing represents the aggregation of feature vectors through similarity constraints. Dimensional comparison processing represents the item-by-item comparison of each dimension component of the feature vectors. The similarity calculation result represents the overall similarity between features. Preset similarity ranges represent limited intervals used to filter stable feature changes. Candidate intervals represent time periods that continuously meet the conditions. Segmented aggregation represents the division and classification of candidate intervals. Centralized grouping represents the aggregation of feature vectors within the same interval. Index sets represent data structures that record the relationships between feature vectors.

[0042] The unique identifier code represents the marking information that distinguishes different groups. The coding sequence represents the set of identifier information arranged in the time dimension. The unique hazard identifier sequence represents the coding expression result of the same hazard in continuous time. The attribution determination represents the classification processing of feature vectors according to the identifier code. The unified index structure represents the structural form for organizing and managing feature vectors in the same set. The same fire hazard represents the same object that exists continuously in the time series. The identification result represents the output information after the hazard object is uniformly classified, which is used for subsequent fire hazard identification record and target association processing.

[0043] S4. Based on the unique hazard identification sequence and spatial evolution trajectory group, the continuous identification data is collected and processed, and an association judgment constraint relationship is constructed based on the degree of feature consistency maintenance and spatial overlap change trend. The target association strategy in the fire hazard identification record is adjusted according to the identification results so that the identification objects with the same unique hazard identification remain uniformly associated under the association judgment constraint relationship. In this embodiment, S4 specifically includes the following steps: S401. Based on the unique hazard identification sequence and the spatial evolution trajectory group, the continuous identification data is collected and processed. The continuous identification data is divided into time order according to the unique hazard identification sequence, and the continuous identification data with the same unique hazard identification is matched with the corresponding spatial area in the spatial evolution trajectory group. The matched continuous identification data is combined and collected to form a continuous identification data set with the corresponding unique hazard identification. In the specific implementation process, the continuous identification data can first be organized chronologically based on the unique hazard identification sequence, forming an ordered data stream according to the collection time. Then, by reading the identifier code in the unique hazard identification sequence, continuous identification data with the same unique hazard identification are grouped into the same group. Next, combined with the spatial region information of the corresponding time position in the spatial evolution trajectory group, spatial location matching processing is performed on the continuous identification data at each time point, establishing a mapping relationship between each continuous identification data and its corresponding spatial region. After matching, the continuous identification data corresponding to the same unique hazard identification are connected and combined chronologically to form a complete data set. For example, in actual industrial monitoring experiments, when analyzing a persistent smoke hazard, although the detection results in different time frames may have spatial offsets, the unique hazard identification sequence can group these detection results into the same group. Furthermore, by combining the spatial evolution trajectory group to match the corresponding spatial regions in each frame, the detection data of the same hazard in continuous time are uniformly collected, thus avoiding the problem of duplicate recording caused by spatial offset and providing a consistent data foundation for subsequent correlation processing.

[0044] The unique hazard identification sequence and spatial evolution trajectory group are used to describe the changes in the identification of fire hazards in the time dimension and the changes in the location in the spatial dimension, respectively. Continuous identification data represents the set of hazard detection information acquired in a continuous time series. Aggregation processing represents the process of grouping and integrating data based on identification and spatial information. Time order division represents the arrangement and segmentation of continuous identification data according to the data collection order. Combination aggregation represents connecting the data corresponding to the same identification according to the time order to form a whole set. The continuous identification data set corresponding to the unique hazard identification represents the set of all identification data belonging to the same fire hazard in a continuous time series. This set forms a stable data structure through unified identification and spatial trajectory constraints, which is used to support the status judgment and target association processing in subsequent fire hazard identification records.

[0045] S402. Based on the continuous identification data set, extract the feature consistency retention degree and spatial overlap change trend corresponding to each continuous identification data set, and arrange the feature consistency retention degree and spatial overlap change trend synchronously in chronological order. By jointly constraining the feature consistency retention degree and spatial overlap change trend, construct the association judgment constraint relationship. S403. Based on the association determination constraint relationship and combined with the identification results, adjust the target association strategy in the fire hazard identification record. By dividing the association range of continuous identification data in the continuous identification data set, perform unified association processing on continuous identification data that meets the association determination constraint relationship, and perform separation processing on continuous identification data that does not meet the association determination constraint relationship, so as to achieve the adjustment of the target association strategy.

