A cargo quality monitoring and analysis method and system based on image recognition

Abstract

The present application provides a method and system for monitoring and analyzing the quality of goods based on image recognition, including: obtaining a real-time video stream of the goods to be identified; constructing a multi-dimensional quality standard atlas library related to the products to be monitored; matching the real-time video stream with the atlas in the multi-dimensional quality standard atlas library through an adaptive comparison algorithm to determine whether there is an area in the real-time video stream that matches the preset quality defects, and generating a matching result; if it is determined that there is an area in the real-time video stream that matches the preset quality defects, a response strategy for the defect type corresponding to the area is automatically triggered; and the multi-dimensional quality standard atlas library is dynamically updated to optimize the monitoring accuracy of the quality of goods. The technical solution provided by the present application improves the monitoring effect of the quality of goods.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of data processing technology, and in particular to a cargo quality monitoring and analysis method and system based on image recognition. Background Art

[0002] With the development of modern logistics, the quality monitoring of goods during transportation has become particularly important. Traditional cargo quality inspection mainly relies on manual spot checks, which is not only inefficient but also easily affected by subjective factors, making it difficult to meet the needs of modern efficient logistics. Therefore, the development of a technology that can monitor the quality of goods in real time during transportation has become an urgent problem to be solved. Especially for long-distance transportation and high-value goods, the technical demand for real-time monitoring of the status of goods is particularly prominent to ensure that the goods are delivered to the destination safely and intact.

[0003] Currently, there are some solutions based on sensors and barcode / QR code scanning technology on the market to track the location and status of goods. However, these technologies are mainly used to track the location information of goods, and cannot provide effective monitoring methods for the physical status of goods during transportation (such as damage, deformation, etc.). In addition, there are some preliminary attempts to use fixed-position cameras to perform quality inspections when goods enter and leave the warehouse, but this method cannot cover the changes in the status of goods during the entire transportation process.

[0004] Existing technologies based on sensors and barcode / QR code scanning can only provide location information of goods, but cannot directly reflect the actual status of goods. Although fixed-position cameras can monitor the status of goods in and out of the warehouse to a certain extent, they cannot guarantee the safety of goods during the entire transportation process due to the lack of full-time tracking capabilities. In addition, these methods often perform poorly in the face of complex lighting conditions, especially in the various lighting environments encountered during transportation. The recognition accuracy of traditional methods will drop significantly, which in turn affects the overall monitoring effect. Therefore, a new technical solution is needed to overcome these problems and achieve more comprehensive and accurate cargo quality monitoring. Summary of the invention

[0005] The embodiments of the present application provide a cargo quality monitoring and analysis method and system based on image recognition, so as to solve the problem of poor cargo quality monitoring effect in the prior art.

[0006] In a first aspect, an embodiment of the present application provides a cargo quality monitoring and analysis method based on image recognition, comprising:

[0007] Obtain real-time video stream of the goods to be identified;

[0008] Constructing a multi-dimensional quality standard atlas library related to the goods to be identified, wherein the multi-dimensional quality standard atlas library includes defect type atlas, qualified product atlas, and reference atlas under different lighting conditions;

[0009] Matching the real-time video stream with the atlas in the multi-dimensional quality standard atlas library by an adaptive comparison algorithm to determine whether there is an area in the real-time video stream that matches the preset quality defect, and generating a matching result;

[0010] If it is determined that there is an area in the real-time video stream that matches the preset quality defect, a response strategy for the defect type corresponding to the area is automatically triggered, and the response strategy includes generating alarm information;

[0011] According to the matching results and the execution of the response strategy, the multi-dimensional quality standard atlas library is dynamically updated to optimize the monitoring accuracy of the goods quality.

[0012] Optionally, the constructing of a multi-dimensional quality standard atlas library related to the goods to be identified includes:

[0013] Obtain sample images containing various preset quality defects, classify and annotate each defect, and form a defect type map;

[0014] Selecting samples of goods that meet quality standards from warehouses or logistics centers, taking images of qualified products at different angles and in different states, and classifying and marking the images of qualified products to generate a qualified product map;

[0015] Take images of the same product under different lighting conditions to ensure that all lighting environments encountered by the product during the actual production process are covered, and classify the images of the same product according to the lighting conditions to form a reference map under different lighting conditions;

[0016] The defect type map, the qualified product map and the reference map are integrated into a multi-dimensional quality standard map library.

