Method and system for verifying supplier capability based on real scene video and iot data, electronic device and storage medium

Abstract

Embodiments of the present application provide a supplier capability verification method and system based on real scene video and Internet of Things data, electronic equipment and storage medium, the method comprises: obtaining real scene video and Internet of Things data set of supplier site. Based on the preset event in the real scene video and the production process of the supplier, the summary video of the real scene video is generated, and the initial summary video is obtained. Based on the acquisition time of the Internet of Things data in the Internet of Things data set and the shooting time of the initial summary video, the Internet of Things data is added to the initial summary video, and the target summary video is obtained. Output the target summary video, so that the demander can verify the supplier based on the target summary video. In this way, the analysis cost of the demander can be reduced, and the verification efficiency can be improved. At the same time, the demander can more intuitively combine the video content and the information of the two modalities of the Internet of Things data to verify the supplier, further improving the verification efficiency.

Description

Technical Field

[0001] This invention relates to the field of video technology, and in particular to a method, system, electronic device, and storage medium for supplier capability verification based on real-scene video and IoT data. Background Technology

[0002] Currently, in modern industrial supply chain management, demanders need to verify the capabilities of suppliers before selecting them. Supplier selection is based on the verification results. To facilitate capability verification for demanders, on-site video recordings of suppliers are often collected.

[0003] In related technologies, one approach is to directly acquire live video footage from the supplier's site and provide it to the demander so that the demander can verify the supplier's capabilities by reviewing the original video content. However, this method incurs high analysis costs and low verification efficiency for the demander. Summary of the Invention

[0004] To overcome the problems existing in related technologies, this invention provides a supplier capability verification method, system, electronic device, and storage medium based on real-scene video and IoT data.

[0005] In a first aspect, the present invention provides a supplier capability verification method based on real-world video and IoT data, the method comprising:

[0006] Obtain real-time video and IoT datasets from the supplier's site;

[0007] Based on the real-scene video and preset events in the supplier's production process, a summary video is generated for the real-scene video to obtain an initial summary video.

[0008] Based on the acquisition time of the IoT data in the IoT dataset and the shooting time of the initial summary video, the IoT data is added to the initial summary video to obtain the target summary video;

[0009] The target summary video is output so that the demand side can verify the supplier based on the target summary video.

[0010] Secondly, the present invention provides a supplier capability verification system based on real-scene video and IoT data, the system comprising:

[0011] The acquisition module is used to acquire real-time video footage and IoT datasets from the supplier's site.

[0012] The first generation module is used to generate a summary video for the real-scene video based on the real-scene video and preset events in the supplier's production process, so as to obtain an initial summary video.

[0013] The first adding module is used to add the IoT data to the initial summary video based on the collection time of the IoT data in the IoT dataset and the shooting time of the initial summary video, so as to obtain the target summary video;

[0014] The first output module is used to output the target summary video so that the demand side can verify the supplier based on the target summary video.

[0015] Thirdly, the present invention provides an electronic device comprising: a processor, a memory, and a computer program stored in the memory and executable on the processor, characterized in that, when the processor executes the program, it implements the supplier capability verification method based on real-scene video and IoT data as described in any one of the first aspects above.

[0016] Fourthly, the present invention provides a readable storage medium that, when the instructions in the storage medium are executed by a processor of an electronic device, enables the electronic device to perform the steps in the supplier capability verification method based on real-world video and IoT data as described in any of the embodiments of the first aspect above.

[0017] In this embodiment of the invention, real-time video footage and IoT datasets from the supplier's site are acquired. Based on the real-time video and preset events in the supplier's production process, a summary video is generated from the real-time video to obtain an initial summary video. Based on the acquisition time of IoT data in the IoT dataset and the shooting time of the initial summary video, IoT data is added to the initial summary video to obtain a target summary video. The target summary video is output for the client to verify the supplier. In this way, generating a target summary video from the real-time video provides the client with a target summary video that is shorter, has higher information density, and lower redundancy. Therefore, it can reduce the client's analysis costs and improve verification efficiency. Simultaneously, by adding IoT data to the initial summary video according to the time dimension to obtain the target summary video, the client can simultaneously view the supplier's actual site video content and IoT data when viewing the target summary video. This allows the client to more intuitively combine the information from both video content and IoT data modalities to verify the supplier, further improving verification efficiency. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the steps of a supplier capability verification method based on real-scene video and IoT data provided in an embodiment of the present invention.

[0020] Figure 2 This is a schematic diagram of a video processing procedure provided in an embodiment of the present invention;

[0021] Figure 3 This is a schematic diagram of another video processing procedure provided in an embodiment of the present invention;

[0022] Figure 4 This is a schematic diagram of a processing step provided by an embodiment of the present invention;

[0023] Figure 5 This is a schematic diagram of the structure of a supplier capability verification system based on real-scene video and IoT data provided in an embodiment of the present invention;

[0024] Figure 6 This is a structural diagram of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] Figure 1 This is a flowchart illustrating the steps of a supplier capability verification method based on real-scene video and IoT data provided in an embodiment of the present invention. Figure 1 As shown, the method may include:

[0027] Step 101: Obtain on-site video and IoT dataset from the supplier.

[0028] In this context, "supplier" can also be referred to as "the party providing the service." The supplier capability verification method based on real-scene video and IoT data provided in this embodiment of the invention can be applied to suppliers in different fields, such as manufacturing, agriculture, and energy. The content supplied by the supplier can vary. For example, the supplier can supply actual goods, energy (e.g., computers, mobile phone accessories, vegetables, electricity), or service results. For instance, the supplier could be a pesticide residue testing agency, providing pesticide residue testing results to the demand side. "Supplier site" refers to the supplier's actual production site, such as the pesticide residue testing operation site, vegetable planting site, or mobile phone accessory manufacturing site. "Real-scene video" refers to video footage taken at the actual production site; it can also be called on-site video.

[0029] The IoT dataset includes at least one IoT data point, also known as Internet of Things (IoT) data. This IoT data is collected by IoT devices in the supplier's actual production site. For example, a single IoT data point may include one or more data such as temperature, humidity, current, voltage, and equipment start-up / shutdown status parameters. IoT devices, also known as IoT devices, collect real-time sensor data. These devices can collect and share on-site data. IoT devices include production equipment used in the actual production site. For example, a sprinkler system in a vegetable growing area can collect environmental parameters (humidity) through a connected humidity sensor, and temperature, current, and voltage through connected temperature, current, and voltage sensors, respectively. It also collects parameters characterizing the equipment's start-up / shutdown status based on sensors in the device actuators, reflecting the equipment's real-time operating status. Furthermore, it can receive IoT data actively uploaded by IoT devices in the supplier's field, and multiple received IoT data points form an IoT dataset. Alternatively, it can interface with the supplier's deployed industrial IoT platform to actively extract IoT datasets from the platform. The industrial IoT platform is used to store IoT data shared by IoT devices in the supplier's field.

[0030] The system can receive video files actively uploaded by the supplier's on-site video acquisition module as real-time video. Specifically, embodiments of the present invention can be applied to a proficiency testing system, which can be deployed on a computing device. This proficiency testing system may include a video acquisition module, which includes cameras (e.g., industrial cameras) deployed at the supplier's site. The cameras are used to capture the supplier's on-site production process in real time via industrial protocols, acquiring video streams at a preset resolution and frame rate (e.g., 1920×1080 resolution, 25fps frame rate). For example, monitoring cameras can be deployed in vegetable greenhouses or experimental vegetable growing areas to continuously acquire high-definition video streams around the clock. Monitoring cameras can also be deployed in pesticide residue testing laboratories to capture the entire pesticide residue testing process, obtaining high-definition video streams.