[0046] Based on the association judgment constraint relationship and the identification results, association control processing can be performed on the continuous identification data in the continuous identification dataset. Specifically, the continuous identification data is parsed one by one in chronological order, and combined with the unique hazard identifier and hazard category information in the identification results, constraint matching judgment is performed on the data at each time position. During the matching process, the constraint conditions corresponding to the degree of feature consistency and spatial overlap change trend are used as the judgment criteria. The continuity of data that meets the constraint conditions in a continuous time range is detected. When data is detected to continuously meet the constraint conditions in the time dimension, the time range is defined as the same association range, and the data in the range is uniformly associated, that is, the corresponding data is merged into the same record entry in the fire hazard identification record. For data that does not meet the constraint conditions in the time series, it is separated from the current association range by the breakpoint identification method, and a new association range is re-established. For example, in actual monitoring experiments, a certain smoke hazard may show gradual feature changes and a gradual shift in spatial location in consecutive image frames. By using constraint matching, it is possible to identify that the data segment belongs to the same hazard and perform unified association. When subsequent feature abrupt changes or discontinuous spatial changes occur, the corresponding data is assigned to a new association range, thereby avoiding the incorrect merging of different hazards.

[0047] The association determination constraint relationship represents a joint constraint expression structure composed of the degree of feature consistency maintenance and the spatial overlap change trend, used to limit whether continuous identification data at different time locations have association conditions; the identification result represents the output information after classifying and labeling the continuous identification data; the fire hazard identification record represents the data set for organizing, classifying and storing hazard information; the target association strategy represents the processing logic for performing merging or separation control on the continuous identification data; the association range division represents the division of the continuous identification data into intervals in the time dimension according to the association determination constraint relationship; the continuous identification data that satisfies the association determination constraint relationship represents the data set that simultaneously satisfies feature consistency and spatial change constraints within a continuous time range; the unified association processing represents merging the data in this set into the same hazard record; the continuous identification data that does not satisfy the association determination constraint relationship represents the data set that does not meet the constraint conditions; the separation processing represents splitting this type of data from the original association range and forming independent records, thereby ensuring the distinguishability between different fire hazards and the accuracy of the identification records.

[0048] In this embodiment, S402 specifically refers to: Based on the continuous recognition dataset, the continuous recognition data is traversed in chronological order to extract the feature consistency retention degree and spatial overlap change trend corresponding to each time position. The extraction results are then serialized and arranged in chronological order to form a feature consistency retention degree sequence and a spatial overlap change trend sequence. The sequence of feature consistency preservation degree and the sequence of spatial overlap change trend are synchronized and aligned at the same time position. The feature consistency preservation degree and spatial overlap change trend at each time position are combined and recorded to form a corresponding combined sequence arranged in chronological order. For continuous time periods in the corresponding combined sequence, interval division processing is performed. The degree of feature consistency and spatial overlap change trend within each interval are jointly constrained and integrated. By uniformly associating and labeling the data within the same interval, and connecting and expressing the association labels of each interval in chronological order, an association judgment constraint relationship covering continuous time series is constructed.

[0049] When processing a continuous recognition dataset, the dataset can be traversed sequentially according to time. By reading the recognition data at adjacent time positions and comparing the feature vectors of the same dimension, the changes in each dimension are summarized to form a feature change description. The degree of feature consistency is obtained through stability analysis of changes within a continuous time range. At the same time, the overlapping range of spatial regions at adjacent time positions is extracted for spatial regional information. By continuously tracking the boundary changes of the overlapping regions, a spatial change description sequence is constructed to obtain the spatial overlap change trend. Subsequently, the two types of data are arranged in time order to form a feature consistency preservation sequence and a spatial overlap change trend sequence with temporal correlation.

[0050] After forming the sequence of feature consistency preservation degree and the sequence of spatial overlap change trend, the two types of sequences can be synchronized and aligned based on the time index. By establishing a time position mapping relationship, the feature consistency preservation degree and spatial overlap change trend at the same time position are bound and recorded. The bound data is then structured and organized so that each time position forms a combined data item containing feature change information and spatial change information, thereby forming a corresponding combined sequence arranged in chronological order. This sequence is used to uniformly describe the dual change characteristics of the same hidden danger in the time dimension.

