Video lossless compression transmission and storage method and device

Abstract

The invention relates to a video lossless compression transmission and storage method and device. The method comprises the following steps: analyzing an obtained video frame sequence; executing quality pre-verification, and if yes, determining that the video frame sequence is a first frame sequence; based on the scene detection target, determining a target attention area in the plurality of first sequence frames; selecting a plurality of target feature frames from the plurality of first sequence frames to execute importance evaluation, and determining corresponding feature levels; mapping the plurality of target feature frames into a one-dimensional pixel sequence; dividing the one-dimensional pixel sequence into a plurality of pixel segments, and performing cross-frame aggregation on the plurality of pixel segments according to the feature level to obtain a plurality of feature sequences; and carrying out hierarchical compression coding on the plurality of feature sequences. By adopting the method, the video can be subjected to quality pre-verification, the target concerned area can be adaptively identified, and differential compression and priority transmission are performed, so that the transmission quality and the storage efficiency of the inspection video can be improved in a complex electromagnetic environment and a limited network condition.

Description

Technical Field

[0001] This application relates to the field of video transmission and storage technology, and in particular to a method and apparatus for lossless video compression transmission and storage. Background Technology

[0002] With the intelligent development of power systems, video monitoring has become an important means of power equipment inspection and operational status assessment. By deploying video equipment to capture real-time images of substations, transmission lines, and distribution equipment, and transmitting the video data to the monitoring center for identification and analysis, the automation and safety of power inspections can be effectively improved.

[0003] However, power line inspection scenarios exhibit significant electromagnetic complexity and dynamic environmental characteristics. On one hand, strong electromagnetic interference signals generated during high-voltage power transmission, circuit breaker operation, and transformer operation directly affect the imaging quality of video acquisition equipment and the stability of communication modules, leading to problems such as brightness fluctuations, increased frequency domain noise, frame timing errors, and image distortion in video frames. If these interfered video frames directly enter the compression and transmission stage, the motion prediction and redundancy elimination mechanisms of the compression algorithm will further amplify frame-level noise and structural anomalies, causing interference errors to propagate in subsequent frames, resulting in blurred video images, prominent block effects, or pseudo-motion phenomena, thus severely affecting the clarity and readability of the inspection video. On the other hand, power line inspection systems are often deployed in areas with limited communication conditions. In scenarios such as power transmission channels in mountainous areas and remote substations, video transmission is often limited by wireless network bandwidth, equipment load, and signal attenuation. This often results in bandwidth constraints, link fluctuations, and packet loss. In such cases, traditional video transmission solutions typically use a uniform compression rate and a fixed transmission order, which cannot be adaptively optimized according to the importance of the video content or scene characteristics. This leads to a decrease in video quality or delayed transmission of key target areas (such as conductor joints, insulator surfaces, and hardware connection points), affecting the accuracy of defect detection. Furthermore, when electromagnetic interference, bandwidth constraints, and traditional compression mechanisms are combined, the negative effects are further amplified, significantly impacting the quality of the video transmitted to the backend. Summary of the Invention

[0004] Therefore, it is necessary to provide a lossless video compression transmission and storage method and apparatus that can perform video quality pre-verification, adaptively identify target areas of interest, and perform differentiated compression and priority transmission to address the problems of video quality degradation caused by electromagnetic interference, transmission delay caused by bandwidth limitation, and compression redundancy amplification. This would improve the transmission quality and storage efficiency of inspection videos under complex electromagnetic environments and limited network conditions.

[0005] On the one hand, a method for lossless video compression transmission and storage is provided, applied at the sending end, the method comprising: Acquire and parse video data to obtain a video frame sequence consisting of multiple video frames; A quality pre-check is performed on the video frame sequence. If the quality pre-check passes, the video frame sequence is determined to be the first frame sequence, and the corresponding video frame is identified as the first sequence frame. Based on the scene detection target, the target interest region and the corresponding region feature parameters in multiple first sequence frames are determined, and multiple target feature frames are selected from multiple first sequence frames according to the change characteristics of the target interest region. An importance assessment is performed on the target interest regions identified in multiple target feature frames to determine the corresponding feature levels; Based on the frame quality of the target feature frames, the change characteristics of the target region of interest, and the feature level, multiple target feature frames are mapped into a one-dimensional pixel sequence through a pre-constructed dimensionality reduction mapping model. The one-dimensional pixel sequence is divided into multiple pixel segments according to the region feature parameters, and the multiple pixel segments are aggregated across frames according to the feature level to obtain multiple feature sequences; Based on the feature level and the current environmental state, the encoding method and transmission order of multiple feature sequences are determined, and the multiple feature sequences are layered, compressed, encoded, and encapsulated into a video aggregated data stream, which is then sent to the receiving end.

[0006] On the other hand, a lossless video compression transmission and storage method is provided, applied at a receiving end, the method comprising: Receive the video aggregation data stream transmitted from the sender, parse it to obtain multiple encoded feature sequences and determine the corresponding feature levels; Integrity checks are performed on multiple encoded feature sequences. In response to the integrity check passing, the decoding method of the multiple encoded feature sequences is determined based on the feature level, and layered decoding is performed to obtain multiple decoded feature sequences. According to the feature level, the multiple decoded feature sequences are stored in the corresponding hierarchical storage area, and the spatiotemporal correlation information of the sequences is determined according to the regional feature parameters of the multiple decoded feature sequences to construct a sequence storage correlation index; In response to receiving a video reading instruction, multiple decoded feature sequences are retrieved based on the sequence storage association index and concatenated to obtain a fused feature sequence; Based on a preset dimensionality reduction mapping inverse model, multiple fused feature sequences are restored into multiple two-dimensional video frames, and frame interpolation and correction are performed to generate a video frame sequence.

[0007] In another aspect, a lossless video compression transmission and storage device is provided, comprising a transmitting end and a receiving end, wherein the transmitting end is used to implement the method applied to the transmitting end as described in any one of the above claims, and the receiving end is used to implement the above-described method applied to the receiving end.

[0008] The aforementioned video lossless compression transmission and storage method and apparatus, by performing quality pre-verification on the video frame sequence at the sending end, only processes video frame sequences that meet the quality requirements, thus preventing low-quality frames affected by electromagnetic interference from directly entering the compression link. This reduces the error propagation caused by abnormal frames during predictive coding and improves the stability of the video source input quality. Furthermore, this application determines the target interest region in the video frame based on scene detection targets, focusing on key components in power inspection scenarios. It then divides the video content into feature sequences with different feature levels according to their change characteristics and importance. Through dimensionality reduction mapping and cross-frame pixel segment aggregation, different levels of region information can be independently encoded and managed. When bandwidth is limited or network environment fluctuates, the encoding method and transmission order can be adaptively determined based on feature level and current environmental state, ensuring that the content of the target area of ​​interest receives priority encoding resources and transmission guarantees, avoiding the decrease in clarity and information delay of the target area of ​​interest caused by traditional uniform compression transmission methods. Simultaneously, through layered compression encapsulation of multiple feature sequences, the flexibility and robustness of transmission scheduling can be improved, mitigating the impact of link instability on overall video quality. Furthermore, this application achieves comprehensive suppression of the combined effects of electromagnetic interference, compression error accumulation, and bandwidth limitation by coordinating video source quality control, content importance classification, target area of ​​interest extraction and aggregation, and environmental adaptive transmission mechanisms. Therefore, this application can effectively suppress video quality degradation under the influence of electromagnetic interference in power inspection scenarios, and maintain the clarity and legibility of the target area of ​​interest under bandwidth-limited conditions, thereby weakening the video quality degradation caused by the superposition of multiple factors and achieving high fidelity, high robustness, and high reliability in power inspection video transmission. Attached Figure Description

[0009] Figure 1 This is a diagram illustrating the application environment of a video lossless compression transmission and storage method in one embodiment. Figure 2 This is a flowchart illustrating a lossless video compression transmission and storage method applied to the sending end in one embodiment; Figure 3 This is a schematic diagram illustrating the steps of dimensionality reduction mapping in one embodiment; Figure 4 This is a flowchart illustrating a lossless video compression transmission and storage method applied at the receiving end in one embodiment. Figure 5This is a structural block diagram of a video lossless compression transmission and storage device in one embodiment. Detailed Implementation