[0031] Furthermore, the high-definition video stream is segmented according to a preset segmentation time and uploaded to the video processing module of the proficiency verification system. The continuously uploaded video files constitute the live video of the supplier's site. The preset segmentation time can be set as needed, for example, it can be one day, one hour, or 10 minutes, etc. The acquired video stream can be in Transport Stream (TS) format and can be uploaded to the video processing module via HTTP Live Streaming (HLS) protocol. The video processing module can receive the data, buffer it, and then store the complete live video. The video processing module can be an edge node or a cloud device. After receiving the live video, it can perform format standardization processing to obtain the final HLS format live video. It should be noted that this proficiency verification system can connect to multiple suppliers, and the connected suppliers are the suppliers to be processed. These suppliers can be distributed in different regions and produce different product categories. Each supplier can be processed using the supplier proficiency verification method based on live video and IoT data provided in this embodiment of the invention, facilitating cross-regional procurement by the demand side.

[0032] Step 102: Based on the real-scene video and preset events in the supplier's production process, generate a summary video for the real-scene video to obtain an initial summary video.

[0033] The preset events are pre-set based on the supplier's production processes. Different suppliers may have different production processes, and correspondingly, different preset events. Preset events are used to characterize key operations and variable events in the supplier's production processes. For example, for a vegetable supplier, preset events may include sowing operations, mid-term management operations (e.g., sprinkler irrigation, fertilization), harvesting operations, quality inspection operations, germination events, seedling emergence events, elongation events, flowering events, color change events, maturity events, etc. For a pesticide residue testing supplier, preset events may include preparing pesticide solutions, removing containers, placing containers, calibrating reagents, spraying pesticides, reading instrument data, sealing, test strip reaction events, etc.

[0034] Furthermore, a summary video refers to a shorter and more concise video created by analyzing and processing the original video to extract representative segments (e.g., segments containing key information). This preserves as much of the original video's key information as possible, allowing users to quickly understand the video content. In this embodiment of the invention, segments in the live-action video that embody preset events are identified, and videos are generated based on these segments as initial summary videos. This allows the initial summary video to centrally showcase key events in the supplier's actual operational process, enabling the client to quickly browse key information segments in the live-action video. This facilitates the client's evaluation of the supplier's production process and verification of the supplier's actual processing capabilities based on the video content.

[0035] Step 103: Based on the acquisition time of the IoT data in the IoT dataset and the shooting time of the initial summary video, add the IoT data to the initial summary video to obtain the target summary video.

[0036] Specifically, the initial summary video after adding IoT data becomes the target summary video. By synchronizing the IoT data collection time with the initial summary video's recording time, the IoT data and video frames in the initial summary video can be matched synchronously along a timeline. Matching IoT data is then added to each video frame, effectively linking the two. This allows the summary video and IoT data to be displayed synchronously and in parallel, enabling the video content and IoT data to interact over time. Furthermore, this synchronized display allows the production process to be presented more transparently and visually to the client, and the IoT data to be presented more intuitively to the user, enhancing the client's trust in the target summary video. This, in turn, improves the transparency of supplier capability verification and the client's perceived credibility of the supplier's capabilities.

[0037] With the development of the Industrial Internet of Things (IIoT) and video surveillance technology, suppliers are using on-site video combined with IoT data to visualize and support production activities, thereby achieving real-time monitoring and historical review of supplier production sites. However, in related technologies, video data and IoT data are often stored independently, making it difficult to integrate and analyze them. For example, demanders can only view the supplier's process flow through video surveillance; for instance, taking a sensor supplier as an example, they can view the actual video of sensor calibration, aging testing, automated assembly, and other production processes. Alternatively, they can analyze IoT data separately. In this embodiment of the invention, video content and IoT data are linked in the time dimension, facilitating cross-integration analysis by demanders and cross-verifying the supplier's actual capabilities.

[0038] Meanwhile, although the source of IoT data is highly authentic, the collection and processing of IoT data is closed. If the demander can only see the data results (i.e., the IoT data itself) but cannot understand the background of data generation, it is often difficult to establish intuitive trust. Furthermore, the duration of live-action videos is long, the content is complex, and the information density is low with high redundancy, making it difficult for the demander to quickly observe content that reflects the supplier's technical capabilities based on the video content. Therefore, the analysis cost is high. In this embodiment of the invention, the target summary video is shorter. Therefore, the information density is higher and the redundancy is lower, allowing the demander to conveniently and quickly view content that reflects the supplier's technical capabilities based on the target summary video, thereby reducing the analysis cost. Moreover, since the target summary video can simultaneously display video content and IoT data, the demander can clearly understand the background of IoT data generation, facilitating mutual corroboration between video content and IoT data, assisting the demander in matching and verifying, and enabling the demander to directly establish trust in the IoT data and video content, thereby improving the credibility of capability verification. For example, if two consecutive video frames in the target summary video show the vegetables before and at the start of sprinkler irrigation, and the humidity data in the IoT data added to these two video frames shows changes, then it can be confirmed that sprinkler irrigation actually occurred during the actual production process. As another example, if a video frame shows the device is running, and the temperature, current, and voltage data added to that video frame all fall within the device's power-on temperature, current, and voltage ranges, and the device's start / stop status parameter is "started," then it can be confirmed that the device is actually operating during the actual production process.

[0039] Step 104: Output the target summary video so that the demand side can verify the supplier based on the target summary video.

[0040] In this invention, the demand side, also known as the purchaser, uses the target summary video to verify the supplier's capabilities (i.e., capacity verification). The target summary video is generated and output to the demand side, allowing them to evaluate the supplier's entire process and multi-dimensional capabilities based on the video content. This assessment evaluates the supplier's actual processing capabilities and thus verifies their capabilities. For example, by viewing the target summary video, the demand side can determine whether the supplier's actual production process includes preset events in the production steps. If so, it indicates that the supplier's production process capabilities and technological processes comply with specifications. Furthermore, by viewing the generated products recorded in the target summary video, the demand side can determine the supplier's quality control level, and so on. By viewing the equipment operation status recorded in the target summary video, the demand side can determine the supplier's equipment maintenance capabilities.

[0041] In summary, the supplier capability verification method based on real-scene video and IoT data provided in this embodiment of the invention acquires real-scene video and IoT datasets of the supplier's site. Based on the real-scene video and preset events in the supplier's production process, a summary video is generated from the real-scene video to obtain an initial summary video. Based on the acquisition time of IoT data in the IoT dataset and the shooting time of the initial summary video, IoT data is added to the initial summary video to obtain a target summary video. The target summary video is output for the demand side to verify the supplier based on the target summary video. In this way, the target summary video is generated based on the real-scene video and provided to the demand side. Since the target summary video is shorter, has higher information density, and lower redundancy, it can reduce the analysis cost for the demand side and improve verification efficiency. At the same time, by adding IoT data to the initial summary video according to the time dimension to obtain the target summary video, the demand side can simultaneously view the video content and IoT data of the supplier's actual site when viewing the target summary video. This allows the demand side to more intuitively combine the information of these two modalities, video content and IoT data, to verify the supplier, further improving verification efficiency.

[0042] Furthermore, in real-world industrial scenarios, the results of supplier capability verification directly impact product delivery quality, project schedule, and service assurance capabilities. Traditionally, capability verification involves manually reviewing static supplier documentation (e.g., qualification documents, meeting minutes, historical transaction records, laboratory test reports, etc.) or conducting on-site inspections. However, with the increasing complexity of industrial environments and the faster pace of equipment upgrades, this approach has significant limitations. Specifically, the static documentation relied upon by traditional methods cannot accurately reflect the actual production environment, resulting in low accuracy and timeliness of information, and static documentation is easily forged. Therefore, its credibility is also significantly limited. Manual on-site verification is costly, inefficient, and cannot be scaled up.

[0043] In this embodiment of the invention, a target summary video with added IoT data is generated for the supplier based on real-scene video. This target summary video is then provided to the demander, eliminating the need for manual on-site verification. Therefore, verification costs are reduced and verification efficiency is improved. Simultaneously, target summary videos of suppliers can be provided to multiple demanders, or multiple supplier target summary videos can be provided to a demander, enabling large-scale verification and meeting the diverse procurement needs of large demanders (e.g., multi-batch, multi-category, and cross-regional procurement needs). Furthermore, since real-scene video accurately reflects the actual production site, the accuracy and timeliness of the information in the final target summary video can be ensured to a certain extent. Moreover, because IoT data is synchronously added to the target summary video, the IoT data and video content can be simultaneously displayed to the demander, providing an evidence chain model combining the video content of the real-scene video and the IoT data for the demander's review. The IoT data and video content corroborate each other, resulting in higher credibility.