[0051] After the corresponding combined sequence is constructed, interval division processing can be performed on the continuous time period. Specifically, by continuously scanning the combined sequence, the continuity of the changes in the degree of feature consistency and spatial overlap trend in the time dimension is detected. When the feature changes remain stable and the spatial changes remain continuous within the continuous time range, the time range is defined as the same interval. The interval boundary is determined by recording the start and end positions of the interval. At the same time, a unified labeling process is performed on the combined data in each interval so that the data in the same interval have a consistent identifier.

[0052] After the intervals are divided, joint constraint integration processing can be performed on the degree of feature consistency and spatial overlap change trend within each interval. Specifically, the degree of feature consistency and spatial overlap change trend of all time positions within the interval are centrally organized, and an interval-level association description structure is established. This allows the data within the interval to express the combination relationship between feature stability and spatial change in a unified form. Then, the association markers of each interval are connected and organized according to the time sequence to form the connection relationship between continuous intervals, thereby constructing an association judgment constraint relationship covering continuous time series.

[0053] The continuous identification dataset represents the entire set of identification data acquired within a continuous time range. Continuous identification data represents hazard identification information at a single time location. Traversal processing represents the process of reading and processing data one by one in chronological order. Feature consistency preservation degree represents the result obtained through the stability analysis of feature changes over a continuous time period. Spatial overlap change trend represents the change description obtained through continuous tracking of overlapping changes in spatial regions. Feature consistency preservation degree sequence and spatial overlap change trend sequence represent the sequence expression of the two types of change information in the time dimension. Synchronization alignment processing represents the process of corresponding the two types of sequences in terms of position through time index. Corresponding combination sequence represents the data structure that contains both feature change information and spatial change information at each time location. Interval division processing represents the segmentation of the time series according to the continuity of change. Joint constraint integration represents the unified organization and expression of the two types of change information within the interval. Unified association label represents the assignment of a consistent identifier to the data within the interval. Connection expression represents the temporal sequential connection organization of each interval. The association judgment constraint relationship covering the continuous time series represents the overall constraint expression form composed of multiple interval association structures, used to describe the joint constraint relationship of feature change and spatial change in the time dimension, and to provide a constraint basis for the target association strategy in fire hazard identification records.

[0054] S5. By combining the changing trends of the spatial evolution trajectory group and the cross-sequence consistent identifier vector group, the association determination constraint relationship is updated, and dynamic control processing is performed through artificial intelligence to make the target association range adaptively adjusted, thereby completing the continuous control of fire hazard identification records.

[0055] In this embodiment, S5 specifically refers to: By combining the changing trends of the spatial evolution trajectory group and the cross-sequence consistent identifier vector group, the spatial region changing trend in the spatial evolution trajectory group and the uniform dimension feature vector changing trend in the cross-sequence consistent identifier vector group are extracted in chronological order. The spatial region changing trend and the uniform dimension feature vector changing trend are synchronized and aligned. The changing trends at each time position are combined accordingly to form a changing trend combination sequence. The association determination constraint relationship is updated based on the combination sequence of changing trends. By dividing the combination sequence of changing trends within a continuous time period into intervals, the association determination constraint relationship is adjusted according to the changing trend within the interval. At the same time, artificial intelligence is called to continuously adjust the updated association determination constraint relationship, and the target association range is divided and reconstructed based on the recognition results to achieve dynamic control processing.

[0056] The changing trends of spatial evolution trajectory groups and cross-sequence consistent identifier vector groups can be constructed through continuous time series analysis. Specifically, this involves performing position tracking processing on spatial regions at continuous time locations within the spatial evolution trajectory group to extract the direction of movement and expansion state of the spatial regions in the time dimension. Difference analysis of continuous position changes forms the spatial region change trend. Simultaneously, continuous time comparison processing is performed on the unified dimension feature vectors in the cross-sequence consistent identifier vector group. By performing dimensional change analysis on the feature vectors of adjacent time locations, the direction and stability of feature changes are extracted, forming a unified dimension feature vector change trend. Subsequently, a time correspondence is established based on a unified time index, and the spatial region change trend is matched with the unified dimension feature vector change trend at each time location. A combined trend sequence is constructed through a combined recording method, ensuring that each time location contains both spatial change information and feature change information. For example, in actual video surveillance experiments, a smoke area exhibits boundary expansion and stable feature distribution across consecutive image frames. By continuously tracking the changes in the spatial region boundary and comparing feature vector changes, a unified change trend sequence can be formed to describe the hazard evolution process.