[0010] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0011] This application provides a lossless video compression transmission and storage method, which can be applied to, for example... Figure 1 In the application environment shown, the video acquisition terminal 100 can include, but is not limited to, fixed surveillance cameras, pan-tilt video acquisition devices, UAV inspection video acquisition modules, and mobile terminal video acquisition modules, etc., for acquiring raw video data containing target equipment and its operating environment at the power inspection site. The video acquisition terminal 100 can be set with different resolutions, frame rates, and exposure parameters according to the inspection task requirements to adapt to imaging needs under different electromagnetic interference intensities and lighting conditions. The transmitting terminal 101 can include, but is not limited to, video processing servers, embedded edge computing nodes, industrial control hosts, or terminal intelligent processing modules, for processing the video data acquired by the video acquisition terminal 100, including parsing, quality pre-verification, feature extraction, feature level evaluation, dimensionality reduction mapping, and layered compression encoding. Furthermore, the transmitting end 101 can also be a microserver or embedded edge processing device integrated on the video acquisition end 100, thereby realizing an edge intelligent processing architecture that integrates video acquisition and transmission; after completing encoding and layered encapsulation, the transmitting end 101 can transmit the generated video aggregated data stream to the receiving end 102 through a wired or wireless network; the receiving end 102 can be, but is not limited to, a video monitoring center server, a cloud data processing platform, an inspection management terminal, or a video storage server, used to receive and parse the video aggregated data stream from the transmitting end 101, perform integrity verification, layered decoding and hierarchical storage on the received data, and splice and restore multiple decoded feature sequences when needed to generate a high-fidelity video frame sequence for display or analysis. Through the above structure, the video acquisition end 100, the transmission end 101, and the receiving end 102 form a complete lossless video compression transmission and storage process from video acquisition, feature layer encoding, network transmission to decoding and storage at the receiving end. Combined with the lossless video compression transmission and storage method provided in this application, high-quality preprocessing, differentiated encoding, and reliable transmission of video data can be achieved in power inspection environments with strong electromagnetic interference and limited communication bandwidth, thereby improving the clarity, stability, and availability of inspection videos.

[0012] In one embodiment, such as Figure 2 As shown, a lossless video compression transmission and storage method is provided, which can be applied to... Figure 1Taking the sender 101 as an example, the explanation includes the following steps: Step 201: Acquire and parse the video data to obtain a video frame sequence consisting of multiple video frames; Step 202: Perform quality pre-check on the video frame sequence. If the quality pre-check passes, determine that the video frame sequence is the first frame sequence and identify the corresponding video frame as the first sequence frame. Step 203: Based on the scene detection target, determine the target interest region and the corresponding region feature parameters in multiple first sequence frames, and select multiple target feature frames from multiple first sequence frames according to the change characteristics of the target interest region. Step 204: Perform importance assessment on the target interest regions identified in multiple target feature frames to determine the corresponding feature levels; Step 205: Based on the frame quality of the target feature frames, the change characteristics of the target region of interest, and the feature level, multiple target feature frames are mapped into a one-dimensional pixel sequence through a pre-constructed dimensionality reduction mapping model. Step 206: Divide the one-dimensional pixel sequence into multiple pixel segments according to the region feature parameters, and aggregate the multiple pixel segments across frames according to the feature level to obtain multiple feature sequences; Step 207: Based on the feature level and the current environment state, determine the encoding method and transmission order of multiple feature sequences, encapsulate the multiple feature sequences into a video aggregated data stream after hierarchical compression encoding, and send it to the receiving end.

[0013] In the aforementioned lossless video compression transmission and storage method, by performing quality pre-verification on the video frame sequence at the sending end, and only processing video frame sequences that meet the quality requirements, it is possible to avoid low-quality frames affected by electromagnetic interference from directly entering the compression link. This reduces the error propagation caused by abnormal frames during predictive coding and improves the stability of the video source input quality. Based on this, this application determines the target interest region in the video frame based on scene detection targets, focusing on key components in power inspection scenarios. Furthermore, based on change characteristics and importance, the video content is divided into feature sequences with different feature levels. Then, through dimensionality reduction mapping and cross-frame pixel segment aggregation, regional information of different levels can be independently encoded and managed. When bandwidth is limited or network environment fluctuates, the encoding method and transmission order can be adaptively determined based on feature level and current environmental state, ensuring that the content of the target area of ​​interest receives priority encoding resources and transmission guarantees, avoiding the decrease in clarity and information delay of the target area of ​​interest caused by traditional uniform compression transmission methods. Simultaneously, through layered compression encapsulation of multi-feature sequences, the flexibility and robustness of transmission scheduling can be improved, mitigating the impact of link instability on overall video quality. Furthermore, this application achieves comprehensive suppression of the combined effects of electromagnetic interference, compression error accumulation, and bandwidth limitation by coordinating video source quality control, content importance classification, target area of ​​interest extraction and aggregation, and environmental adaptive transmission mechanisms. Therefore, this application can effectively suppress video quality degradation under the influence of electromagnetic interference in power inspection scenarios, and maintain the clarity and legibility of the target area of ​​interest under bandwidth-limited conditions, thereby weakening the video quality degradation caused by the superposition of multiple factors and achieving high fidelity, high robustness, and high reliability in power inspection video transmission.

[0014] In one embodiment, performing quality pre-check on a video frame sequence includes: Based on the detection scenario, the scene frame quality detection indicators are determined. The detection scenario includes at least the power inspection scenario, and the scene frame quality detection indicators include at least the following: inter-frame brightness stability, frequency domain noise energy, and frame temporal continuity. The average brightness change rate between frames is obtained from the frame brightness statistics of the video frame sequence, and the inter-frame brightness stability score of the video frame sequence is determined. The frequency domain noise energy score of the video frame sequence is determined by obtaining the proportion of high-frequency noise components based on the spectral energy distribution of the video frame sequence. The continuity of inter-frame time intervals and frame numbers is determined based on the timestamp sequence of the video frame sequence, and a frame temporal continuity score of the video frame sequence is generated. The scene frame quality score of the video frame sequence is determined based on the inter-frame luminance stability score, frequency domain noise energy score, and frame temporal continuity score. Based on the basic noise characteristics, exposure settings, and ambient light intensity of the video acquisition equipment, the ground-state frame quality score of the video frame sequence is determined. A comprehensive quality score for the video frame sequence is generated based on the scene frame quality score and the ground state frame quality score. If the overall quality score is greater than or equal to the overall quality score threshold, the quality pre-verification is deemed to have passed. If the overall quality score is less than the overall quality score threshold, the quality pre-verification is deemed to have failed.

[0015] Specifically, in this embodiment, by performing quality pre-verification on the video frame sequence, the quality evaluation is extended from a single frame to the video sequence level, achieving a comprehensive assessment of the quality of the power inspection video source. Combining the basic noise, exposure settings, and ambient light intensity of the video acquisition equipment, the imaging baseline quality of the equipment in the current scene is determined, achieving dual quality constraints based on equipment status and scene characteristics. Furthermore, through joint evaluation of scene frame quality scoring and ground-state frame quality scoring, frame sequences that do not meet quality requirements are pre-screened during the video generation stage, reducing the entry of low-quality frames affected by electromagnetic interference into the encoding and compression stages, and preventing noise accumulation and compression impairment from further degrading transmission quality. Through the above mechanisms, this embodiment improves the front-end imaging quality and subsequent analysis reliability of power inspection videos, while effectively reducing data redundancy and misjudgment risks in the transmission link, and improving video availability and inspection accuracy in complex electromagnetic environments.

[0016] It is worth noting that the power inspection scenario, characterized by strong electromagnetic interference, complex lighting, and equipment vibration, is used as the detection scenario. The scene frame quality detection index is determined based on the characteristics of this detection scenario, including: Inter-frame brightness stability is used to characterize whether the brightness change of a video is smooth and continuous between adjacent frames, reflecting whether there is flicker, sudden exposure, or lighting disturbance during the imaging process. In power inspection scenarios, arc discharge, equipment switching, and electromagnetic pulses can cause instantaneous brightness jumps in CMOS (Complementary Metal-Oxide-Semiconductor) or CCD (Charge-Coupled Device) photosensitive elements. Therefore, inter-frame brightness stability can characterize whether the video sequence has been affected by transient electromagnetic interference or exposure instability. Frequency domain noise energy is used to characterize the strength of high-frequency noise components in video images, and thus reflects whether the imaging signal is contaminated by electromagnetic interference. In power inspection scenarios, high-voltage power transmission, circuit breaker operation, and high-current interruption will generate broadband electromagnetic radiation, causing phenomena such as particle noise and stripe interference in the acquired images. Therefore, frequency domain noise energy reflects the signal-to-noise ratio and interference noise intensity of the image. Frame timing continuity is used to characterize whether the output and transmission times of video frames maintain the correct order and interval, reflecting whether frame skipping, out-of-order, or frame loss occurs during the acquisition and buffering process. In power inspection environments, equipment vibration, electromagnetic interference, and network link jitter can all lead to frame loss or buffering delays. Therefore, timing continuity can be used to detect whether the video is affected by unstable transmission or abnormal buffering at the acquisition end.