[0044] Optionally, step 103 above may specifically include:

[0045] Step 1031: For any IoT data in the IoT dataset, find the video frame whose shooting timestamp matches the collection timestamp of the IoT data from the initial summary video, and use it as the target video frame corresponding to the IoT data.

[0046] Step 1032: Overlay the IoT data onto the target video frame corresponding to the IoT data.

[0047] In this embodiment of the invention, the IoT dataset can be composed of IoT data and is synchronized with time. Specifically, each piece of IoT data is bound to a timestamp, which is the collection timestamp of that IoT data and is used to uniquely indicate the collection time of that IoT data. In specific implementation, the IoT devices at the supplier's site periodically collect IoT data according to a preset collection cycle and bind a collection timestamp to each piece of IoT data collected. The preset collection cycle can be set in advance as needed; for example, the preset collection cycle can be no more than 10 seconds.

[0048] The capture timestamp is used to indicate the capture time of video frames in the initial summary video. In a specific embodiment, the start capture time of the live-action video can be determined first. For each video frame in the initial summary video, the sum of the start capture time and the playback time of the video frame on the live-action video playback axis is calculated as the capture timestamp of the video frame. For example, assuming the start capture time is 08:00:00 on April 21, 2025, and the playback time of the video frame on the live-action video playback axis is 15 seconds, then the capture timestamp of the video frame can be determined as 08:00:15 on April 21, 2025.

[0049] Alternatively, the video acquisition module can obtain the actual time when each video frame is captured and add a capture timestamp to each video frame. Correspondingly, the capture timestamp carried by the video frame can be obtained. Furthermore, alignment can be performed according to a time synchronization mechanism. Specifically, for any video frame in the initial summary video, the acquisition timestamp of the IoT data is compared with the capture timestamp of the video frame. If they match, the video frame is identified as the target video frame corresponding to the IoT data. For IoT data that corresponds to the target video frame, the IoT data is overlaid onto the target video frame, so that the IoT data is displayed in the target video frame, achieving video frame-level alignment according to the timeline. For example, the IoT data can be used as the text content of a layer, and this layer can be overlaid on the target video frame at the top. The transparency of this layer can be less than a preset transparency threshold to avoid obscuring the original content in the target video frame. The preset transparency threshold can be set as needed, for example, 50%, 80%, etc., and this embodiment of the invention does not limit this. Understandably, environmental monitoring data, such as temperature and humidity curves, can also be generated based on IoT data and overlaid on the initial summary video to facilitate viewing by the user.

[0050] It should be noted that the execution order of acquiring the on-site real-time video of the supplier and acquiring the on-site IoT dataset in this embodiment of the invention is not unique. The operations of acquiring the on-site real-time video and acquiring the on-site IoT dataset can be executed in parallel or sequentially. For example, after generating the initial summary video, the playback timestamp of the first video frame in the initial summary video can be used as the start time, and the playback timestamp of the last video frame in the initial summary video can be used as the end time. The IoT data collected between the start time and the end time can then be used to form the IoT dataset. In this way, by synchronously collecting IoT data within the corresponding time period according to the start and end times using the timestamps of the initial summary video, unnecessary IoT data acquisition can be avoided.

[0051] Currently, traditional methods relying on static documents and manual static review, or simply providing live video footage, are insufficient to accurately represent a supplier's actual capabilities, lacking both sensory and data-driven support. In this invention, an initial summary video of the live video footage is synchronously overlaid with IoT data according to timestamps, creating a one-to-one correspondence between the operational behaviors presented in the video frames and the IoT data. Thus, the client's credibility with the video content depends not only on the image content but also on the corresponding IoT data support. When the client views the target summary video, the synchronously displayed video content and IoT data provide both sensory and data-driven support, allowing the client to more intuitively perceive the supplier's actual capabilities. Simultaneously, it facilitates subsequent alignment and analysis of IoT data and video frames along the timeline, combining video content and IoT data for dual cross-validation, enhancing the client's trust in the production process, and significantly improving the credibility of supplier capability verification.

[0052] Optionally, embodiments of the present invention may further include the following steps:

[0053] Step S21: Identify the device identifier of the production equipment captured in the real-scene video as the first identifier, and obtain the device identifier carried in the IoT data set as the second identifier.

[0054] In this embodiment of the invention, each IoT data entry also includes the identifier of the device that collected the IoT data. Accordingly, the identifier carried in each IoT data entry is extracted, and then deduplication is performed on the obtained identifier to obtain a second identifier. Image recognition is performed on video frames in the real-world video to extract video frames containing production equipment. Then, optical character recognition is performed on the video frames containing production equipment to extract the device identifier of the production equipment in the video frames. Finally, deduplication is performed on the obtained device identifier to obtain a first identifier.

[0055] Step S21: If the first identifier matches the second identifier, add a first trusted tag to the target summary video.

[0056] For any given first identifier, it can be compared with each second identifier. If a second identifier that is identical to the first identifier exists, then the first identifier is considered to match the second identifier. Accordingly, if all first identifiers match the second identifiers, the video content can be considered to match the IoT data, and a first trusted tag can be added to the target summary video to facilitate subsequent verification by the requesting party.

[0057] Step S23: If the first identifier and the second identifier do not match, add a first suspicious tag to the target summary video.

[0058] If at least one first identifier and a second identifier do not match, the video content can be considered mismatched with IoT data, and a first suspicious tag can be added to the target summary video. The first trusted tag indicates that the video content matches IoT data, while the first suspicious tag indicates that the video content does not match IoT data. Both the first trusted tag and the first suspicious tag can be text-based prompts. For example, the first trusted tag can be the text: "Video content matched with IoT data identified: No device identifier in the video does not match the IoT data!" This allows the client to intuitively perceive the authenticity of the video content, further enhancing their trust in the supplier's demonstrated capabilities. The first suspicious tag can be the text: "Video content mismatched with IoT data identified: A device identifier in the video does not match the IoT data; please be careful during verification!" Specifically, when adding the first trusted tag or the first suspicious tag, it can be used as the text content of a layer, added to the top layer of the first video frame of the target summary video. The transparency of this layer can be less than a preset transparency threshold to avoid obscuring the original content in the target video frame. Alternatively, the first trusted tag or the first suspicious tag can be associated with the target summary video as a text file. During output, this text file and the target summary video can be output simultaneously.

[0059] In this embodiment of the invention, the device identifier of the production equipment captured in the real-world video is identified as a first identifier, and the device identifier carried in the IoT dataset is obtained as a second identifier. If the first identifier and the second identifier match, a first trusted label is added to the target summary video. If the first identifier and the second identifier do not match, a first suspicious label is added to the target summary video. In this way, by adding a first trusted label or a first suspicious label, more dimensions of reference information are provided to the requester, thereby improving the verification efficiency of the requester.

[0060] Optionally, embodiments of the present invention may further include the following steps:

[0061] Step S31: Select video frames from the initial summary video as video frames to be verified, and perform authenticity detection on each video frame to be verified to obtain the frame detection result of each video frame to be verified; the frame detection result is used to characterize whether the video frame to be verified has any forgery traces.

[0062] Step S32: Based on the frame detection results of each of the video frames to be verified, generate an authenticity detection result; the authenticity detection result is a second trusted label or a second suspicious label.

[0063] Step S33: Output the authenticity detection result to the requester.

[0064] In this embodiment of the invention, periodic frame sampling and random frame sampling strategies can be used to identify potential forgery traces. Specifically, video frames can be periodically selected from the initial summary video as video frames to be verified at preset intervals, for example, selecting one video frame as the video frame to be verified every 10 video frames. And / or, video frames can be randomly selected as video frames to be verified; this embodiment of the invention does not limit this. Considering users' concern for the authenticity of the video, the authenticity of the initial summary video is verified by performing authenticity verification (referred to as verification) on these video frames to be verified. Forgery traces can include: editing, splicing, inconsistent lighting, static background deception (e.g., using a background that causes visual illusion), artificial intelligence (AI) synthesis, etc.