[0017] Optionally, matching the real-time video stream with the atlas in the multi-dimensional quality standard atlas library by an adaptive comparison algorithm to determine whether there is an area in the real-time video stream that matches a preset quality defect, and generating a matching result, includes:

[0018] Extracting the key area image of the goods to be identified from the real-time video stream;

[0019] Normalizing the key area image, and segmenting the normalized key area image into a plurality of sub-areas using image segmentation technology;

[0020] For each of the sub-regions, applying a feature descriptor extraction method to obtain a visual feature vector of the sub-region;

[0021] Selecting a corresponding reference atlas in the multi-dimensional quality standard atlas library, and obtaining a visual feature vector of the reference atlas using a feature descriptor extraction method;

[0022] Calculating a similarity score between a visual feature vector of each sub-region of the key region image and a visual feature vector of a corresponding sub-region of the reference atlas;

[0023] The similarity score is judged according to a preset similarity threshold to determine whether there is an area in the real-time video stream that matches the preset quality defect; if so, the specific position of the area and the defect type corresponding to the area are located to generate a matching result.

[0024] Optionally, if it is determined that there is an area matching a preset quality defect in the real-time video stream, automatically triggering a response strategy for the defect type corresponding to the area includes:

[0025] Generate an alarm message containing detailed information of the defect according to the specific location of the area and the defect type corresponding to the area;

[0026] Based on the alarm information, a matching response strategy is selected according to a preset rule base, and after the response strategy is implemented, defect detection is re-performed using an adaptive comparison algorithm to verify whether the occurrence of defects has been reduced or eliminated; wherein the preset rule base stores response strategies corresponding to different defect types.

[0027] Optionally, dynamically updating the multi-dimensional quality standard atlas library according to the matching result and the execution of the response strategy to optimize the accuracy of goods quality monitoring includes:

[0028] Collecting the execution status of the response strategy after execution, wherein the execution status includes the accuracy of the alarm information;

[0029] Determining the implementation effect of the response strategy according to the implementation of the response strategy;

[0030] Analyze the defect types in the matching results and the defect features corresponding to the defect types to determine whether there are defect types that are not fully covered, and add the defect types that are not fully covered as defect samples to the defect type atlas to update the multi-dimensional quality standard atlas library;

[0031] In combination with the updated multi-dimensional quality standard atlas library and the execution effect of the response strategy, the relevant parameters of the multi-dimensional quality standard atlas library are dynamically updated to optimize the accuracy of cargo quality monitoring, and the relevant parameters include at least the defect type weight coefficient.

[0032] Optionally, it also includes:

[0033] Dynamically update the defect type weight coefficient in the multi-dimensional quality standard atlas library through the following calculation formula :

[0034] ;

[0035] in, is the preset adjustment factor, is the basic weight coefficient, It's in time Time The detection frequency of the defect type, is the historical maximum detection frequency of this defect type, It is with Nonlinear influencing factors related to defect types, The standardized testing frequency is increased to Second power, It is Severity ratings for each defect type, It is with Factors related to the severity of the defect types, is a positive smoothing parameter, The impact on the weight coefficient, is an exponential decay function, is the total number of defect types; It's in time Time The average color intensity of the key area images, is the average color intensity of the corresponding key area image in the reference atlas, It is The standard deviation of the color intensity of the key area images, is the number of key region images, is the adjustment factor, It is a time series prediction model based on machine learning. It is an adaptive adjustment function based on Bayesian optimization.

[0036] Optionally, it also includes:

[0037] The preset similarity threshold is dynamically updated by the following calculation formula :

[0038] ;

[0039] in, is the basic similarity threshold, is the preset adjustment factor, It is in time The average color intensity of the key area images, is the average color intensity of the corresponding key area image in the reference atlas, It is The root mean square error of the color intensity of the key area image, is the number of key region images, It's in time Time The Euclidean distance between the feature vector of a sub-region and the feature vector of the corresponding sub-region in the reference atlas, is the ideal Euclidean distance, It is The standard deviation of the Euclidean distances of the sub-region feature vectors, is the number of sub-regions, is a periodic adjustment factor, is the current time point, is an adaptive adjustment function based on machine learning. is an adaptive adjustment function based on reinforcement learning.

[0040] is the color intensity difference term,

[0041] is the eigenvector difference term, It is a periodic adjustment item.