[0017] In one specific embodiment, the inter-frame average luminance change rate is obtained based on the frame luminance statistics of the video frame sequence, and the inter-frame luminance stability score of the video frame sequence is determined, including: The brightness distribution of adjacent frames in a video frame sequence is statistically analyzed, and the average brightness of the i-th frame is denoted as L. i Then the average brightness change rate between frames can be expressed as: ; Where ΔL represents the average inter-frame luminance change rate of the video frame sequence, and N represents the total number of frames in the video frame sequence; the inter-frame luminance stability score of the video frame sequence is determined based on the average inter-frame change rate, and is expressed as: ; Among them, S L k represents the inter-frame luminance stability score of the video frame sequence. L The brightness fluctuation attenuation coefficient is taken as 10-30, when S L A value close to 100 indicates good brightness stability, while a value below 70 indicates significant flickering or electromagnetic interference.

[0018] In one specific embodiment, the proportion of high-frequency noise components is obtained based on the spectral energy distribution of the video frame sequence, including: Performing a Fast Fourier Transform on each frame of the video frame sequence yields the spectral energy distribution. High-frequency energy is defined as the energy component whose frequency amplitude exceeds 0.6 times the total frequency bandwidth. The proportion of high-frequency noise energy is then expressed as: ; Among them, R H This represents the proportion of high-frequency noise energy in the video frame sequence. `f` represents the spatial frequency amplitude, reflecting the rate of change in image grayscale; a larger value corresponds to more drastic changes in detail or noise components in the image. `E(f)` represents the energy value of the spatial frequency component at frequency `f`, which is the square of the spectral energy amplitude at frequency `f` after Fast Fourier Transform, used to characterize the contribution of this frequency component to the overall image energy. max The maximum spatial frequency that can be sampled (i.e., the Nyquist frequency) is determined by the image resolution and sampling rate. The frequency domain noise energy score is determined based on the proportion of noise energy, and is expressed as follows: ; Among them, S N k represents the frequency domain noise energy score of the video frame sequence. N This represents the noise sensitivity coefficient, preferably between 0.8 and 1.2; if S N If the value is less than 60, it can be determined that the noise in the image is significantly enhanced, indicating a risk of electromagnetic interference.

[0019] In one specific embodiment, the continuity of inter-frame time intervals and frame numbers is determined based on the timestamp sequence of the video frame sequence, and a frame temporal continuity score of the video frame sequence is generated, which can be expressed as: ; Among them, S T T represents the frame temporal continuity score of the video frame sequence. i This represents the i-th video frame in the video frame sequence, where T0 represents the desired frame interval, determined by the acquisition frame rate, and ΔT. i This represents the inter-frame time interval between the (i+1)th video frame and the ith video frame in the video frame sequence; if S T If the value is less than 80, it can be determined that the video frame sequence has problems such as frame skipping, frame loss, or out-of-order issues.

[0020] In one specific embodiment, the scene frame quality score of the video frame sequence is determined based on the inter-frame luminance stability score, frequency domain noise energy score, and frame temporal continuity score, and is expressed as follows: ; Among them, S S W represents the scene frame quality score of this video frame sequence. L W T W N These represent the corresponding scoring weight coefficients, with 0.35, 0.25, and 0.40 being preferred.

[0021] In one specific embodiment, the fundamental frame quality score of the video frame sequence is determined based on the fundamental noise characteristics, exposure settings, and ambient light intensity of the video acquisition device, including: Acquire basic noise characteristics of the video acquisition device, including the sensor's signal-to-noise ratio, dark current noise level, and quantization accuracy, to characterize the lower limit of the imaging noise of the device when there is no external interference. Specifically, when the device has a high signal-to-noise ratio and a stable noise level, the basic noise score is high. When the sensor is aging, the dark current is large, or the noise fluctuates randomly, the basic noise score decreases. Obtain the exposure parameters of the current video capture device, including exposure time, ISO and aperture value, and evaluate the adaptability of the exposure settings in combination with the actual ambient light level. Specifically, when the exposure time matches the ambient brightness and the ISO setting is reasonable, the brightness distribution of the image is uniform and the details are clear, resulting in a higher exposure adaptability score. When the exposure is too short, the image is too dark, or the exposure is too long, the image is overexposed, the exposure adaptability score is reduced accordingly. The ambient light intensity is estimated by a light intensity sensor or the average brightness of video frames. Specifically, when the ambient light intensity is within the imaging illuminance range recommended by the device, the ambient light score is higher, and when the light is insufficient or there is strong glare, the ambient light score is lower. The base frame quality score of the video frame sequence is determined by comprehensively evaluating the base noise score, exposure adaptation score, and ambient light score. This score reflects the baseline imaging quality that the device can achieve under ideal conditions and is used to measure the hardware imaging capability of the video source and the suitability of the current acquisition conditions. Specifically, when the base frame quality score is higher than the set reference standard, it indicates that the video acquisition device is in a stable and usable imaging state. When the score is lower than the reference standard, it indicates that there may be problems such as excessive noise, abnormal exposure, or poor lighting under the current acquisition conditions. Based on this, the overall weight of the video frame sequence can be reduced or a re-acquisition prompt can be triggered.

[0022] In one specific embodiment, a comprehensive quality score for the video frame sequence is generated based on the scene frame quality score and the ground state frame quality score, which can be expressed as: ; Among them, S C S represents the overall quality score of the video frame sequence. S S represents the scene frame quality score of this video frame sequence. B This represents the base-state frame quality score of the video frame sequence, where α represents the scene weight coefficient, which can be between 0.6 and 0.8. Optionally, the overall quality score threshold is 80 for comparison with the overall quality score.

[0023] In one embodiment, the method further includes: In response to a failure of the quality pre-verification, multiple video frames in the video frame sequence are traversed at least based on standard anomaly detection and scene interference anomaly detection to determine the anomalous frames and their corresponding anomaly types. Based on the anomaly type, perform corresponding frame-level pre-repair processing on the abnormal frame to obtain the repaired frame and update the video frame sequence; Re-perform quality pre-check on the updated video frame sequence; If the quality pre-verification fails after a preset number of consecutive attempts, the video frame sequence is determined to be an abnormal frame sequence. The spatiotemporal correlation sequence corresponding to the abnormal frame sequence is determined and sent to the receiving end, so that the receiving end can perform frame-level reconstruction of the abnormal frame sequence based on the spatiotemporal correlation sequence and generate a reconstructed frame sequence.

[0024] Preferably, standard anomaly detection includes frame brightness anomaly detection, edge feature mutation detection, and inter-frame structural similarity detection, used to identify quality anomalies caused by overexposure, blurring, or frame skipping; scene interference anomaly detection includes electromagnetic interference noise detection, flash interference detection, and local texture distortion detection, used to identify image anomalies caused by power equipment discharge, strong light flicker, or sensor interference.

[0025] Preferably, the abnormal frames undergo corresponding frame-level pre-repair processing, including: when the abnormality type is brightness abnormality, using a grayscale adaptive correction algorithm based on inter-frame brightness compensation for brightness equalization; when the abnormality type is noise contamination, using frequency domain noise suppression or temporal filtering algorithms to reduce random noise energy; when the abnormality type is structural distortion or frame skipping abnormality, using a frame interpolation reconstruction algorithm based on inter-frame motion vector estimation to restore local texture and temporal continuity. The repaired frames replace the original abnormal frames to form an updated video frame sequence.

[0026] Preferably, the updated video frame sequence is re-performed for quality pre-verification to determine whether the repair effect meets the quality requirements. If the re-verification passes, the repaired video frame sequence is used as valid data in subsequent processes. If the quality pre-verification fails after a preset number of consecutive times (e.g., three times), the video frame sequence is determined to be an abnormal frame sequence.

[0027] Specifically, this embodiment introduces a joint mechanism of standard anomaly detection and scene interference anomaly detection when the quality pre-verification fails. This mechanism can accurately locate the abnormal frames and their types that cause quality degradation, avoiding the resource waste caused by discarding or re-acquiring the entire video in traditional solutions. Frame-level pre-repair is performed based on the anomaly type, including brightness compensation, noise suppression, texture correction, and frame skipping rearrangement, improving the quality of video materials from the source. The "detection-pre-repair-verification" quality closed loop reduces the number of low-quality frames affected by electromagnetic interference, illumination fluctuations, or acquisition link jitter from entering the compression process, reducing the quality degradation caused by noise accumulation and compression artifacts. When multiple pre-repairs still fail the quality verification, this embodiment constructs a spatiotemporal correlation sequence of abnormal frames and sends it to the receiving end. This allows the receiving end to perform targeted reconstruction based on precise temporal and spatial locations, eliminating the need for full reconstruction of the entire video, significantly reducing computational costs, while maintaining the continuity and structural consistency of video content, and improving the accuracy and efficiency of abnormal segment reconstruction.