[0065] Specifically, the video frame to be verified can be used as input to a pre-trained video verification model. This model extracts vector features representing forgery traces, such as abnormal textures, time jumps, artifacts, editing, and splicing, from the video frame and performs anomaly detection. For example, an image forensics algorithm can be used to extract vector features representing forgery traces and perform anomaly detection. Image forensics algorithms are techniques used to detect and verify the authenticity of images, primarily for identifying whether images have been tampered with. Then, based on these vector features representing forgery traces, anomaly scores are generated for the video frame to be verified. For example, the larger the vector features representing forgery traces, the higher the anomaly score. Correspondingly, if the anomaly score is greater than a preset anomaly score threshold, a frame detection result indicating the presence of forgery traces (e.g., "0") is generated for the video frame to be verified. Otherwise, a frame detection result indicating the absence of forgery traces (e.g., "1") is generated for the video frame to be verified. This video verification model is a deep learning model and can also be called a video tampering detection model.

[0066] Furthermore, if the frame detection results of all video frames to be verified indicate that there are no signs of forgery, a second trustworthy label is output as the authenticity detection result. Otherwise, a second suspicious label is output as the authenticity detection result. This provides the requester with more comprehensive reference information, thereby improving the efficiency of verification. The second trustworthy label indicates that no forgery has been detected in the video content, while the second suspicious label indicates that forgery has been detected. Both the second trustworthy and second suspicious labels are text-based prompts. For example, the second trustworthy label could be the text: "The video content shows no signs of forgery, trustworthy!" and the second suspicious label could be the text: "The video content shows signs of forgery; please be careful during verification!" This provides a risk warning to the requester, facilitating verification. Simultaneously, explaining the specific reasons for trustworthiness / suspicion in the authenticity detection result helps the requester independently determine whether the video was genuinely filmed.

[0067] Specifically, outputting the authenticity detection result to the demand side may include simultaneously outputting the authenticity detection result when outputting the supplier's target summary video to the demand side. Of course, the target summary video can also be output first, followed by the authenticity detection result; this embodiment of the invention does not limit this.

[0068] Because videos can be forged through editing and splicing, even if the client sees the on-site video provided by the supplier, it is difficult to determine whether it was filmed on-site, whether it depicts the actual production process, or whether it has been tampered with. This can lead to distrust at the sensory level, affecting the effectiveness of the entire proficiency verification mechanism. Since video demonstration verification is a key basis for supplier selection, if falsification leads the client to choose an unqualified supplier, it will directly result in a flawed procurement decision and potential losses for the client. In this embodiment of the invention, frame-level authenticity analysis is performed on the initial summary video, and video frames to be verified are selected for authenticity testing. Based on the frame detection results of each video frame to be verified, authenticity detection results are generated for the target summary video. Finally, by outputting the authenticity detection results to the client, the client can intuitively perceive the authenticity of the video content, which can enhance the client's sensory trust to a certain extent, further strengthening the client's perceived trust in the supplier's capability demonstration content, and providing technical support for highly reliable and intelligent procurement decisions.

[0069] It should be noted that, in this embodiment of the invention, the generated authenticity detection result and the first trust label or the first suspicious label can also be written into the same document as a risk warning report and output to the requester. The risk warning report can provide the requester with effective annotations for the target summary video, facilitating the requester's judgment. Furthermore, in this embodiment of the invention, if the authenticity detection result is a second suspicious label, the target summary video of that supplier and the authenticity detection result can be not output. In this way, suppliers with suspicious authenticity detection results can be directly eliminated for the requester, ensuring the authenticity and reliability of the target summary video provided to the requester, improving the credibility of the provided target summary video, and avoiding the requester's distrust of the video used to demonstrate the supplier's capabilities.

[0070] In this embodiment of the invention, before generating a summary video for the real-scene video and identifying the first identifier from the real-scene video, the following steps can be performed: identifying invalid video frames in the real-scene video and removing them from the real-scene video. Invalid video frames can include duplicate video frames and video frames taken in dark conditions. For example, video frames with image brightness below a preset brightness threshold can be detected as video frames taken in dark conditions. Alternatively, an infrared filtering mechanism can be used to automatically identify and remove video data from dark time periods. Video frames with a similarity greater than a preset similarity threshold are identified as duplicate video frames. Because there is insufficient light or complete darkness in nighttime environments, video frames taken at night often lack effective visual information. By removing video frames taken in dark conditions, the dark parts of the real-scene video in nighttime environments can be removed. The content of duplicate video frames is often static, causing content redundancy; therefore, duplicate video frames can also be removed. In this embodiment of the invention, by removing invalid video frames, interference from invalid frames can be reduced, the workload of subsequent processing can be reduced, and the efficiency of subsequent processing can be improved.

[0071] Optionally, step 102 above may specifically include:

[0072] Step 1021: According to the preset events in the supplier's production process, divide the real-scene video into multiple frame segments and obtain the segment features of each frame segment.

[0073] In this embodiment of the invention, steps 1021 to 1023 can be implemented through a video summarization and compression module. This module can support compressing long-duration videos and generating summaries, and outputting short-duration summary videos. Specifically, event detection can be performed on the real-scene video to extract frame segments containing preset events. A frame segment includes multiple video frames, and each frame segment corresponds to a preset event. For example, the real-scene video can be used as input to a pre-trained video analysis algorithm for the supplier. This algorithm extracts frame features from each video frame in the real-scene video. The pre-trained video analysis algorithm can be a video analysis model based on a Spatio-Temporal Convolutional Neural Network (ST-CNN). This model is trained and fine-tuned using data from the supplier's production scenario to ensure that the model is adapted to the supplier's production scenario. When the model processes data, it can perform deep feature extraction, capturing the spatial features of static frames and the dynamic features between consecutive frames. Specifically, the frame features of video frames in the sequence of video frames of the real-world video are extracted one by one through convolutional layers and pooling layers. The frame features of the video frames can include spatial features and dynamic features.

[0074] Spatial features can include vectors representing the items used in the supplier's production process, while dynamic features can include vectors representing the operations performed and changing events during the supplier's production process—that is, vectors representing pre-defined events in the supplier's production process. For example, in a pesticide residue testing scenario, spatial features can include vectors representing items such as testing tools, containers, and reagents, while dynamic features can include vectors representing events such as pesticide application, container removal, placement, reagent calibration, pesticide spraying, instrument data reading, sealing, and test strip reaction. In a vegetable supply scenario, spatial features can include vectors representing items such as seeders, sprinkler irrigation equipment, and harvesting equipment, while dynamic features can include vectors representing events such as sowing, sprinkler irrigation, fertilization, harvesting, quality inspection, germination, emergence, elongation, flowering, color change, and maturity.

[0075] The semantic understanding layer performs semantic recognition on the real-scene video based on the frame features of each video frame, selecting a segment that semantically represents a preset event as a frame segment. This ensures that the content of the final summary video covers the preset event, showcasing key scenes of the complete operational process and thus demonstrating the supplier's key capabilities. For example, the semantic understanding layer can use inter-frame motion estimation algorithms (e.g., optical flow or background modeling) to determine the degree of dynamic change in consecutive video frames. For consecutive video frames with a dynamic change degree greater than a preset threshold, the layer identifies whether the consecutive video frames represent a preset event based on their frame features. For instance, for a vegetable supplier, preset events such as sprouting, elongation, flowering, and color change are usually accompanied by significant morphological and color changes. The corresponding video frames tend to have a greater degree of dynamic change. Therefore, frame segments can be divided based on motion states to generate the summary video. Consequently, frame segments composed of consecutive video frames representing preset events have a higher correlation with the preset events. Therefore, these frame segments will be assigned higher importance scores, thus giving them a higher selection priority. The initial summary video can also be called a video summary. Figure 2 This is a schematic diagram of a video processing process provided by an embodiment of the present invention, such as... Figure 2 As shown, for real-world videos, dark areas can be identified and discarded (i.e., invalid video frames can be removed), and then selection can be made based on motion state to finally obtain a video summary.