[0057] After the trend combination sequence is formed, interval division processing can be performed on the data within a continuous time range. Specifically, by continuously scanning the trend combination sequence, the consistency of the spatial region trend and the uniform dimension feature vector trend in the time dimension is detected. When the trend within a continuous time range remains in a continuous or stable state, the time range is defined as the same interval, and the interval boundary is determined by recording the start and end positions of the interval. After the interval division is completed, the trend combination within each interval is subjected to centralized analysis processing. The trend within the interval is uniformly expressed and an interval-level change description structure is formed. The association judgment constraint relationship is updated based on the interval-level change description, so that the association judgment constraint relationship can reflect the change characteristics within different time intervals.

[0058] After the association determination constraint relationship is updated, artificial intelligence can be invoked to perform dynamic adjustment processing on the updated association determination constraint relationship. Specifically, by comparing and analyzing the combination sequence of change trends in historical time series with the current combination sequence of change trends, the difference information of change is extracted, and the constraint conditions in the association determination constraint relationship are continuously adjusted based on the difference of change. At the same time, combined with the unique hazard identifier and hazard category information corresponding to the identification results, the target association range is divided. By adjusting the boundary and reconstructing the range of association in continuous time series, the target association range is dynamically updated with the change trend. For example, in the actual monitoring process, when the smoke area changes from local diffusion to rapid expansion, the association determination constraint relationship can be adjusted by comparing the change trend, and the target association range can be expanded simultaneously, so that the same hazard remains continuously associated during the expansion phase.

[0059] The changing trends of the spatial evolution trajectory group and the cross-sequence consistent identifier vector group represent the continuous changes in spatial location and feature changes in the time dimension. The spatial region changing trend represents the location migration and morphological change process of the spatial region in continuous time. The changing trend of the unified dimension feature vector represents the direction and stability of the feature vector in continuous time. The synchronous alignment processing represents the processing of corresponding the spatial region changing trend and the unified dimension feature vector changing trend to each time position based on the time index. The changing trend combination sequence represents the data structure that simultaneously contains the spatial changing trend and the feature changing trend at each time position. The association determination constraint relationship represents the structured expression result used to constrain the association conditions between continuously identified data.

[0060] Interval division represents segmenting the time series based on the continuity of the changing trend. Adjustment processing represents updating and correcting the correlation judgment constraint relationship based on the changing trend within the interval. The updated correlation judgment constraint relationship represents the constraint expression result that can reflect the changing characteristics of different time intervals. The target correlation range represents the time range or spatial range used to merge continuous identification data in the fire hazard identification record. Range division and range reconstruction represent the process of redefining and adjusting the boundary of the target correlation range according to the changing trend. By continuously updating the correlation judgment constraint relationship and performing dynamic control processing, the target correlation range can adapt to the evolution process of the hazard, thereby realizing the continuous control and dynamic adjustment of the fire hazard identification record.

[0061] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions according to the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. Computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless means (e.g., infrared, wireless, microwave, etc.). A computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. Available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media. Semiconductor media can be solid-state drives.

[0062] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0063] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0064] In the several embodiments provided in this application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0065] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0066] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0067] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for identifying and recording fire hazards based on artificial intelligence, characterized in that, Specifically, the following steps are included: S1. Based on artificial intelligence, extract unified dimension feature vectors from continuously acquired fire hazard identification data and assign them unique identifier codes to form a cross-sequence consistent identifier vector group. At the same time, continuously map the overlapping relationship of corresponding spatial regions to generate a spatial evolution trajectory group. S2. Synchronously compare the cross-sequence consistent identifier vector group with the spatial evolution trajectory group. By calculating the coupling relationship between the degree of feature consistency maintenance and the spatial overlap change trend, determine whether there is a situation where the feature expression is consistent and the spatial overlap relationship gradually changes, and generate a state identifier. S3. When there are consistent feature expressions and the spatial overlap relationship gradually changes, artificial intelligence is invoked to perform unified mapping processing on cross-sequence consistent identifier vector groups to generate a unique hidden danger identifier sequence for determining the identification result. S4. Based on the unique hazard identification sequence and spatial evolution trajectory group, the continuous identification data is collected and processed, and the association judgment constraint relationship is constructed based on the degree of feature consistency maintenance and spatial overlap change trend. The target association strategy in the fire hazard identification record is adjusted according to the identification results. S5. Combining the changing trends of the spatial evolution trajectory group and the cross-sequence consistent identifier vector group, the association judgment constraint relationship is updated, and dynamic control processing is performed through artificial intelligence to complete the continuous control of fire hazard identification records.