[0042] In a second aspect, the embodiment of the present application provides a cargo quality monitoring and analysis system based on image recognition, including:

[0043] An acquisition module, used to acquire a real-time video stream of the goods to be identified;

[0044] A construction module, used to construct a multi-dimensional quality standard atlas library related to the goods to be identified, wherein the multi-dimensional quality standard atlas library includes defect type atlas, qualified product atlas and reference atlas under different lighting conditions;

[0045] A matching module, used to match the real-time video stream with the atlas in the multi-dimensional quality standard atlas library through an adaptive comparison algorithm to determine whether there is an area in the real-time video stream that matches the preset quality defect, and generate a matching result;

[0046] A trigger module, configured to automatically trigger a response strategy for the defect type corresponding to the area if it is determined that there is an area matching a preset quality defect in the real-time video stream, the response strategy including generating alarm information;

[0047] An updating module is used to dynamically update the multi-dimensional quality standard atlas library according to the matching results and the execution of the response strategy to optimize the monitoring accuracy of the goods quality.

[0048] In a third aspect, an embodiment of the present application provides a computing device, comprising a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement a cargo quality monitoring and analysis method and system based on image recognition as described in any one of the first aspects above.

[0049] In a fourth aspect, an embodiment of the present application provides a computer storage medium storing a computer program. When the computer program is executed by a computer, it implements a cargo quality monitoring and analysis method and system based on image recognition as described in any one of the first aspects above.

[0050] In the embodiment of the present application, a real-time video stream of the goods to be identified is obtained; a multi-dimensional quality standard atlas library related to the products to be monitored is constructed; the real-time video stream is matched with the atlas in the multi-dimensional quality standard atlas library through an adaptive comparison algorithm to determine whether there is an area in the real-time video stream that matches the preset quality defects, and a matching result is generated; if it is determined that there is an area in the real-time video stream that matches the preset quality defects, a response strategy for the defect type corresponding to the area is automatically triggered, and the response strategy includes generating alarm information and / or adjusting relevant process parameters on the production line to reduce the incidence of defects; the multi-dimensional quality standard atlas library is dynamically updated to optimize the monitoring accuracy of the goods quality. The technical solution provided by the present application improves the quality of goods and production efficiency, and brings significant cost savings and competitive advantages to enterprises.

[0051] These and other aspects of the present application will become more apparent from the description of the following embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0053] Figure 1A flowchart of a cargo quality monitoring and analysis method based on image recognition provided in an embodiment of the present application;

[0054] Figure 2 A structural diagram of a cargo quality monitoring and analysis system based on image recognition provided in an embodiment of the present application;

[0055] Figure 3 A schematic diagram of the structure of a computing device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0056] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0057] In some of the processes described in the specification and claims of this application and the above-mentioned figures, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this article or executed in parallel. The serial numbers of the operations, such as 101, 102, etc., are only used to distinguish between different operations, and the serial numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this article are used to distinguish different messages, devices, modules, etc., do not represent the order of precedence, and do not limit the "first" and "second" to be different types.

[0058] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of this application.

[0059] Figure 1 A flowchart of a method for monitoring and analyzing cargo quality based on image recognition is provided for an embodiment of the present application, such as Figure 1 As shown, the method includes:

[0060] 101. Obtain real-time video stream of goods to be identified;

[0061] In the on-board video cargo quality recognition scenario, it is first necessary to install a camera or other video acquisition equipment at a suitable position of the transport vehicle to capture the status of the cargo in the compartment in real time. These video streams can be continuous video clips or key frames extracted from the video.

[0062] Suppose a truck is transporting a batch of fragile items, such as beverages in glass bottles. A high-definition camera is installed in the vehicle compartment, and the camera is set to record video at a certain frame rate (such as 30 frames per second). When the vehicle starts, the camera starts working, capturing the status of the goods in the compartment in real time, and transmitting the video data to the image processing system for analysis.

[0063] 102. Construct a multi-dimensional quality standard atlas library related to the goods to be identified, wherein the multi-dimensional quality standard atlas library includes defect type atlas, qualified product atlas, and reference atlases under different lighting conditions;

[0064] In order for the system to accurately identify the status of the goods, it is necessary to first establish a map library containing various quality standards, including known defect type maps, qualified product maps, and reference maps under different lighting conditions.

[0065] Among them, the defect type atlas refers to the collection of sample images containing various preset quality defects. These samples can come from historical data, laboratory tests or defective products in actual production; each defect is labeled and classified in detail, such as scratches, cracks, color difference, shape deviation, etc.; the labeled information includes the location, size, type, etc. of the defect; key features such as edges, textures, color distribution, etc. are extracted from each defect sample. These features will be used in the subsequent matching and identification process; the annotation and feature information are integrated into a defect type atlas. The atlas contains different types of defects and their feature descriptions, providing a reference for defect detection.