[0076] Then, the remaining video frames (i.e., the unselected video frames) are divided into multiple frame segments according to a preset unit time. The preset unit time can be pre-set according to the duration of the supplier's live video; the longer the live video, the longer the preset unit time can be. For example, for a vegetable supplier, the live video may include full-process monitoring video of vegetables from sowing to harvest, which is often very long, for example, up to 60 days. Therefore, the preset unit time can be set to one day. For a pesticide residue testing supplier, the live video is often in the minute range; for example, the live video may be a pesticide residue testing video approximately 20 minutes long. Therefore, the preset unit time can be set to 5 minutes, or 10 minutes, etc., and this embodiment of the invention does not limit this.

[0077] For any given frame segment, keyframes are extracted. For example, keyframes are automatically identified using an automatic keyframe recognition algorithm. For instance, the automatic keyframe recognition algorithm filters video frames containing key information (e.g., items used in the production process, operations performed, and existing changes) based on the frame features of the video frames. For example, the automatic keyframe recognition algorithm can be based on inter-frame difference algorithms, K-means clustering algorithms, etc., and this embodiment of the invention does not limit this. The frame features of the video frames in the frame segment are processed, for example, by performing average pooling on the frame features of all video frames to obtain the global features of the frame segment. Since frame features include dynamic features, the global features can characterize the change features of the frame segment. The extracted keyframes and global features are used as the segment features of the frame segment.

[0078] Step 1022: Based on the segment features of each frame segment, determine the importance score of each frame segment; the importance score is positively correlated with the correlation between the frame segment and the preset event.

[0079] In this embodiment of the invention, the segment features of each frame are used as input to the attention layer of the video analysis model to perform attention fusion between different frames. Attention fusion is a technique that combines attention mechanisms and feature fusion. By simulating the characteristics of human visual attention, it focuses on the most important parts of the input data and effectively integrates different feature information. The video analysis model is trained to learn which segment features are more relevant to a preset event and assigns weights based on this relevance. Relevance represents the similarity between segment features and the preset event; the higher the semantic similarity, the higher the relevance. Specifically, the attention layer assigns attention weights to frame segments based on their segment features, using an attention mechanism. For example, an attention mechanism can be used on the temporal dimension to identify the relevance of segment features to a preset event and assign higher attention weights to frame segments with higher relevance. The higher the relevance of a frame segment to the preset event, the greater its attention weight. These attention weights can be input into a preset classifier, which multiplies the attention weight of each frame segment by the total score (e.g., 100 points) and outputs this as the importance score for each frame segment. Thus, the higher the importance score, the more critical the frame segment is to the live video, and the better it reflects the pre-set events in the production process.

[0080] Step 1023: Select target frame segments according to the preset summary video duration, the importance score of each frame segment, and the segment duration, and compose the initial summary video based on the target frame segments.

[0081] The preset summary video length can be set based on the length of the live video; the longer the live video, the longer the summary video can be. For example, for pesticide residue testing suppliers, the summary video length can be 15-20 seconds. For vegetable suppliers, the summary video length can be 60 seconds.

[0082] In this embodiment of the invention, the live video is divided into multiple frame segments according to preset events in the supplier's production process, and the segment features of each frame segment are obtained. The importance score of each frame segment is determined based on its segment features. Target frame segments are selected to form an initial summary video according to a preset summary video length, the importance score of each frame segment, and the segment length. In this way, generating a summary video can effectively shorten the video length and highlight the preset events in the production process, thereby reducing the viewing cost for the client and allowing them to obtain comprehensive and accurate information in a very short time. This reduces labor costs and improves the efficiency of the client's assessment of the supplier's capabilities.

[0083] For example, 60 days of continuous monitoring video in a crop planting scenario can be compressed into a summary video of less than one minute to showcase the key stages and changes in the crop growth cycle. Similarly, live video footage from a pesticide residue testing scenario can be compressed into a 15-20 second summary video to retain key information, highlight pre-set events in the production process, and emphasize key operations and equipment operation, thereby improving understanding of critical processes and allowing the client to quickly assess whether the supplier possesses standardized processing capabilities. In this embodiment of the invention, the proficiency testing system can be linked with the supplier's monitoring platform for online compression.

[0084] Optionally, the step of selecting the target frame segment according to the preset summary video duration, the importance score of each frame segment, and the segment duration may include:

[0085] Step 1023a: Compress each frame segment and determine the segment length of each compressed frame segment.

[0086] In this step, each frame segment can be compressed, and the actual duration of the compressed frame segment is calculated as the segment duration. For example, an inter-frame prediction algorithm can be used to encode the pixel information of the previous or next frame to achieve compression. Motion information between adjacent frames can be used to reduce redundancy in frame segments by representing and encoding motion information through motion vectors, thereby achieving compression. Methods such as signal decomposition-based compression are also employed, and this embodiment of the invention does not limit the scope of the method. By compressing frame segments, the duration and size of the frame segments are shortened.

[0087] Step 1023b: Using the duration of the summary video as a constraint, select target frame segments based on the importance scores and segment durations of each compressed frame segment; the total duration of the target frame segments shall not exceed the duration of the summary video.

[0088] Specifically, each compressed frame segment is associated with an importance score and segment duration. The set of segment durations and importance scores associated with all compressed frame segments constitutes a duration-importance score sequence. The preset summary video duration and duration-importance score sequence are used as input to the knapsack algorithm. The knapsack algorithm selects target frame segments within the limited summary video duration, thus obtaining the initial summary video composed of target frame segments. For example, the knapsack algorithm can be used to model the frame segment selection problem as a "0-1 knapsack problem" within the limited summary video duration. By solving for the segment combination corresponding to the maximum total score (i.e., the most representative segment combination), the target frame segments are selected. The frame segments included in the final selected segment combination are the target frame segments.

[0089] Of course, other selection algorithms can also be used. For example, a greedy algorithm can be used to calculate the ratio of the importance score to the segment length for each frame segment. The frame segments are then sorted in descending order of the ratio. The segments with the highest ratios are selected sequentially until the total duration reaches the preset summary video length, or until there are no more frame segments to choose from. Figure 3 This is a schematic diagram of another video processing procedure provided by an embodiment of the present invention, such as... Figure 3 As shown, for real-world videos, depth feature extraction can be performed first to obtain the frame features of each video frame. Figure 3 The features of each video frame (represented by features from frame 0 to frame n) are used to perform attention fusion between different frames, resulting in attention weights for each frame segment. Next, a classifier is used to determine the importance score for each frame segment according to its attention weight. Finally, a knapsack algorithm is used to select the appropriate frame segment.

[0090] In this embodiment of the invention, each frame segment is compressed, and the duration of each compressed frame segment is determined. Target frame segments are selected based on the summary video duration, the importance score of each compressed frame segment, and the segment duration. This compression improves subsequent processing efficiency and reduces storage bandwidth pressure during video output.

[0091] Optionally, before the step of outputting the target summary video described above, the following may also be included:

[0092] Step S41: Generate narration text for the initial summary video based on the video content of the initial summary video.

[0093] In this embodiment of the invention, the narration text can also be referred to as the voice-over text. Specifically, content description text can be generated based on the video content of each video frame in the initial summary video. The narration text is then generated based on the content description text of each video frame. For example, content description text describing the production process and equipment status can be extracted according to a pre-set production process description and used as the narration text. The production process description is pre-set information; for example, the production process description may include text information such as equipment purpose, operating steps, safety risks, and safety tips. The content description text can be matched with this text information for similarity. If the similarity is greater than a preset similarity threshold, the content description text is extracted. In this way, the final voice-over text is used to describe text information such as equipment purpose, operating steps, safety risks, and safety tips.

[0094] Step S42: Based on the narration text, generate narration audio and narration subtitles corresponding to the narration text for the initial summary video.