2. The fire hazard identification and recording method based on artificial intelligence according to claim 1, characterized in that, S1 specifically refers to: Based on artificial intelligence, feature extraction processing is performed on continuously acquired fire hazard identification data. By performing multi-scale feature mapping on image data and performing unified dimension compression processing, a unified dimension feature vector is generated. The unified dimension feature vector is then numbered according to time sequence to assign a unique identifier code. The uniform dimension feature vector and the unique identifier code are arranged sequentially in chronological order, and then combined according to the same number order to form a cross-sequence consistent identifier vector group; For each unified dimension feature vector in the cross-sequence consistent identifier vector group, the corresponding spatial region location information is extracted. By aligning the spatial regions in adjacent time series and calculating the area ratio of the overlapping part of the spatial regions, the overlapping range is determined. Then, according to the order of change of the overlapping range, the spatial regions in the continuous time series are matched and connected one by one to establish a continuous correspondence between spatial regions and realize the continuous mapping of the overlapping relationship of spatial regions. Based on the continuous mapping results of the overlapping relationships of spatial regions, the spatial regions are connected in chronological order. By recording the location change paths of the spatial regions in the continuous time series, a set of spatial evolution trajectories is generated.

3. The fire hazard identification and recording method based on artificial intelligence according to claim 1, characterized in that, S2 specifically includes the following steps: S201. Synchronous comparison is performed between the cross-sequence consistent identifier vector group and the spatial evolution trajectory group. The feature vectors of each unified dimension in the cross-sequence consistent identifier vector group are aligned one by one with the corresponding spatial regions in the spatial evolution trajectory group according to the time order. A one-to-one correspondence between feature information and spatial evolution trajectory is established based on the unique identifier code. S202. After completing the synchronization comparison, by performing adjacent difference analysis on the unified dimension feature vectors of continuous time series in the cross-sequence consistent identifier vector group, the continuous time period with difference changes below the preset change range is divided into stable intervals of feature consistency maintenance. At the same time, the spatial region overlap relationship of continuous time series in the spatial evolution trajectory group is extracted, and the continuous time period with continuous change of overlap relationship is divided into continuous change intervals of spatial overlap change trend. The stable intervals and continuous change intervals are matched in chronological order to form a coupling relationship. S203. Based on the coupling relationship, the continuous time series is judged and processed. By comparing the time coverage of the stable interval and the continuous change interval one by one, when the coverage of the stable interval and the coverage of the continuous change interval meet the continuous correspondence in the same time period, it is judged that there is a situation where the feature expression is consistent and the spatial overlap relationship gradually changes. The time series in the corresponding time period is assigned a preset mark value to generate a status identifier.

4. The fire hazard identification and recording method based on artificial intelligence according to claim 3, characterized in that, S202 specifically refers to: After completing the synchronization comparison, the differences of the unified dimension feature vectors of the continuous time series in the cross-sequence consistent identifier vector group are calculated one by one, and the difference values ​​between adjacent unified dimension feature vectors are arranged in chronological order. The time periods in which the continuous difference values ​​are within the preset range of change are segmented and marked to determine the stable interval of the degree of feature consistency maintenance. The spatial region overlap relationship of continuous time series in the spatial evolution trajectory group is extracted frame by frame. By obtaining the change in the spatial region overlap range in adjacent time series and arranging them in chronological order, continuous time periods with unidirectional change are segmented and marked to determine the continuous change interval of the spatial overlap change trend. The stable intervals that maintain the consistency of features in chronological order are aligned with the continuous intervals that show spatial overlap. By comparing the start and end positions of each time period, corresponding matching is performed on the intervals that overlap in time range to form a coupling relationship.