[0028] In one specific embodiment, determining the spatiotemporal correlation sequence corresponding to the abnormal frame sequence and sending it to the receiving end includes: The system counts the continuous range of abnormal frames on the timeline, determines the start and end frame numbers of the abnormal segments, and records the corresponding video timestamp information to generate time-related information. Spatial region localization is performed for each abnormal frame. By using masks of abnormal brightness regions, noise enhancement regions, or texture distortion regions, the spatial coordinate range of the abnormal pixels is determined, forming spatial correlation information. The spatiotemporal correlation sequence consists of temporal correlation information and spatial correlation information, which are used to describe the position of the abnormal segment in time sequence and the distribution of the abnormal region in the image space. After determining the spatiotemporal correlation sequence, the abnormal frame sequence and its corresponding spatiotemporal correlation sequence are sent to the receiving end.

[0029] In a more specific embodiment, the receiving end performs frame-level reconstruction of the abnormal frame sequence based on the spatiotemporal correlation sequence to generate a reconstructed frame sequence, including: When the spatiotemporal correlation sequence indicates that abnormal frames exist continuously in the time dimension and the spatial abnormal regions are relatively concentrated, a temporal interpolation algorithm based on neighboring normal frames is used to reconstruct the abnormal regions. When the abnormal region exhibits random noise characteristics, frequency domain noise backoff and texture restoration algorithms are used to repair the local region; When the anomalous region involves structural distortion or texture loss, an image inpainting algorithm based on motion estimation and spatial context filling is used for reconstruction. Finally, the receiving end completes the partial or overall repair of the abnormal frame according to the instructions of the spatiotemporal correlation sequence, and generates a reconstructed frame sequence.

[0030] Spatiotemporal correlation sequences enable the receiver to accurately locate the time period and spatially abnormal region of abnormal frames, thereby enabling targeted reconstruction without having to perform full reconstruction of the entire video. This helps reduce the amount of reconstruction computation and improve reconstruction efficiency, while ensuring the continuity of content and the consistency of spatial structure in the reconstructed video.

[0031] In one embodiment, based on the scene detection target, the target interest region in multiple first sequence frames is determined, including: Based on historical fault records and operational risk models of the detection scenarios, risky components in multiple scenarios are identified. Multiple risk components in various scenarios, as well as multiple key objectives pre-set based on the scenario detection task, are used as scenario detection targets; Extract the corresponding feature parameters from the identified scene detection targets to generate scene detection target feature vectors; Based on the consistency between intra-frame structural features and region features, candidate region extraction is performed on multiple first sequence frames to obtain multiple candidate regions and generate corresponding candidate region vectors. Based on the similarity comparison between the candidate region vector and the scene detection target feature vector, the target region of interest is selected from multiple candidate regions, and the region feature parameters of the target region of interest are generated based on the selected candidate region vector.

[0032] It is worth noting that the target focus mode in this embodiment is highly compatible with the power inspection scenario. The video acquisition points for power inspection are strategically deployed around key equipment and risky components such as transformers, switchgear, insulators, and signal lights. The core information in the footage is concentrated, and the business objectives are clear. Based on this, this embodiment identifies risky components in the scenario through historical fault records and operational risk models, and constructs scenario detection targets by combining them with key task objectives. This makes the determination of the area of ​​interest risk-driven and task-specific, no longer relying on fixed templates or single visual features. By extracting candidate regions from the video sequence and filtering target areas of interest based on vector similarity, stable and accurate positioning of the area of ​​interest is achieved, effectively suppressing background interference and false detections. It can highlight areas containing high-risk components or potential defects. The entire process, from risk modeling to area extraction, achieves adaptive optimization, which not only improves the front-end filtering capability but also reduces the back-end recognition computational burden, improving the accuracy and operational efficiency of the power inspection system in complex scenarios.

[0033] In one specific embodiment, taking a substation equipment inspection scenario as an example, the inspection video is collected by multiple cameras deployed in the transformer area, switchgear area, and signal indicator area. The target area of ​​interest is determined from multiple first sequence frames, including: Based on the historical fault records and operational risk model of this detection scenario, risk assessments are conducted on equipment that has previously exhibited abnormal temperature rise, poor contact, or discharge signs, thereby identifying risky components in several scenarios, such as main transformer terminals, busbar connection points, surge arresters, and signal light modules. The aforementioned risky components, along with the key objectives preset according to the inspection task (such as equipment nameplates, oil level indicators, status indicator lights, etc.), are collectively used as the scene detection targets. For each scene detection target, corresponding feature parameters such as texture, color, shape, and brightness gradient are extracted. After normalization and principal component dimensionality reduction, a scene detection target feature vector is generated.

[0034] Candidate region extraction is performed on multiple first sequence frames collected. Specifically, based on the consistency between intra-frame structural features (such as edge density and regional distribution stability) and regional features, a multi-scale sliding window and segmentation fusion strategy is adopted to extract multiple candidate regions, and a corresponding candidate region vector is generated for each candidate region. Calculate the similarity between the candidate region vector and the scene detection target feature vector, such as using cosine similarity or Euclidean distance, and select the region with the highest similarity and that satisfies the spatial location constraints from multiple candidate regions as the target region of interest; For the selected target region of interest, further extract local geometric structure, gray-scale distribution and dynamic change features to generate regional feature parameters of the target region of interest for subsequent state recognition and anomaly detection; Specifically, this embodiment can automatically locate target areas containing critical equipment or risky components in complex power inspection videos, achieving targeted front-end feature extraction and selection of areas of interest, providing high-quality input data support for subsequent fault identification and intelligent diagnosis.

[0035] In a more specific embodiment, a representation of the operational risk model is given as follows: ; Among them, S i,t This represents the risk score of the i-th component in the detection scenario at time t. For the sigmoid function, x i,t This represents a risk feature vector constructed from several risk parameters, including but not limited to: historical failure rate, recent periodic alarm count, load rate, ambient temperature and humidity, and vibration amplitude; w T The transpose of the weight matrix representing the risk feature vector and b, representing the bias term, are both obtained by logistic regression maximum likelihood estimation of labeled samples.

[0036] In one embodiment, multiple target feature frames are selected from multiple first sequence frames based on the changing characteristics of the target region of interest, including: Inter-frame difference calculation is performed on the target interest region in multiple first sequence frames to obtain the rate of change parameter characterizing the dynamic change degree of the target interest region. The inter-frame difference calculation can be performed by differentiating the pixel grayscale of the target interest region to obtain the change energy as the rate of change parameter, so as to characterize the dynamic activity degree of the target in the time dimension. Based on preset image quality indicators, the image quality parameters of the target interest region in multiple first sequence frames are determined. The image quality parameters can be comprehensively evaluated based on a weighted average of brightness uniformity, sharpness (gradient variance), noise estimation, or contrast indicators. A change feature score is generated based on the change rate parameter and image quality parameter to characterize the change features of the target region of interest in the first frame sequence. The change feature score is preferably obtained by weighted fusion. Based on the change feature score, multiple first sequence frames are sorted, and several frames with the highest change feature score are selected as target feature frames. The number of target feature frames is preferably 3-5 frames.

[0037] Specifically, in this embodiment, by introducing a comprehensive evaluation mechanism that combines the rate of change parameter of the target area of ​​interest with the image quality parameter, key frames with significant dynamic changes and high visual clarity can be automatically identified in multi-frame videos. This effectively avoids the problem of invalid frame selection caused by motion blur or lighting interference, making the extracted target feature frames more representative of the real features of target state changes, thereby improving the accuracy and stability of subsequent recognition, detection or diagnosis tasks.

[0038] In one embodiment, an importance assessment is performed on the target interest regions identified in multiple target feature frames to determine the corresponding feature levels, including: For multiple target interest regions, the target importance of the corresponding scene detection targets is obtained. The target importance is determined based on one or more of the following: the frequency of occurrence of the scene detection target in the video frame sequence, the task weight in the scene detection task, and the functional criticality in the scene detection entity. Based on the importance of the target, a first evaluation factor for the target area of ​​interest is generated, and combined with a second evaluation factor obtained by mapping the change feature score, an importance score for the target area of ​​interest is generated. Based on the pre-stored rank mapping table, the rank intervals to which the importance scores of multiple target attention regions belong are determined, and the feature ranks of multiple target attention regions are determined. Set the frame regions in the target feature frame that are not in the region of interest to be non-in-interest regions; Based on the regional correlation between non-interested regions and target interested regions, an importance score for the non-interested regions is generated, and the feature level of the non-interested regions is determined according to the level mapping table. The feature level of the non-interested regions is lower than that of the target interested regions.