[0095] The narration audio, also known as voice-over audio, can be input into a preset speech synthesizer. The synthesizer converts the narration text into natural speech, resulting in narration audio. Accordingly, the narration audio can introduce users to the device's purpose, operating procedures, safety risks, safety tips, etc., achieving a natural integration of video and audio content. This helps users quickly grasp the capabilities demonstrated in the video. For example, for the narration text "Day 15: Vegetables Sprouting," corresponding audio and subtitles can be generated, thus more intuitively demonstrating the vegetable growth cycle and quality.

[0096] Furthermore, the narration text can be translated into multiple preset languages, making it convenient for users with different language preferences to view. For content description text included in the narration text, the timestamp of the corresponding video frame is used as the text timestamp, and a text timestamp is added to the content description text. Then, the narration text is converted into a subtitle file, resulting in narration subtitles in multiple languages.

[0097] Step S43: Add the narration audio and narration subtitles to the initial summary video.

[0098] For example, a pre-defined command-line tool (e.g., FFmpeg) can be used to embed the narration audio and subtitles into the video file of the initial summary video to achieve the addition. It should be noted that the operation of adding narration audio and subtitles can be performed before or after adding IoT data to the initial summary video, and this embodiment of the invention does not limit this.

[0099] To address the issue of traditional methods lacking effective explanations and technical clarifications in videos, this invention generates explanatory text for the initial summary video based on its content. Based on the explanatory text, corresponding audio and subtitles are then generated for the initial summary video. Finally, the audio and subtitles are added to the initial summary video. This increases the comprehensibility of the final target summary video, helping stakeholders quickly understand the video content, improving user comprehension, and meeting the needs of non-technical personnel or those conducting cross-industry procurement.

[0100] Optionally, embodiments of the present invention may further include the following steps:

[0101] Step S51: Receive the demand description information input by the demander.

[0102] Step S52: Based on the demand description information and the capability tag information of the candidate suppliers, determine the supplier that matches the demander's demand as the target supplier.

[0103] The steps for outputting the target summary video described above may specifically include: Step 1041, if the supplier is the target supplier, outputting the target summary video to the requester.

[0104] In this embodiment of the invention, the demand description information can be natural language, for example, the demand description information can be "a smart sensor with sound and light alarm function". The demand description information can represent the user's purchasing intention. The candidate suppliers can be suppliers connected to the proficiency verification system. Through the above steps 101 to 103, the target summary video and authenticity test results of each candidate supplier are first determined. The capability tag information can also be called capability tag text. The capability tag information is used to represent the content that the supplier can supply (e.g., what items, services, etc.) and its performance indicators, as well as the supplier's process capabilities and manufacturing capabilities in the production process, such as sensor calibration capability, aging test capability, automated detection capability, etc. One capability tag information is used to represent one capability element. The capability element can also be called capability parameter. The capability tag information of each supplier can be pre-set for the supplier based on the actual situation of the supplier.

[0105] After receiving the requirement description text, the capability verification system can use this information as input to its inference engine, which includes a large language model, an external expert knowledge base, and a knowledge graph. The large language model is a pre-trained language model with recommendation capabilities, such as a Personalization Aware LLMs for Recommendation (PALR). The expert knowledge base is a pre-built knowledge base containing industry experts' knowledge of the suppliers. This includes detailed descriptions of different suppliers' products, performance indicators, process capabilities, manufacturing capabilities, and technical requirements for specific products, helping the large language model to more accurately understand supplier capabilities and user needs. Text from the expert knowledge base can be pre-converted into vectors and added to the large language model's vector library. The knowledge graph is a pre-built graph covering elements such as supplier name, products, performance indicators, process capabilities, manufacturing capabilities, and industry, along with their relationships. For example, the supplier's products can be treated as entities, other elements as attributes of those entities, and relationships can be established to create the knowledge graph.

[0106] The inference engine can intelligently parse the requirement description text. Specifically, it performs natural language processing on the requirement description information, such as word segmentation, to obtain the keywords included in the requirement description. These keywords are then converted into text vectors. The vector distance between the target vector and the corresponding text vector in the large language model's vector library is calculated as the first vector distance. Vectors in the vector library with a first vector distance less than a preset distance threshold are selected as target vectors. The smaller the first vector distance, the more similar the two vectors are, indicating a greater similarity between the semantics of the requirement description and the corresponding text in the vector library. The target vector represents the capabilities that the requester requires from the supplier. Then, a matching analysis is performed between the target vector and the capability tag information of each candidate supplier to filter out suitable target suppliers for the requester's final decision. Specifically, the vector distance between the target vector and the capability tag information vectors of each candidate supplier can be calculated as the second vector distance. Suppliers corresponding to vectors with a second vector distance less than a preset distance threshold are selected as candidate suppliers, and a recommendation score is determined for each candidate supplier based on the second vector distance. The smaller the second vector distance, the higher the recommendation score.

[0107] Furthermore, the demand description text can be transformed into a knowledge graph query statement. For example, the entity required by the demander (e.g., product name) and the attributes possessed by that entity can be identified based on the demand description text. Using a knowledge graph query algorithm, the required entity is located, and the suppliers corresponding to the located entities are selected as candidate suppliers. Simultaneously, a recommendation score is determined for each candidate supplier based on the number of attributes required by the demander. The higher the number of required attributes a candidate supplier possesses, the higher its recommendation score. Finally, candidate suppliers with recommendation scores greater than a preset threshold are identified as target suppliers and output in list form, thereby achieving precise matching of supply and demand capabilities.

[0108] In this embodiment of the invention, based on the demand description information of the demander and the capability tag information of the candidate suppliers, suppliers matching the demander's needs are automatically selected as target suppliers. The demander is then provided with a target supplier summary video and authenticity verification results. This avoids the problem of low accuracy caused by subjective judgment in manual screening, improving the scientific and targeted nature of the demander's procurement decisions. Only a demand description information needs to be input; the inference engine performs intelligent matching to determine the target suppliers that meet the requirements. This optimizes the matching efficiency between demanders and suppliers.

[0109] Specifically, in this embodiment of the invention, for each target supplier, if the authenticity detection result of the target supplier is a second trusted label, then the target supplier is recommended to the demand party, that is, the target summary video of the supplier and the authenticity detection result are output. Otherwise, the target supplier is removed. For example, the capability matching and recommendation module can be entered, where the capability matching and recommendation module determines the recommendation score according to the above method, and outputs a list of recommended target suppliers according to the recommendation score.

[0110] It should be noted that after receiving the requirement description text, the system can first check the historical search results to see if there are any historical search results containing the same requirement description text. If so, the historical supplier in that search result will be directly selected as the target supplier. Each historical search result represents a historical case.

[0111] In this embodiment of the invention, the target summary videos of each target supplier that meet the client's requirements, along with the authenticity detection results, are ultimately output to the client. The target summary videos include multimodal information such as IoT data, audio narration, and subtitles to facilitate capability verification by the client. It should be noted that a recommendation list containing supplier identifiers for all target suppliers can also be output. Based on the authenticity detection results of the target summary videos and the accompanying first suspicious tag / first trust tag, a capability assessment report is generated and output to the client for easy review. In this embodiment of the invention, target suppliers are automatically determined based on the requirement description text, and the target summary videos and authenticity detection results of the target suppliers are output to the client; therefore, the solution possesses a high level of automation and intelligence.

[0112] Figure 4 This is a schematic diagram of a processing step provided by an embodiment of the present invention, such as... Figure 4 As shown, video capture is performed first to obtain real-world video. Next, video summarization and compression are performed to obtain an initial summary video. After speech synthesis and subtitle generation, narration subtitles and audio are added to the initial summary video. A video verification step is then performed to obtain the authenticity test results. An IoT data matching step is then performed to add IoT data to the initial summary video. Finally, a knowledge reasoning and capability recommendation step is performed to identify target suppliers, and target summary videos and authenticity test results for each supplier are output to the client.