5. The fire hazard identification and recording method based on artificial intelligence according to claim 1, characterized in that, S3 specifically refers to: When consistent feature expressions and gradually changing spatial overlap relationships occur, artificial intelligence is invoked to perform unified mapping processing on cross-sequence consistent identifier vector groups. By extracting the unified dimension feature vectors of continuous time series in the cross-sequence consistent identifier vector group according to the time order, the similarity of the unified dimension feature vectors in adjacent time series is calculated one by one, and the similarity calculation results are arranged in time order. Continuous time periods with similarity within a preset similarity range are segmented and grouped, and the unified dimension feature vectors in the same segment are grouped together. Based on the grouping results, a corresponding association set between unified dimension feature vectors is established to form a unified mapping relationship. Based on the unified mapping process, the unified dimension feature vectors in each unified mapping relationship are arranged in chronological order, and a corresponding unique identifier code is assigned to each unified mapping relationship. The unique identifier codes in the corresponding time series are combined sequentially to generate a unique hidden danger identifier sequence. Based on the unique hazard identification sequence, the unified dimension feature vector in the cross-sequence consistent identification vector group is assigned to the same fire hazard. The identification result is output according to the correspondence.

6. The fire hazard identification and recording method based on artificial intelligence according to claim 1, characterized in that, S4 specifically includes the following steps: S401. Based on the unique hazard identification sequence and the spatial evolution trajectory group, the continuous identification data is collected and processed. The continuous identification data is divided into time order according to the unique hazard identification sequence, and the continuous identification data with the same unique hazard identification is matched with the corresponding spatial area in the spatial evolution trajectory group. The matched continuous identification data is combined and collected to form a continuous identification data set with the corresponding unique hazard identification. S402. Based on the continuous identification data set, extract the feature consistency retention degree and spatial overlap change trend corresponding to each continuous identification data set, and arrange the feature consistency retention degree and spatial overlap change trend synchronously in chronological order. By jointly constraining the feature consistency retention degree and spatial overlap change trend, construct the association judgment constraint relationship. S403. Based on the association determination constraint relationship and combined with the identification results, adjust the target association strategy in the fire hazard identification record. By dividing the association range of continuous identification data in the continuous identification data set, perform unified association processing on continuous identification data that meets the association determination constraint relationship, and perform separation processing on continuous identification data that does not meet the association determination constraint relationship, so as to achieve the adjustment of the target association strategy.

7. The fire hazard identification and recording method based on artificial intelligence according to claim 6, characterized in that, S402 specifically refers to: Based on the continuous recognition dataset, the continuous recognition data is traversed in chronological order to extract the feature consistency retention degree and spatial overlap change trend corresponding to each time position. The extraction results are then serialized and arranged in chronological order to form a feature consistency retention degree sequence and a spatial overlap change trend sequence. The sequence of feature consistency preservation degree and the sequence of spatial overlap change trend are synchronized and aligned at the same time position. The feature consistency preservation degree and spatial overlap change trend at each time position are combined and recorded to form a corresponding combined sequence arranged in chronological order. For continuous time periods in the corresponding combined sequence, interval division processing is performed. The degree of feature consistency and spatial overlap change trend within each interval are jointly constrained and integrated. By uniformly associating and labeling the data within the same interval, and connecting and expressing the association labels of each interval in chronological order, an association judgment constraint relationship covering continuous time series is constructed.

8. The fire hazard identification and recording method based on artificial intelligence according to claim 1, characterized in that, S5 specifically refers to: By combining the changing trends of the spatial evolution trajectory group and the cross-sequence consistent identifier vector group, the spatial region changing trend in the spatial evolution trajectory group and the uniform dimension feature vector changing trend in the cross-sequence consistent identifier vector group are extracted in chronological order. The spatial region changing trend and the uniform dimension feature vector changing trend are synchronized and aligned. The changing trends at each time position are combined accordingly to form a changing trend combination sequence. The association determination constraint relationship is updated based on the combination sequence of changing trends. By dividing the combination sequence of changing trends within a continuous time period into intervals, the association determination constraint relationship is adjusted according to the changing trend within the interval. At the same time, artificial intelligence is called to continuously adjust the updated association determination constraint relationship, and the target association range is divided and reconstructed based on the recognition results to achieve dynamic control processing.

Images