[0066] The qualified product atlas refers to selecting samples of goods that meet quality standards from warehouses or logistics centers and taking images of qualified products at different angles and states. Ensure that the samples cover all important areas of the product; Labeling and classification: Classify and mark the qualified product images and record their feature information. This helps the system distinguish between good and bad products during the inspection process; Extract key features from qualified product images, such as edges, textures, color distribution, etc. These features will serve as standard references for normal products; Integrate the labeling and feature information into a qualified product atlas. The atlas provides standard features of high-quality products for comparison and verification of the quality of real-time video streams.

[0067] Reference maps under different lighting conditions refer to images of the same product taken under different lighting conditions (such as natural light, artificial light, and lighting of different intensities and angles). Ensure that all lighting environments that the product may encounter during the actual production process are covered; annotate images under different lighting conditions and record lighting parameters (such as light source type, intensity, angle, etc.); extract key features from images under different lighting conditions, such as edges, textures, color distribution, etc. These features will be used to deal with the impact of lighting changes; integrate annotation and feature information into a reference map under different lighting conditions. This map can help the system maintain stable detection performance under different lighting environments.

[0068] Atlas library integration and standardization refers to integrating defect type atlases, qualified product atlases, and reference atlases under different lighting conditions into a comprehensive database, namely a multi-dimensional quality standard atlas library; through image processing technology (such as image registration, color correction, etc.), all atlases have a unified standard to ensure consistency under different lighting and viewing angles; the atlas library supports real-time updates to reflect the latest quality standard changes and new defect cases. The atlas library content can be continuously optimized and expanded through manual review and machine learning algorithms.

[0069] The application of the atlas library means that during the actual inspection process, the system will use the atlases in the multi-dimensional quality standard atlas library to adaptively compare with the real-time video stream to determine whether there are preset quality defects; if defects are found, the system will locate and classify the defects according to the feature information in the atlas library and generate detailed matching results; based on the matching results, the system will automatically trigger the corresponding response strategy, such as generating alarm information.

[0070] Collect images of different types of goods in good condition as qualified product maps, and classify and annotate known defect types (such as breakage, scratches, etc.) to form defect type maps. At the same time, take images of the same goods under different lighting conditions (such as natural light, dim environment, strong direct light, etc.) to cover various possible lighting environments and form a lighting condition reference map. Integrate these maps into a multi-dimensional quality standard map library.

[0071] Furthermore, the present application considers that in the existing methods for monitoring the quality of goods, although some technologies based on image recognition have been applied, these methods still have some significant problems in the actual production environment. In order to solve the above problems, the present application proposes a method for constructing a multi-dimensional quality standard atlas library, including defect type atlas, qualified product atlas and reference atlas under different lighting conditions. Through this method, the accuracy of defect detection and the adaptability of the system can be improved, thereby better meeting the needs of modern manufacturing industry for high-quality monitoring.

[0072] The options are as follows:

[0073] Optionally, the "constructing a multidimensional quality standard atlas library related to the goods to be identified" in step 102 includes: obtaining sample images containing various preset quality defects, and classifying and labeling each defect to form a defect type atlas; selecting goods samples that meet the quality standards from a warehouse or logistics center, taking images of qualified products at different angles and states, and classifying and labeling the qualified product images to generate a qualified product atlas; taking images of the same product under different lighting conditions to ensure that all lighting environments encountered by the product in the actual production process are covered, and classifying the images of the same product according to the lighting conditions to form a reference atlas under different lighting conditions; integrating the defect type atlas, the qualified product atlas and the reference atlas into a multidimensional quality standard atlas library.

[0074] In an embodiment of the present application, first, sample images of quality defects that may occur in the transportation of various types of goods are collected, such as broken glass bottles, indentations on cartons, scratches on plastic products, etc. These defective samples are carefully classified, for example, all images of broken glass bottles are labeled as "glass breakage", and all images of indentations on cartons are labeled as "packaging deformation", etc. The sample images under each category need to describe in detail the specific characteristics of the defects, such as size, shape, location, etc., to facilitate subsequent comparative identification. Create a database to store these classified defective sample images to form a defect type atlas.

[0075] Secondly, select a batch of undamaged and well-packaged goods as samples to ensure that they meet quality standards. Photograph these qualified goods from different angles, including the front, side, and top, to ensure that the status of the goods can be fully displayed. Mark each qualified product image, indicating the type of goods, shooting angle, and other information. Integrate these marked qualified product images into a qualified product atlas.