[0113] In the field of intelligent manufacturing and supply chain management, there exists a complex supply chain environment, where suppliers' production processes are often quite intricate. In this embodiment of the invention, a video summary is generated and IoT data is synchronously matched over time. Simultaneously, the summary video is verified, and narration subtitles and audio are added, thereby constructing a more credible, comprehensive, efficient, and intelligent supplier capability verification framework, which facilitates quick capability verification for demanders.

[0114] Figure 5 This is a schematic diagram of the structure of a supplier capability verification system based on real-scene video and IoT data provided in an embodiment of the present invention, as shown below. Figure 5 As shown, the system may specifically include:

[0115] The acquisition module 201 is used to acquire real-time video and IoT datasets from the supplier's site.

[0116] The first generation module 202 is used to generate a summary video for the real-scene video based on the real-scene video and preset events in the supplier's production process, so as to obtain an initial summary video.

[0117] The first adding module 203 is used to add the IoT data to the initial summary video based on the collection time of the IoT data in the IoT dataset and the shooting time of the initial summary video, so as to obtain the target summary video;

[0118] The first output module 204 is used to output the target summary video so that the demander can verify the supplier based on the target summary video.

[0119] Optionally, the first adding module 203 is specifically used for:

[0120] For any IoT data in the IoT dataset, find the video frame whose shooting timestamp matches the collection timestamp of the IoT data from the initial summary video, and use it as the target video frame corresponding to the IoT data.

[0121] The target video frame corresponding to the IoT data is overlaid with the IoT data.

[0122] Optionally, the system further includes:

[0123] The identification module is used to identify the device identifier of the production equipment captured in the real-scene video as a first identifier, and to obtain the device identifier carried in the IoT data set as a second identifier;

[0124] The second adding module is used to add a first trusted tag to the target summary video when the first identifier matches the second identifier;

[0125] The third adding module is used to add a first suspicious tag to the target summary video when the first identifier and the second identifier do not match.

[0126] Optionally, the system further includes:

[0127] The detection module is used to select video frames from the initial summary video as video frames to be verified, and to perform authenticity detection on each video frame to be verified to obtain the frame detection result of each video frame to be verified; the frame detection result is used to characterize whether the video frame to be verified has any forgery traces.

[0128] The second generation module is used to generate an authenticity detection result based on the frame detection results of each of the video frames to be verified; the authenticity detection result is a second trusted label or a second suspicious label.

[0129] The second output module is used to output the authenticity detection result to the requester.

[0130] Optionally, the system further includes:

[0131] The receiving module is used to receive the demand description information input by the demander;

[0132] The determination module is used to determine, based on the demand description information and the capability tag information of the candidate suppliers, a supplier that matches the demander's demand and is selected as the target supplier.

[0133] The first output module is specifically used to: output the target summary video to the demander when the supplier is the target supplier.

[0134] Optionally, the first generation module 202 is specifically used for:

[0135] According to the preset events in the supplier's production process, the real-scene video is divided into multiple frame segments, and the segment features of each frame segment are obtained.

[0136] Based on the segment features of each frame segment, an importance score is determined for each frame segment; the importance score is positively correlated with the correlation between the frame segment and the preset event.

[0137] According to the preset summary video duration, the importance score of each frame segment, and the segment duration, target frame segments are selected, and the initial summary video is composed based on the target frame segments.

[0138] Optionally, the first generation module 202 is further configured to:

[0139] Each of the frame segments is compressed, and the duration of each compressed frame segment is determined.

[0140] Using the duration of the summary video as a constraint, target frame segments are selected based on the importance scores and segment durations of each compressed frame segment; the total duration of the target frame segments is not greater than the duration of the summary video.

[0141] Optionally, the system further includes:

[0142] The third generation module is used to: generate explanatory text for the initial summary video based on the video content of the initial summary video;

[0143] The fourth generation module is used to generate narration audio and narration subtitles corresponding to the narration text for the initial summary video based on the narration text;

[0144] The fourth module is used to add the narration audio and narration subtitles to the initial summary video.

[0145] In summary, the supplier capability verification system based on real-scene video and IoT data provided in this embodiment of the invention acquires real-scene video and IoT datasets of the supplier's site. Based on the real-scene video and preset events in the supplier's production process, a summary video is generated from the real-scene video to obtain an initial summary video. Based on the acquisition time of IoT data in the IoT dataset and the shooting time of the initial summary video, IoT data is added to the initial summary video to obtain a target summary video. The target summary video is output for the demand side to verify the supplier based on the target summary video. In this way, the target summary video is generated based on the real-scene video and provided to the demand side. Since the target summary video is shorter, has higher information density, and lower redundancy, it can reduce the analysis cost for the demand side and improve verification efficiency. At the same time, by adding IoT data to the initial summary video according to the time dimension to obtain the target summary video, the system can simultaneously display the video content and IoT data of the supplier's actual site when the demand side views the target summary video. This allows the demand side to more intuitively combine the information from these two modalities of video content and IoT data to verify the supplier, further improving verification efficiency.

[0146] It should be noted that the functional module division method of the proficiency verification system is not unique. Optionally, the present invention also provides a supplier proficiency verification system based on real-scene video and IoT data. This proficiency verification system includes multiple functional modules, and the above-mentioned supplier proficiency verification method based on real-scene video and IoT data is implemented based on these multiple functional modules. For example, the proficiency verification system may include a video acquisition module, a video summarization and compression module, a speech synthesis and subtitle generation module, a video verification module, an IoT data matching module, and a knowledge reasoning and capability recommendation module. The multiple functional modules in this system work together to implement the steps in the above-mentioned supplier proficiency verification method embodiment based on real-scene video and IoT data in actual operation, and achieve the same technical effect. For example, the video acquisition module is used to implement step 101, the video summarization and compression module is used to implement step 102, the speech synthesis and subtitle generation module is used to implement steps S41 to S43, the video verification module is used to implement steps S31 to S33, the IoT data matching module is used to implement step 103 and steps S21 to S22, the knowledge reasoning and ability recommendation module is used to implement steps S51 to S52 and step 104.

[0147] This system is applicable to various industrial scenarios and diverse device access environments. Specifically, the IoT data matching module supports mainstream industrial communication protocols, facilitating rapid integration with smart factory systems from various suppliers. This enables platform-based capability verification services, providing a solid technical foundation and significant potential for the promotion of intelligent industrial procurement and supply chain management. The capability verification system employs a distributed streaming media processing architecture, supporting concurrent batch processing of multiple video sources. This allows for concurrent processing of real-world video from multiple suppliers, facilitating simultaneous capability verification of multiple suppliers.

[0148] This proficiency testing system targets procurement processes in industrial scenarios. It synchronously displays summary videos of real-world processes alongside IoT data, intelligently analyzes the videos to generate authenticity verification results, and adds narration and subtitles. This assists demanders in intelligently verifying supplier capabilities, improving the efficiency and credibility of supplier verification and ultimately aiding decision-making. It's worth noting that the functional modules can be flexibly integrated according to business scenarios. For example, the video summarization and compression module and the video verification module can be merged into a unified video processing unit, and the expert knowledge base of the knowledge reasoning and capability recommendation module can be integrated into other systems as an external service interface. Furthermore, some modules can be trimmed or replaced as needed. For instance, in trusted scenarios where video verification is not required, the video verification module can be skipped, thus meeting customized needs and demonstrating high scalability.

[0149] Furthermore, the capability verification system provided by this invention can be deployed on a set of computing devices with high-performance data processing and inference capabilities. For example, the capability verification system can be flexibly deployed on various types of device platforms, including but not limited to industrial edge servers, enterprise private cloud clusters, or public cloud computing environments. The system can provide a unified interface as a service, thus supporting concurrent access by multiple users and cross-vendor data access, giving the system good scalability and practical deployment value.

[0150] It should be noted that the module division, system structure, operation sequence and processing strategy in the embodiments of the present invention are all illustrative technical implementation methods and do not limit the variations and extensions of the present invention in practical applications.