[0039] Preferably, the frequency of occurrence of the scene detection target in the video frame sequence is determined based on the number of times the target is detected in the video frame sequence and the duration between frames; the task weight in the scene detection task can be determined based on the business importance of the target in the detection task. For example, components that directly reflect the operating status of equipment or safety risks (such as signal lights, temperature indicators, switch position markers, etc.) can be assigned higher task weights, while auxiliary or decorative components can be assigned lower weights; the functional criticality of the scene detection entity is determined based on the system level and function of the equipment to which the target belongs. For example, key components in the main circuit or primary equipment section (such as busbar connection points, main transformer bushings, surge arresters, etc.) have high functional criticality, while devices in the monitoring or auxiliary control layer have relatively low functional criticality. Preferably, a first evaluation factor for the target area of ​​interest is generated based on the target importance, and a second evaluation factor obtained by mapping the change feature score is combined to generate an importance score for the target area of ​​interest. This includes: first, normalizing the target importance to eliminate differences in the dimensions of different indicators, thus obtaining the first evaluation factor for the target area of ​​interest; then, weighting and fusing the second evaluation factor obtained by mapping the change feature score with the first evaluation factor, and generating a comprehensive importance score by setting a fusion ratio or dynamic adjustment coefficient to simultaneously reflect the task importance and dynamic activity level of the target; wherein, the mapping function from the change feature score to the second evaluation factor is preferably a first linear mapping function. Preferably, an importance score for the non-interested region is generated based on the regional correlation between the non-interested region and the target interest region. This includes generating a regional correlation index based on features such as spatial distance, boundary overlap rate, texture or optical flow consistency between the non-interested region and adjacent target interest regions, and obtaining the corresponding importance score through a second linear mapping function based on the regional correlation index. For example, when the non-interested region and the high-level target interest region have a strong spatial coupling relationship, its importance score can be increased accordingly; otherwise, its score can be decreased to reflect the degree of indirect correlation in the scene. Preferably, the second linear mapping function is smaller than the first linear mapping function, so that the importance score of the non-interested region is smaller than the importance score of the interested region; Preferably, the grading mapping table includes at least feature grades and corresponding importance score intervals. For example, the highest grade A corresponds to the first importance score interval and is used to characterize core targets that have a key impact on the detection task; grade B corresponds to the second importance score interval and is used to characterize targets with high variation characteristics or strong task relevance; grade C corresponds to the third importance score interval and is used to characterize areas of general importance or indirect association; grade D corresponds to the fourth importance score interval and is used to characterize areas with no significant features or that are only used as background references. Grades A and B are preferably applicable to areas of interest, while grades C and D are preferably applicable to areas of non-interest. In practical applications, the number of grade divisions and the range of intervals can be dynamically adjusted according to the needs of different inspection tasks.

[0040] Specifically, in this embodiment, the importance of a target is generated by comprehensively considering the frequency of its occurrence, task weight, and functional criticality in the scene detection target. This importance score is then combined with the change feature score to form an importance score, which is mapped to a feature level. This allows the selection and classification of target areas of interest to have clear business semantics and risk orientation. Through this classification mechanism, key areas related to equipment status recognition, hidden danger warning, and anomaly diagnosis can be prioritized in video frames, thereby reducing the interference of irrelevant background on subsequent recognition and reasoning. At the same time, non-interested areas are assigned lower feature levels based on regional correlation, enabling video content to be processed differently according to its importance. This not only improves the accuracy of front-end selection but also provides a quantifiable priority basis for subsequent encoding scheduling, thereby significantly improving the overall task sensitivity and resource utilization efficiency of the system.

[0041] In one embodiment, such as Figure 3 As shown, based on the frame quality of the target feature frames, the variation characteristics of the target region of interest, and the feature level, multiple target feature frames are mapped into a one-dimensional pixel sequence through a pre-constructed dimensionality reduction mapping model, including: Step 2051: Based on the quality pre-verification results of the video frame sequence in which the target feature frame is located, and combined with the regional average change feature score of the target feature frame, determine the frame quality of the target feature frame, and obtain the model order based on the pre-verification quality. Step 2052: Based on the change characteristics of the target region of interest and the corresponding feature level, generate the region weights of multiple regions in the target feature frame and determine the sampling density of the model. Step 2053: Based on the dimensionality reduction mapping model determined by the model order and sampling density, obtain information on multiple pixels in multiple regions of the target feature frame, and generate a single-frame one-dimensional pixel sequence based on the information on multiple pixels. Step 2054: After obtaining pixel information from multiple target feature frames, the corresponding multiple single-frame one-dimensional pixel sequences are spliced ​​and fused to generate a multi-frame fused one-dimensional pixel sequence.

[0042] Specifically, in this embodiment, adaptive dimensionality reduction modeling for different video frame qualities and regional differences is achieved by comprehensively considering the frame quality of the target feature frame, the change characteristics of the target region of interest, and the feature level during the dimensionality reduction mapping process. By dynamically determining the model order based on the frame quality, high-quality frames can retain more spatial details under higher-order mapping, while low-quality frames can effectively suppress noise interference under lower-order mapping, thereby improving the temporal stability and inter-frame consistency of the dimensionality reduction mapping. At the same time, region weights are generated based on the change characteristics and feature levels of the target region of interest, and the sampling density is determined accordingly. This results in higher sampling accuracy in regions with significant changes or semantic importance, thus maintaining the detailed features of key regions after dimensionality reduction and improving the sensitivity and recognition accuracy of feature representation. Furthermore, the dimensionality reduction mapping model is constructed based on structure-preserving curves, ensuring that adjacent pixels in two-dimensional space retain local continuity in a one-dimensional sequence, thereby guaranteeing the spatial coherence and learnability of the reduced features. Building upon this, by splicing and fusing multiple frames of single-frame one-dimensional pixel sequences, the feature change trends between consecutive frames can be captured in the temporal dimension, reducing the impact of single-frame noise and further improving overall robustness and temporal expressiveness. In summary, this embodiment, through a space-time dual adaptive dimensionality reduction mapping mechanism, achieves efficient compression and feature extraction of video data while maintaining feature integrity, significantly improving data utilization and intelligent recognition performance during video transmission and storage.

[0043] In one specific embodiment, the frame quality of the target feature frame is determined based on the quality pre-verification result of the video frame sequence containing the target feature frame, combined with the regional average change feature score of the target feature frame. The model order is then obtained based on the pre-verification quality, expressed as: ; Where, N H N represents the model order. min With N max Let represent the minimum and maximum possible orders of the model, respectively, and let f() represent the frame quality normalization function, used to map the frame quality score to the range of 0-1. The calculation result represents the frame quality score, where S C τ represents the overall quality score of the video frame sequence containing the target feature frame, μ represents the weighting coefficient of the overall quality score of the video frame sequence containing the target feature frame, and τ represents the weighting coefficient of the region change feature score of the target feature frame. The average change feature score of the target feature frame is represented by the region.

[0044] In one specific embodiment, region weights for multiple regions in the target feature frame are generated based on the change characteristics of the target region of interest and the corresponding feature level. These regions include both target regions of interest and non-interest regions, as shown below: ; Among them, Are i This represents the region weight of the i-th region. Z represents the change feature score of the i-th region. When this region is a target area of ​​interest, this value is equal to the change feature score of the target area of ​​interest; when this region is not a target area of ​​interest, this value is set to 0. i This represents the feature level value of the i-th region. Z represents the maximum score of the change feature in the region of the target feature frame. max This represents the maximum value of the feature level, and γ represents the fusion weight coefficient, which takes a value between 0 and 1, preferably 0.4. The model sampling density is determined based on the region weights, and is expressed as follows: ; Where, ρ i ρ represents the sampling density of the i-th region in the dimensionality reduction mapping model. min and ρ max Let represent the minimum and maximum sampling densities of the dimensionality reduction mapping model, respectively, and g() represent the normalized mapping function, preferably a nonlinear growth function, to enhance the sampling sensitivity of high-weight regions.