[0076] Furthermore, take images of the same batch of goods under different lighting conditions (such as direct sunlight during the day, cloudy days, lighting inside the warehouse, etc.). Ensure that these lighting conditions can represent the various lighting environments that the goods may encounter during actual transportation. Categorize and store images of the same goods under different lighting conditions, for example, "taken under natural light", "taken under indoor lighting", etc. These images will serve as part of the reference atlas under different lighting conditions.

[0077] Finally, the three atlases mentioned above are integrated into the same database or management system to form a multi-dimensional quality standard atlas library. This atlas library should have a good search function to facilitate the rapid search of specific types of images. The atlas library should also have an update mechanism to allow new sample images to be added or images that are no longer applicable to be deleted at any time to maintain the validity and accuracy of the atlas library.

[0078] Through the above steps, a multi-dimensional quality standard atlas library suitable for vehicle-mounted video cargo quality identification scenarios can be constructed. Such an atlas library will help the system to more accurately identify whether the quality of the cargo has changed during transportation and take corresponding measures in a timely manner.

[0079] 103. Match the real-time video stream with the atlas in the multi-dimensional quality standard atlas library by using an adaptive comparison algorithm to determine whether there is an area in the real-time video stream that matches the preset quality defect, and generate a matching result;

[0080] Use an adaptive comparison algorithm to compare each frame of the real-time video stream with the atlas in the multi-dimensional quality standard atlas library, identify whether there are areas in the video that match the preset quality defects, and generate matching results.

[0081] When the video stream enters the image processing system, the algorithm begins to analyze the images in the video frame by frame. For each key frame, the algorithm will automatically identify the goods in the image and compare it with the images in the standard atlas library. If the goods are found to be obviously damaged or not in compliance with the standards, these areas will be marked and the corresponding matching results will be generated.

[0082] This application also takes into account that in the industrial production process, real-time monitoring of goods quality is an important part of ensuring product consistency and improving production efficiency. Traditional image matching methods are often based on the comparison of the entire image. This method has certain limitations when dealing with complex backgrounds, lighting changes, and images taken from different perspectives, such as prone to false alarms and missed detections. In addition, when the key area features of the product to be monitored are not obvious or are obscured, it is difficult for traditional methods to accurately identify quality defects.

[0083] Therefore, this application proposes a more sophisticated and targeted quality inspection solution. This solution improves the accuracy of defect detection by segmenting the key areas in the real-time video stream and comparing each sub-area with the reference atlas in the standard atlas library. This not only overcomes the influence of environmental factors, but also locates the specific defect location and type more accurately, providing reliable data support for subsequent quality control.

[0084] Optionally, in step 103, "matching the real-time video stream with the atlas in the multi-dimensional quality standard atlas library by an adaptive comparison algorithm to determine whether there is an area in the real-time video stream that matches the preset quality defect, and generating a matching result" includes: extracting the key area image of the goods to be identified from the real-time video stream; normalizing the key area image, and using image segmentation technology to segment the normalized key area image into multiple sub-areas; for each of the sub-areas, applying a feature descriptor extraction method to obtain a visual feature vector of the sub-area; selecting a corresponding reference atlas in the multi-dimensional quality standard atlas library, and using a feature descriptor extraction method to obtain a visual feature vector of the reference atlas; calculating a similarity score between the visual feature vector of each sub-area of ​​the key area image and the visual feature vector of the corresponding sub-area of ​​the reference atlas; judging the similarity score according to a preset similarity threshold to determine whether there is an area in the real-time video stream that matches the preset quality defect; if so, locating the specific position of the area and the defect type corresponding to the area to generate a matching result.

[0085] Normalization: Convert the image into a unified standard format, such as adjusting the size and brightness, to eliminate the impact of external conditions.

[0086] Image segmentation technology: decomposes an image into multiple meaningful parts (sub-regions) for further analysis.

[0087] Feature descriptor extraction method: extract information fragments from an image that can represent its unique properties for comparison with other images.

[0088] Visual feature vector: A set of numerical representations obtained using feature descriptors, used to quantitatively describe the visual features of an image.

[0089] Similarity score: A measure of the similarity between two visual feature vectors, where a higher value indicates that the two are closer.

[0090] Preset similarity threshold: A pre-set boundary used to determine whether two sets of data are similar enough.

[0091] In the embodiment of the present application, first, when the video stream enters the image processing system, the system processes each frame of the image. Use object detection algorithms (such as YOLO, SSD, etc.) to identify and select the cargo area in the video. For example, when the system detects a carton, it marks the location of the carton in the video frame and extracts the image of the area as the key area image.