[0151] The present invention also provides an electronic device, see [link to relevant documentation]. Figure 6 The system includes a processor 801, a memory 802, and a computer program 8021 stored in the memory and executable on the processor. When the processor executes the program, it implements the supplier capability verification method based on real-world video and IoT data as described in the preceding embodiments. The present invention also provides a readable storage medium, wherein instructions in the storage medium, when executed by the processor of an electronic device, enable the electronic device to execute the supplier capability verification method based on real-world video and IoT data as described in the preceding embodiments.

[0152] For system embodiments, since they are substantially similar to method embodiments, the description is relatively simple; relevant details can be found in the descriptions of the method embodiments. The algorithms and displays provided herein are not inherently related to any particular computer, virtual system, or other device. Various general-purpose systems can also be used in conjunction with the teachings herein. Based on the above description, the required structure for constructing such systems is readily apparent. Furthermore, this invention is not directed to any particular programming language. It should be understood that the contents of the invention described herein can be implemented using various programming languages, and the above description of specific languages ​​is for the purpose of disclosing the best mode of implementation of the invention.

[0153] Numerous specific details are set forth in the specification provided herein. However, it will be understood that embodiments of the invention may be practiced without these specific details. In some instances, well-known methods, structures, and techniques have not been shown in detail so as not to obscure the understanding of this specification.

[0154] Similarly, it should be understood that, in order to simplify the invention and aid in understanding one or more of the various inventive aspects, in the above description of exemplary embodiments of the invention, various features of the invention are sometimes grouped together in a single embodiment, figure, or description thereof. However, this disclosure should not be construed as reflecting an intention that the claimed invention requires more features than are expressly recited in each claim. Rather, as reflected in the following claims, inventive aspects lie in fewer than all features of a single foregoing disclosed embodiment. Therefore, the claims following the detailed description are hereby expressly incorporated into this detailed description, wherein each claim itself is a separate embodiment of the invention.

[0155] Those skilled in the art will understand that modules in the device of the embodiments can be adaptively changed and placed in one or more devices different from that embodiment. Modules, units, or components in the embodiments can be combined into a single module, unit, or component, and further, they can be divided into multiple sub-modules, sub-units, or sub-components. Except where at least some of such features and / or processes or units are mutually exclusive, any combination can be used to combine all features disclosed in this specification (including the accompanying claims, abstract, and drawings) and all processes or units of any method or device so disclosed. Unless expressly stated otherwise, each feature disclosed in this specification (including the accompanying claims, abstract, and drawings) may be replaced by an alternative feature that serves the same, equivalent, or similar purpose.

[0156] The various component embodiments of the present invention can be implemented in hardware, or as software modules running on one or more processors, or a combination thereof. Those skilled in the art will understand that microprocessors or digital signal processors (DSPs) can be used in practice to implement some or all of the functions of some or all of the components in the sorting device according to the present invention. The present invention can also be implemented as a device or apparatus program for performing part or all of the methods described herein. Such a program implementing the present invention can be stored on a computer-readable medium, or can be in the form of one or more signals. Such signals can be downloaded from an Internet website, provided on a carrier signal, or provided in any other form.

[0157] It should be noted that the above embodiments are illustrative of the invention and not restrictive, and that those skilled in the art can devise alternative embodiments without departing from the scope of the appended claims. In the claims, any reference signs placed between parentheses should not be construed as limiting the claims. The word "comprising" does not exclude the presence of elements or steps not listed in the claims. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. The invention can be implemented by means of hardware comprising several different elements and by means of a suitably programmed computer. In the unit claims enumerating several means, several of these means may be embodied by the same item of hardware. The use of the words first, second, and third, etc., does not indicate any order. These words can be interpreted as names.

[0158] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be noted that all actions involving the acquisition of signals, information, or data in this application are performed in accordance with the relevant data protection laws and policies of the country where the application is located, and with authorization from the owner of the corresponding device. The above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention. The above descriptions are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A supplier capability verification method based on real-scene video and IoT data, characterized in that, The method includes: Obtain real-time video and IoT datasets from the supplier's site; Based on the real-scene video and preset events in the supplier's production process, a summary video is generated for the real-scene video to obtain an initial summary video. Based on the acquisition time of the IoT data in the IoT dataset and the shooting time of the initial summary video, the IoT data is added to the initial summary video to obtain the target summary video; The target summary video is output so that the demand side can verify the supplier based on the target summary video.

2. The method according to claim 1, characterized in that, The step of adding the IoT data to the initial summary video based on the acquisition time of the IoT data in the IoT dataset and the shooting time of the initial summary video includes: For any IoT data in the IoT dataset, find the video frame whose shooting timestamp matches the collection timestamp of the IoT data from the initial summary video, and use it as the target video frame corresponding to the IoT data. The target video frame corresponding to the IoT data is overlaid with the IoT data.

3. The method according to claim 1, characterized in that, The method further includes: The device identifier of the production equipment captured in the real-scene video is identified as the first identifier, and the device identifier carried in the IoT data set is obtained as the second identifier; If the first identifier matches the second identifier, a first trusted tag is added to the target summary video; If the first identifier and the second identifier do not match, a first suspicious tag is added to the target summary video.

4. The method according to claim 1, characterized in that, The method further includes: Video frames are selected from the initial summary video as video frames to be verified, and authenticity detection is performed on each of the video frames to be verified to obtain the frame detection results of each video frame to be verified; the frame detection results are used to characterize whether there are any traces of forgery in the video frames to be verified. Based on the frame detection results of each of the video frames to be verified, an authenticity detection result is generated; the authenticity detection result is a second trusted label or a second suspicious label. The authenticity test results are output to the requesting party.

5. The method according to claim 1, characterized in that, The method further includes: Receive the demand description information input by the demander; Based on the demand description information and the capability tag information of the candidate suppliers, suppliers that match the demander's needs are identified as target suppliers; The step of outputting the target summary video includes: when the supplier is the target supplier, outputting the target summary video to the requester.

6. The method according to any one of claims 1-5, characterized in that, The process of generating a summary video for the real-scene video based on the real-scene video and preset events in the supplier's production process, to obtain an initial summary video, includes: According to the preset events in the supplier's production process, the real-scene video is divided into multiple frame segments, and the segment features of each frame segment are obtained. Based on the segment features of each frame segment, an importance score is determined for each frame segment; the importance score is positively correlated with the correlation between the frame segment and the preset event. According to the preset summary video duration, the importance score of each frame segment, and the segment duration, target frame segments are selected, and the initial summary video is composed based on the target frame segments.

7. The method according to claim 6, characterized in that, The step of selecting target frame segments according to the preset summary video duration, the importance score of each frame segment, and the segment duration includes: Each of the frame segments is compressed, and the duration of each compressed frame segment is determined. Using the duration of the summary video as a constraint, target frame segments are selected based on the importance scores and segment durations of each compressed frame segment; the total duration of the target frame segments is not greater than the duration of the summary video.

8. The method according to any one of claims 1-5, characterized in that, Before outputting the target summary video, the method further includes: Based on the video content of the initial summary video, generate explanatory text for the initial summary video; Based on the narration text, generate narration audio and narration subtitles corresponding to the narration text for the initial summary video; Add the narration audio and narration subtitles to the initial summary video.

9. A supplier capability verification system based on real-scene video and IoT data, characterized in that, The system includes: The acquisition module is used to acquire real-time video footage and IoT datasets from the supplier's site. The first generation module is used to generate a summary video for the real-scene video based on the real-scene video and preset events in the supplier's production process, so as to obtain an initial summary video. The first adding module is used to add the IoT data to the initial summary video based on the collection time of the IoT data in the IoT dataset and the shooting time of the initial summary video, so as to obtain the target summary video; The first output module is used to output the target summary video so that the demand side can verify the supplier based on the target summary video.

10. An electronic device, characterized in that, include: A processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the supplier capability verification method based on real-world video and IoT data as described in any one of claims 1-8.

11. A readable storage medium, characterized in that, When the instructions in the storage medium are executed by the processor of the electronic device, the electronic device is able to perform one or more of the supplier capability verification methods based on real-world video and IoT data as described in claims 1-8.

Images