[0045] In one specific embodiment, a dimensionality reduction mapping model is used to generate a mapping curve that traverses multiple pixels on a two-dimensional plane. This allows for obtaining pixel information through traversal sampling and constructing a corresponding one-dimensional pixel sequence, thereby achieving structure-preserving dimensionality reduction of two-dimensional plane video frame data. The dimensionality reduction mapping model uses a structure-preserving dimensionality reduction curve as its basic model and employs the frame quality of the target feature frame, the variation characteristics of the target region of interest, and the feature level as weighted input parameters to dynamically control the order of the curve and the local sampling density. This can be expressed as: ; Where M(x,y) represents the dimensionality reduction mapping model function that maps two-dimensional coordinate points (x,y) to one-dimensional pixel sequence positions. Indicates order N H The structure-preserving dimensionality reduction curve is preferably a Hilbert curve. This represents the sampling density adjustment function for different regions during the traversal mapping process. It is used to adaptively adjust the distribution of sampling points on the model curve. Preferably, it can be expressed as: ; in, and These represent the sampling offsets along the x-axis and y-axis, respectively, which are inversely proportional to the regional sampling density. By adjusting the offsets, non-uniform distribution control of sampling points can be achieved during the Hilbert curve traversal process, i.e., more sampling points are gathered in high-level, high-variable regions, and sparse sampling is carried out in low-level regions. Furthermore, to ensure the preservation of the spatial structure in high-weight regions, it is preferable to improve the turning angle sequence of the Hilbert curve, increasing the number of turning steps when entering high-weight regions to enhance the coverage density of local paths. This adjustment is expressed as follows: ; in, This represents the rotation angle of the standard Hilbert curve at the k-th turning point. Indicates the steering angle correction range. This represents the rotation angle of the curve corresponding to the dimensionality reduction mapping model at the k-th turning point. This represents a region-weighted adjustment function used to increase the degree of local steering subdivision in high-weight regions, thereby achieving refined coverage of local curves. It is preferably a composite mapping function based on spatial traversal and principal component projection.

[0046] In one embodiment, a pixel segment includes a target-of-interest pixel segment and a non-interest pixel segment. A one-dimensional pixel sequence is divided into multiple pixel segments based on region feature parameters, and these multiple pixel segments are aggregated across frames according to feature levels to obtain multiple feature sequences, including: The region feature parameters are analyzed to obtain the two-dimensional coordinate information of the target region of interest. The region feature parameters are used to describe the spatial distribution and visual importance of each region in the video frame, including but not limited to region boundaries, size and salience value. Based on the dimensionality reduction mapping model, the one-dimensional coordinate information of the target region of interest is obtained through the two-dimensional coordinate information, and the pixel position range of the target region of interest in the one-dimensional pixel sequence is determined according to the one-dimensional coordinate information. Based on the pixel location range, pixel segments corresponding to the target interest region are extracted from the one-dimensional pixel sequence and used as target interest pixel segments, and the feature level of the corresponding target interest region is labeled. In response to obtaining multiple target interest pixel segments, the remaining multiple pixel segments in the one-dimensional pixel sequence are identified as multiple non-interest pixel segments, and the feature level of the corresponding non-interest region is labeled. Based on the feature level of the target interest region, multiple target interest pixel segments are aggregated at the same level to obtain multiple target interest feature sequences; Based on the feature level of the non-interested region, multiple non-interested pixel segments are aggregated at the same or adjacent levels to obtain multiple non-interested feature sequences. Multiple target-focused feature sequences and multiple non-focused feature sequences are sorted according to their corresponding sequence feature levels to generate multiple ordered feature sequences. The sequence feature level is obtained by the average feature level of the multiple target-focused feature sequences or non-focused feature sequences that constitute it. It is worth noting that pixel segment aggregation can include the following two scenarios: First, aggregating pixel segments of the same region at different times, where the purpose of aggregation is to maintain the continuity and feature consistency of the region in the temporal dimension, so that key regions maintain a stable representation during cross-frame analysis and encoding; Second, aggregating pixel segments of different regions with the same feature level at the same time, where the purpose of aggregation is to uniformly process regions with similar importance or task value in the spatial dimension, thereby reducing data redundancy and improving the statistical stability and compressibility of the sequence.

[0047] Specifically, in this embodiment, the spatial coordinates of the target region of interest are obtained by parsing the regional feature parameters, and then precisely located to a one-dimensional pixel sequence using a dimensionality reduction mapping model. This divides the video frame into target and non-target pixel segments, and generates a feature sequence by cross-frame aggregation based on feature levels. This allows for content structuring before compression encoding, ensuring the temporal continuity of key regions and the integrity of feature expression, reducing spatial discrepancies and information fragmentation. Simultaneously, aggregation at the same or adjacent levels based on feature levels significantly reduces data redundancy, improves sequence compressibility and statistical stability, and facilitates efficiency optimization of the encoder in prediction, transformation, and bit allocation. Therefore, this embodiment achieves an ordered organization from spatial pixels to temporal features, laying an accurate and controllable feature foundation for subsequent hierarchical encoding and on-demand transmission.

[0048] In a preferred embodiment, after obtaining the one-dimensional pixel sequence and dividing it into multiple pixel segments, the method further includes: The quality of each pixel in each pixel segment is evaluated to obtain the corresponding pixel quality score. The pixel quality score is calculated based on one or more of the following indicators: pixel sharpness index, local gradient strength index, color information entropy index, and pixel stability index between adjacent frames. Based on the pixel quality score, a set of low-information pixels is determined. When the pixel quality score is lower than a preset threshold, the pixel is judged as a blurry point or noise point. Low-information pixels are removed and marker pixels are inserted at their positions. The pixel value of the marker pixel can be a preset neutral value (such as mean gray or neutral color) to maintain the consistency of pixel segment length and avoid displacement deviation in subsequent feature aggregation. Furthermore, after performing the label replacement, the pixel fragments are subjected to local smoothing and resampling to reduce gradient abrupt changes caused by the culling operation and improve the continuity and learnability of the pixel fragments.

[0049] Specifically, in this embodiment, the elimination and label replacement steps can filter out pixels with low feature contribution without affecting the continuity of the dimensionality reduction mapping, thereby significantly reducing noise interference and redundant information. Simultaneously, it maintains the structural integrity of pixel segments, ensuring the consistency of the aggregated feature sequence in both time and spatial projection. Therefore, this preferred embodiment not only further improves the signal-to-noise ratio of the one-dimensional pixel sequence but also achieves higher compression rates and stronger discriminative feature representation capabilities, providing a more reliable data foundation for subsequent feature aggregation, pattern recognition, and intelligent video analysis.

[0050] In one embodiment, based on feature levels and the current environmental state, the encoding method and transmission order of multiple feature sequences are determined, and the multiple feature sequences are layered, compressed, encoded, and encapsulated into a video aggregate data stream, which is then sent to the receiving end, including: Obtain the current environmental status parameters, which include at least one or more of the following: network bandwidth, device load, signal interference intensity, and scene operating conditions. Among them, network bandwidth is used to measure the available rate of the video transmission channel, device load is used to evaluate the processing capability of the server or encoding terminal, signal interference intensity is used to determine the stability of the transmission channel, and scene operating conditions are used to reflect the dynamic characteristics or environmental constraints of the acquisition site, preferably the current electromagnetic interference intensity. Based on the feature levels of multiple feature sequences, a basic coding parameter set corresponding to the feature level is configured for each feature sequence. The basic coding parameter set includes parameters such as quantization step size, prediction mode, and bit rate control to control the sequence compression quality and bit allocation. The quantization step size is used to adjust the pixel value representation precision to control compression loss. The prediction mode is used to determine the coding prediction strategy between sequences. The bit rate control is used to constrain the size of the encoded output bit stream to ensure that the coding quality and bandwidth are compatible. Based on the current environmental state parameters, one or more basic coding parameters in the basic coding parameter set are adjusted to obtain the modified coding parameter set. The adjustment of the modified coding parameter set can be dynamically modified according to the environmental parameters to adjust the quantization step size, prediction mode and bit rate control, thereby balancing video quality and transmission stability. Based on the modified coding parameter set configured with feature sequences, corresponding coding weight factors are generated to characterize the impact of modified coding parameters on coding performance. The coding weight factors can be determined by comprehensively considering the compression rate, reconstruction quality, and transmission reliability of the encoded video. Based on the encoding weight factor and the level weight factor obtained from the feature level mapping, transmission priority coefficients of multiple feature sequences are generated respectively. The transmission priority coefficient is obtained by weighted fusion of the level weight factor obtained from the feature level linear mapping and the encoding weight factor. Based on the transmission priority coefficient, the transmission order of multiple feature sequences is determined. The transmission order is linearly related to the magnitude of the transmission priority coefficient. The larger the transmission priority coefficient, the earlier the transmission order. In order to prioritize the transmission reliability and readability of the core area under the conditions of limited bandwidth or electromagnetic interference. Based on the corresponding set of modified coding parameters, multiple feature sequences are subjected to hierarchical compression coding and then encapsulated sequentially to generate a video aggregated data stream. The hierarchical compression coding achieves high-quality preservation of key areas and moderate compression of non-key areas by applying different compression intensities to different feature levels or regions. At the same time, the encapsulation ensures that the data stream can be decoded layer by layer and retrieved on demand.