[0092] Secondly, the extracted key area images are normalized, such as adjusting the image size, brightness and contrast, to make them conform to the image format in the standard atlas library. Use image segmentation technology (such as semantic segmentation algorithm or edge detection algorithm) to segment the normalized key area images into multiple sub-areas. For example, a carton is segmented into several sub-areas such as four corners, four sides and the middle part.

[0093] Furthermore, feature descriptor extraction methods (such as SIFT, SURF, ORB, etc.) are applied to each segmented sub-region to obtain the visual feature vector of the sub-region. These feature descriptors can help the system identify key points in the image and their descriptors. For example, for a corner of a carton, its texture features, color distribution and other information are extracted to form a feature vector.

[0094] Furthermore, reference maps corresponding to the current cargo type are selected from the multi-dimensional quality standard map library, including qualified product maps and defect type maps. The feature descriptor extraction method is also applied to the selected reference maps to obtain the visual feature vector of each reference map.

[0095] Furthermore, a preset similarity threshold is set. For example, if the similarity score is lower than a certain threshold (such as 0.7), it is considered that the sub-region may have quality defects. Whether there is a defective area is determined based on the similarity score. If the score is lower than the threshold, it is considered that there is a defect.

[0096] Finally, for areas that are determined to have defects, the system will record the specific location (such as coordinates) of the area in the video and the defect type (such as cracks, dents, etc.), and generate a matching result report that includes information such as the location, type, severity, and recommended treatment measures.

[0097] Through the above steps, the on-board video cargo quality identification system can accurately detect the quality changes of cargo during transportation and generate alarm information in time so that the logistics company can take corresponding measures quickly.

[0098] 104. If it is determined that there is an area in the real-time video stream that matches a preset quality defect, a response strategy for the defect type corresponding to the area is automatically triggered, and the response strategy includes generating alarm information;

[0099] Once the system detects an area with quality defects in the video, it will take action based on the preset response strategy, such as generating an alarm message to notify relevant personnel.

[0100] If the algorithm detects damaged goods in the video, the system will immediately generate an alarm message and send it to the logistics company's control center or the driver's mobile device via a wireless network. At the same time, the system can also record the time and location of the defect for further investigation.

[0101] Furthermore, the present application considers that in the existing cargo quality monitoring system, although it is possible to detect defects and trigger alarms through image recognition technology, there are still some problems in practical application. The existing alarm information often only contains simple prompts and lacks detailed information on the location and type of defects, which makes it difficult for operators to quickly locate and deal with problems; usually, the system can only generate alarm information. This single response method cannot effectively reduce the incidence of defects; after implementing the response strategy, the existing system does not have a further verification step to confirm whether the defects have been eliminated or reduced, which may lead to the problem not being completely solved.

[0102] In order to solve the above problems, this application proposes a more comprehensive and intelligent response strategy solution. This solution not only generates detailed alarm information, but also selects a matching response strategy based on a preset rule base, and re-detects defects using an adaptive comparison algorithm after implementing the response strategy to verify whether the occurrence of defects has been reduced or eliminated. This can improve the response efficiency and accuracy of the system and ensure that quality problems are effectively resolved.

[0103] The options are as follows:

[0104] Optionally, in step 104, "if it is determined that there is an area in the real-time video stream that matches a preset quality defect, a response strategy for the defect type corresponding to the area is automatically triggered", includes: generating an alarm message containing detailed information of the defect according to the specific location of the area and the defect type corresponding to the area; based on the alarm information, selecting a matching response strategy according to a preset rule base, and re-performing defect detection using an adaptive comparison algorithm after implementing the response strategy to verify whether the occurrence of defects has been reduced or eliminated; wherein the preset rule base stores response strategies corresponding to different defect types.

[0105] Alert information: A notification generated by the system containing defect details to alert relevant personnel to pay attention and deal with the problem.

[0106] Preset rule base: A database that stores different defect types and their corresponding response strategies, which is used to guide the system on how to automatically respond when a specific defect is detected.

[0107] Adaptive comparison algorithm: An algorithm that can dynamically adjust parameters based on real-time data to re-detect defects and verify the effectiveness of response strategies.

[0108] In the embodiment of the present application, first, when the system determines that there is an area in the real-time video stream that matches the preset quality defect through an adaptive comparison algorithm, the specific location of the defect area (for example, the coordinate position in the car) and the defect type (such as damage, scratches, etc.) will be recorded. An alarm message containing detailed information of the defect is automatically generated, for example: "A damaged carton was detected at the car coordinate (120,80), suspected to be caused by an impact."

[0109] Secondly, the preset rule base stores the response strategies corresponding to different defect types. For example, for a damaged carton, the response strategy may be to stop the vehicle for inspection immediately, notify the nearest service station to prepare replacement goods, record the damage for claims, etc. Based on the generated alarm information, the system will query the preset rule base to find the response strategy that matches the detected defect type. For the damaged carton in the above example, the system will select the response strategy of stopping the vehicle for inspection immediately according to the preset rules.

[0110] Finally, when the response strategy is executed, the system will re-detect the repaired area using the adaptive comparison algorithm again. For example, after the driver stops to check and confirm the problem, the system will continue to monitor changes in the area. If the repair measures are effective, the system will no longer detect quality problems in the area. If the problem still exists, the system will continue to generate alarm information and may upgrade the response strategy (such as contacting the logistics center for support).

[0111] The following is a specific scenario example:

[0112] Suppose a transport truck is transporting a batch of electronic devices, and the system monitors the status of the cargo in the car in real time through the on-board camera. The system detects that a carton at the coordinate (120,80) has obvious signs of damage. Automatically generate an alarm message: "A damaged carton was detected, coordinate (120,80), suspected to be caused by a collision." The system queries the preset rule base and finds a response strategy that matches the damaged carton: stop and check immediately, and notify the nearest service station to prepare to replace the goods. After receiving the alarm, the driver stops to check and confirms that the carton is indeed damaged. The driver replaces the damaged carton and reseals it. The system uses the adaptive contrast algorithm to detect the area again and confirms that the defect is no longer detected after the repair.

[0113] Through this series of steps, the on-board video cargo quality identification system can not only promptly detect quality problems of cargo during transportation, but also automatically trigger response strategies and verify the repair effects to ensure that the cargo is in good condition before arriving at the destination.

[0114] 105. According to the matching result and the execution of the response strategy, the multi-dimensional quality standard atlas library is dynamically updated to optimize the monitoring accuracy of the goods quality.

[0115] The system will continuously update and improve the multi-dimensional quality standard atlas library based on the results of each test and the implementation of the response strategy to improve the accuracy and efficiency of future identification.

[0116] Whenever the system detects a new defect type or a certain defect type occurs more frequently, it will add the new defect sample to the defect type map and adjust the corresponding response strategy. In this way, the system's recognition ability and response speed will be improved over time. At the same time, the reference map under lighting conditions can be adjusted according to actual conditions to ensure that the status of the goods can be accurately identified under various lighting conditions.

[0117] Furthermore, the present application considers that in the existing cargo quality monitoring system, although it has been possible to detect defects and trigger response strategies through image recognition technology, there are still some problems in practical applications. The existing multi-dimensional quality standard atlas library is usually static and lacks a dynamic update mechanism; this makes it difficult for the system to adapt to new defect types and changes in the production environment, thereby affecting the detection accuracy; the existing system lacks a detailed evaluation and feedback mechanism for the effect of the strategy after implementing the response strategy; this may cause the response strategy to fail to achieve the expected effect and may even introduce new problems; the existing atlas library may not fully cover all possible defect types, resulting in some types of defects cannot be effectively detected. In order to solve the above problems, the present application proposes a more comprehensive and intelligent dynamic update scheme. This scheme not only updates the atlas library according to the execution of the response strategy, but also discovers defect types that are not fully covered by analyzing the matching results, and adjusts the relevant parameters to optimize the detection accuracy. This can improve the system's adaptability and detection accuracy, and ensure that quality problems are effectively solved.

[0118] The options are as follows:

[0119] Optionally, "dynamically updating the multidimensional quality standard atlas library according to the matching results and the execution of the response strategy to optimize the accuracy of cargo quality monitoring" in step 105 includes: collecting the execution status after executing the response strategy, the execution status including the accuracy of the alarm information; determining the execution effect of implementing the response strategy according to the execution status of the response strategy; analyzing the defect types in the matching results and the defect characteristics corresponding to the defect types to determine whether there are defect types that are not fully covered, and adding the defect types that are not fully covered as defect samples to the defect type atlas to update the multidimensional quality standard atlas library; in combination with the updated multidimensional quality standard atlas library and the execution effect of the response strategy, dynamically updating the relevant parameters of the multidimensional quality standard atlas library to optimize the accuracy of cargo quality monitoring, and the relevant parameters include at least the defect type weight coefficient.