[0051] Specifically, in this embodiment, basic coding parameters are configured according to the feature level of the feature sequence, and the coding strategy is dynamically adjusted in combination with environmental conditions such as network bandwidth, device load, and signal interference intensity. Furthermore, the transmission priority is determined based on the coding weight and level weight, realizing a hierarchical coding and scheduling mechanism that prioritizes key areas and adaptively degrades non-key areas. In scenarios with limited network or fluctuating links, the clarity and transmission reachability of key areas of the task are guaranteed first, avoiding the loss of core status information due to insufficient bitrate. At the same time, the layered encapsulated video aggregation data stream supports progressive decoding and hierarchical retrieval, which can significantly reduce bandwidth consumption and transmission latency, and improve the stability and real-time recognition capability of the system in complex network environments.

[0052] like Figure 4 As shown, this application also provides a lossless video compression transmission and storage method, applied to the receiving end 102, the method including: Step 301: Receive the video aggregation data stream transmitted by the sending end, parse it to obtain multiple encoded feature sequences and determine the corresponding feature levels; The video aggregation data stream includes multiple layered encoded feature sequences. Each feature sequence corresponds to a feature level divided by the sending end according to the target feature frame. The parsing process includes extracting the encoded sequence and its metadata from the encapsulated data so that it can be processed according to the level in subsequent processing. Step 302: Perform integrity verification on multiple encoded feature sequences. In response to the integrity verification passing, determine the decoding method of multiple encoded feature sequences based on feature level and perform layered decoding to obtain multiple decoded feature sequences. Among them, integrity verification is used to detect packet loss or erroneous bitstreams during transmission. The decoding method may include different bitrate decoding strategies or prediction modes depending on the feature level. Layered decoding can restore the pixel information of key areas and non-key areas layer by layer, ensuring that core information is restored first. Step 303: Store multiple decoded feature sequences into corresponding hierarchical storage areas according to feature levels, and determine the spatiotemporal correlation information of the sequences based on the regional feature parameters of the multiple decoded feature sequences, and construct a sequence storage correlation index; Among them, the hierarchical storage area is used to distinguish different feature levels or different time segments. The regional feature parameters include the coordinates of the target area of ​​interest and the markings of the non-interested areas. The spatiotemporal correlation information is used to record the position and spatial mapping relationship of each sequence on the video timeline. The index table can accelerate on-demand retrieval and splicing operations. Step 304: In response to receiving a video reading instruction, multiple decoded feature sequences are retrieved based on the sequence storage association index and concatenated to obtain a fused feature sequence; The splicing process reassembles the hierarchically stored feature sequences according to time order and spatial location based on spatiotemporal correlation information, ensuring that the target area of ​​interest is continuous in the time dimension and that the non-interested areas maintain a reasonable spatial mapping, thereby realizing the feature information recovery of the complete video frame. Step 305: Based on the preset dimensionality reduction mapping inverse model, multiple fused feature sequences are restored into multiple two-dimensional video frames and then frame interpolation and correction are performed to generate a video frame sequence. Among them, the dimensionality reduction mapping inverse model maps the one-dimensional pixel sequence back to the corresponding two-dimensional coordinates, the frame interpolation operation is used to restore the missing or replaced pixels between frames, and the correction is used to eliminate the geometric deviations introduced during compression coding or dimensionality reduction, and finally obtain a continuous and readable two-dimensional video frame sequence.

[0053] Specifically, in this embodiment, the receiving end performs layered parsing and integrity verification on the video aggregated data stream, and performs layered decoding and partitioned storage according to feature levels. At the same time, a spatiotemporal correlation index is constructed to support on-demand retrieval and splicing reconstruction. This ensures that key layer data is restored first, avoids the spread of decoding errors between layers, and improves the availability and reliability of key content. In addition, through inverse dimensionality reduction mapping and frame interpolation correction, the complete video frame can be restored with low computational cost while maintaining temporal continuity and spatial structure consistency. This layered storage and on-demand reconstruction mechanism not only improves the retrieval and maintainability of video management, but also allows for flexible selection of recovery levels according to different business needs, thereby effectively reducing computational and storage overhead and improving system performance and operating efficiency in multi-task concurrent scenarios.

[0054] It should be understood that, although Figures 2 to 4The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figures 2 to 4 At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0055] In one embodiment, such as Figure 5 As shown, a lossless video compression transmission and storage device is provided, including a transmitter and a receiver. The transmitter includes: a first extraction module for acquiring and parsing video data to obtain a video frame sequence composed of multiple video frames; a pre-verification module for performing quality pre-verification on the video frame sequence, and determining the video frame sequence as a first frame sequence and identifying the corresponding video frames as first sequence frames if the quality pre-verification passes; a second extraction module for determining the target interest region and corresponding region feature parameters in the multiple first sequence frames based on scene detection targets, and selecting multiple target feature frames from the multiple first sequence frames according to the changing characteristics of the target interest region; and a level evaluation module for evaluating the multiple target feature frames. The system performs importance assessment on the target region of interest identified in the feature frames to determine the corresponding feature level. A dimensionality reduction mapping module maps multiple target feature frames into a one-dimensional pixel sequence based on the frame quality, variation characteristics of the target region of interest, and feature level, using a pre-built dimensionality reduction mapping model. A feature sequence generation module divides the one-dimensional pixel sequence into multiple pixel segments based on region feature parameters and aggregates these segments across frames according to feature level to obtain multiple feature sequences. An encoding and transmission module determines the encoding method and transmission order of multiple feature sequences based on feature level and current environmental state, layers and compresses the multiple feature sequences, encapsulates them into a video aggregate data stream, and sends it to the receiving end. The receiving end includes: a parsing module, used to receive the video aggregated data stream transmitted by the sending end, parse it to obtain multiple encoded feature sequences and determine the corresponding feature levels; a verification and decoding module, used to perform integrity verification on the multiple encoded feature sequences, and in response to the integrity verification passing, determine the decoding method of the multiple encoded feature sequences based on the feature levels and perform layered decoding to obtain multiple decoded feature sequences; a layered storage module, used to store the multiple decoded feature sequences into corresponding layered storage areas according to the feature levels, and determine the sequence spatiotemporal correlation information based on the regional feature parameters of the multiple decoded feature sequences to construct a sequence storage correlation index; a response splicing module, used to retrieve the multiple decoded feature sequences based on the sequence storage correlation index and splice them in response to the received video reading command to obtain a fused feature sequence; and a video generation module, used to restore the multiple fused feature sequences into multiple two-dimensional video frames based on a preset dimensionality reduction mapping inverse model and perform frame interpolation and correction to generate a video frame sequence.

[0056] Specific limitations regarding lossless video compression transmission and storage devices can be found in the limitations of lossless video compression transmission and storage methods described above, and will not be repeated here. Each module in the aforementioned lossless video compression transmission and storage device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in hardware or independently of the processor in a computer device, or stored in software in the memory of a computer device, so that the processor can call and execute the corresponding operations of each module.

[0057] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0058] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0059] The above embodiments are merely illustrative of several implementation methods of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application.

Claims

1. A method for lossless compression, transmission and storage of video, characterized in that, Applied to a sending end, the method comprises: Obtaining and parsing video data to obtain a video frame sequence composed of multiple video frames; Performing quality pre-checking on the video frame sequence, and in response to the quality pre-checking passing, determining the video frame sequence as a first frame sequence and determining corresponding video frames as first sequence frames; Based on scene detection targets, determining target attention areas in multiple first sequence frames and corresponding area feature parameters, and selecting multiple target feature frames from multiple first sequence frames according to the change characteristics of the target attention areas; Performing importance evaluation on the target attention areas determined in multiple target feature frames to determine corresponding feature levels; Based on the frame quality of the target feature frames, the change characteristics of the target attention areas and the feature levels, mapping multiple target feature frames to one-dimensional pixel sequences through a pre-constructed dimension reduction mapping model; According to the area feature parameters, the one-dimensional pixel sequences are divided into multiple pixel segments, and multiple pixel segments are cross-frame aggregated according to the feature levels to obtain multiple feature sequences; Based on the feature levels and the current environment state, the encoding mode and transmission order of multiple feature sequences are determined, multiple feature sequences are compressed and encoded in layers, packaged into video aggregation data streams, and sent to a receiving end.

2. The method of claim 1, wherein, The quality pre-checking on the video frame sequence comprises: Based on the detection scene, determine the scene frame quality detection index, wherein the detection scene at least includes the power inspection scene, and the scene frame quality detection index at least includes: inter-frame brightness stability, frequency domain noise energy, frame time sequence continuity; According to the frame brightness statistical result of the video frame sequence, the inter-frame average brightness change rate is obtained, and the inter-frame brightness stability score of the video frame sequence is determined; According to the frequency spectrum energy distribution of the video frame sequence, the high-frequency noise component proportion is obtained, and the frequency domain noise energy score of the video frame sequence is determined; According to the time stamp sequence of the video frame sequence, the inter-frame time interval and the continuity of frame number are determined, and the frame time sequence continuity score of the video frame sequence is generated; According to the inter-frame brightness stability score, the frequency domain noise energy score and the frame time sequence continuity score, the scene frame quality score of the video frame sequence is determined; According to the basic noise characteristics, exposure setting and ambient light intensity of the video acquisition device, the ground state frame quality score of the video frame sequence is determined; According to the scene frame quality score and the ground state frame quality score, the comprehensive quality score of the video frame sequence is generated; In response to the comprehensive quality score being greater than or equal to the comprehensive quality score threshold, it is determined that the quality pre-checking passes; In response to the comprehensive quality score being less than the comprehensive quality score threshold, it is determined that the quality pre-checking does not pass.

3. The method of claim 1, wherein, The method further comprises: In response to the quality pre-checking not passing, at least based on standard anomaly detection and scene interference anomaly detection, multiple video frames in the video frame sequence are traversed to determine abnormal frames and corresponding abnormal types; According to the abnormal type, corresponding frame-level pre-repair processing is performed on the abnormal frame to obtain a repaired frame and update the video frame sequence; The updated video frame sequence is re-executed for the quality pre-checking; In response to the quality pre-checking failing for a continuous preset number of times, the video frame sequence is determined as an abnormal frame sequence, a spatiotemporal correlation sequence corresponding to the abnormal frame sequence is determined and sent to the receiving end, so that the receiving end performs frame-level reconstruction on the abnormal frame sequence according to the spatiotemporal correlation sequence to generate a reconstructed frame sequence.

4. The method of claim 1, wherein, The scene detection target is determined based on the target attention region in the first sequence frame, including: Based on the historical fault record of the detected scene and the running risk model, a plurality of scene risk components are determined; The plurality of scene risk components and a plurality of task critical targets pre-set based on the scene detection task are taken as the scene detection target; The determined scene detection target is extracted to generate a scene detection target feature vector; Based on the consistency of the intra-frame structure feature and the region feature, candidate region extraction is performed on the plurality of first sequence frames to obtain a plurality of candidate regions and generate corresponding candidate region vectors; According to the similarity comparison result of the candidate region vector and the scene detection target feature vector, the target attention region is selected from the plurality of candidate regions, and the region feature parameter of the target attention region is generated according to the selected candidate region vector.

5. The method of lossless compression, transmission and storage of video according to claim 1, wherein, The target feature frame is selected from the plurality of first sequence frames according to the change feature of the target attention region, including: Frame difference calculation is performed on the target attention region in the plurality of first sequence frames to obtain a change rate parameter representing the dynamic change degree of the target attention region; Based on a preset image quality index, an image quality parameter of the target attention region in the plurality of first sequence frames is determined; According to the change rate parameter and the image quality parameter, a change feature score is generated to represent the change feature of the target attention region in the first frame sequence; Based on the change feature score, the plurality of first sequence frames are sorted, and a plurality of frames with the highest change feature score are selected as the target feature frame.

6. The method of lossless compression, transmission and storage of video according to claim 5, wherein, The target attention region in the plurality of target feature frames is evaluated to determine the corresponding feature level, including: For a plurality of target attention regions, the target importance of the corresponding scene detection target is obtained, wherein the target importance is determined based on one or more of the following: the frequency of the scene detection target in the video frame sequence, the task weight in the scene detection task, and the functional criticality in the scene detection entity; According to the target importance, a first evaluation factor of the target attention region is generated, and a second evaluation factor mapped from the change feature score is combined to generate an importance score of the target attention region; Based on a pre-stored level mapping table, the importance score of each target attention region is determined to belong to a level interval, and the feature level of each target attention region is determined. The frame region of the target feature frame other than the target attention region is set as a non-attention region; According to the region relevance of the non-attention region and the target attention region, an importance score of the non-attention region is generated, and the feature level of the non-attention region is determined according to the level mapping table, wherein the feature level of the non-attention region is less than the feature level of the target attention region.

7. The method of lossless compression, transmission and storage of video according to claim 6, wherein, The pixel segments include target attention pixel segments and non-attention pixel segments, the one-dimensional pixel sequence is divided into multiple pixel segments according to the region feature parameters, and multiple feature sequences are obtained by aggregating multiple pixel segments across frames according to the feature levels, including: The region feature parameters are analyzed to obtain two-dimensional coordinate information of the target attention region; Based on the dimension reduction mapping model, one-dimensional coordinate information of the target attention region is obtained through the two-dimensional coordinate information, and the pixel position range of the target attention region in the one-dimensional pixel sequence is determined according to the one-dimensional coordinate information; According to the pixel position range, the pixel segment corresponding to the target attention region in the one-dimensional pixel sequence is intercepted as the target attention pixel segment, and the feature level of the corresponding target attention region is calibrated; In response to obtaining multiple target attention pixel segments, multiple non-attention pixel segments remaining in the one-dimensional pixel sequence are determined, and the feature levels of the corresponding non-attention regions are calibrated; Based on the feature level of the target attention region, multiple target attention pixel segments are aggregated at the same level to obtain multiple target attention feature sequences; Based on the feature level of the non-attention region, multiple non-attention pixel segments are aggregated at the same level or adjacent level to obtain multiple non-attention feature sequences; Multiple target attention feature sequences and multiple non-attention feature sequences are sorted according to the corresponding sequence feature levels to generate multiple ordered feature sequences.

8. The method of lossless compression, transmission and storage of video according to claim 1, wherein, Based on the feature level and the current environment state, the encoding mode and the transmission order of multiple feature sequences are determined, multiple feature sequences are compressed and encoded in layers, packaged as video aggregation data stream, and sent to the receiving end, including: Obtain the current environment state parameters, which include one or more of the following: network bandwidth, device load, signal interference intensity and scene running conditions; According to the feature level of multiple feature sequences, multiple feature sequences are respectively configured with a set of basic encoding parameters corresponding to the feature level; According to the current environment state parameters, one or more basic encoding parameters in the basic encoding parameter set are adjusted to obtain a modified encoding parameter set; Based on the modified encoding parameter set configured by the feature sequence, a corresponding encoding weight factor is generated to represent the influence of the modified encoding parameter on the encoding performance; According to the encoding weight factor and the level weight factor obtained by the feature level mapping, a transmission priority coefficient of each feature sequence is generated; Based on the transmission priority coefficient, the transmission order of multiple feature sequences is determined; According to the corresponding modified coding parameter set, the plurality of feature sequences are sequentially packaged after hierarchical compression coding to generate the video aggregation data stream.

9. A method for lossless compression, transmission and storage of video, characterized in that, Applied to a receiving end, the method comprises: Receiving the video aggregation data stream transmitted by the sending end, and parsing to obtain a plurality of coded feature sequences and determining corresponding feature levels; Performing integrity check on the plurality of coded feature sequences, and in response to passing the integrity check, determining a decoding mode of the plurality of coded feature sequences based on the feature levels and performing hierarchical decoding to obtain a plurality of decoded feature sequences; According to the feature levels, the plurality of decoded feature sequences are respectively stored in corresponding hierarchical storage areas, and sequence space correlation information is determined according to regional feature parameters of the plurality of decoded feature sequences to construct sequence storage correlation indexes; In response to receiving a video reading instruction, the plurality of decoded feature sequences are retrieved and spliced based on the sequence storage correlation indexes to obtain a fused feature sequence; Based on a preset dimension reduction mapping inverse model, the plurality of fused feature sequences are restored to a plurality of two-dimensional video frames and are subjected to frame filling and correction to generate a video frame sequence.

10. A device for lossless compression, transmission and storage of video, comprising a sending end and a receiving end, characterized in that, The sending end is configured to implement the method according to any one of claims 1-8.

Images