[0120] Accuracy of alarm information: refers to whether the generated alarm information accurately reflects the actual defects, their locations and types.

[0121] Change in defect incidence rate: refers to the change in the frequency or number of defects after the response strategy is implemented.

[0122] Defect type weight coefficient: A parameter used to adjust the importance of different defect types in the detection process, usually set based on the severity and frequency of the defect.

[0123] In the embodiment of the present application, first, when the system triggers a response strategy, it is necessary to record the execution of the response strategy. This includes the accuracy of the alarm information (i.e., whether the alarm accurately reflects the actual problem). For example, if the system detects that a carton is damaged and issues an alarm, and the driver stops to check and confirms that there is indeed damage, then the alarm information is recorded as accurate; if the driver finds that there is no damage after checking, then the alarm information is recorded as a false alarm.

[0124] Secondly, analyze the accuracy of the alarm information and the effectiveness of the response strategy to evaluate whether the response strategy effectively solves the detected problem. For example, if the alarm information is accurate and the driver replaces the damaged carton according to the response strategy, then the response strategy can be considered to be well implemented; if the alarm information is frequently false alarm or the response strategy fails to solve the problem, the response strategy needs to be further optimized.

[0125] Furthermore, the defect types and their characteristics in the matching results are analyzed. For example, if a new type of damage (such as edge tearing) is detected, and the existing atlas library does not contain such defects, this defect type needs to be added as a new sample to the defect type atlas. For example, if the system detects some cartons with torn edges, but there are no such samples in the atlas library, it is necessary to take some images of cartons with torn edges and add them to the defect type atlas to improve the system's recognition ability.

[0126] Finally, after updating the atlas library, the defect type weight coefficient needs to be adjusted to reflect the importance of the newly added defect type. For example, if edge tearing is a common defect and seriously affects the quality of the goods, the weight coefficient of this defect can be increased so that the system pays more attention to this defect in future inspections. Dynamically adjusting the weight coefficient can enable the system to more accurately identify various defect types in future inspections, thereby improving the accuracy of cargo quality monitoring.

[0127] The following is a specific scenario example:

[0128] Suppose a transport truck is transporting a batch of electronic products. The system monitors the status of the goods in the car in real time through the on-board camera. The system detects that a carton at the coordinate (120,80) is damaged and generates an alarm message: "Damaged carton detected, coordinate (120,80), suspected to be caused by impact." The driver stops the car to check and confirms that the carton is indeed damaged, and records the alarm message as accurate. The driver replaces the damaged carton according to the response strategy and records that the response strategy is executed well. If the system has multiple false alarms, the response strategy needs to be further optimized to reduce false alarms. The system detects a new defect type - edge tear, but this defect is not covered in the existing atlas. Collect some images of carton with edge tear and add them as new samples to the defect type atlas to improve the system's recognition ability. After updating the atlas, adjust the defect type weight coefficient and increase the weight coefficient of edge tear defects to improve the system's sensitivity in future detections. Combined with the updated atlas library and the execution effect of the response strategy, the system can monitor the quality of goods more accurately and take timely measures to ensure the safety and integrity of the goods during transportation.

[0129] Through this series of dynamic update steps, the vehicle-mounted video cargo quality identification system can continuously optimize its detection accuracy and better serve the quality monitoring needs of the logistics industry.

[0130] This application also takes into account that in the existing multi-dimensional quality standard atlas library, the defect type weight coefficient is usually static and is set based on historical data and expert experience. This method has the following problems: once the weight coefficient is set, it often does not adjust dynamically over time and cannot adapt to environmental changes and new problems; the fixed weight coefficient may not be able to reflect the current detection frequency, severity, and changes in image features in a timely manner, resulting in inaccurate quality assessment; only relying on data from a certain dimension (such as historical detection frequency) for weight setting, ignoring other important information (such as image color intensity, machine learning prediction, etc.). In order to solve the above problems, this application proposes a method for dynamically updating the defect type weight coefficient in the multi-dimensional quality standard atlas library through a calculation formula. Through this method, the current quality status can be reflected more comprehensively and flexibly, thereby providing more accurate quality assessment and management recommendations.

[0131] The options are as follows:

[0132] Optionally, the method further comprises:

[0133] Dynamically update the defect type weight coefficient in the multi-dimensional quality standard atlas library through the following calculation formula :

[0134] ;

[0135] in, is the preset adjustment factor, is the basic weight coefficient, It's in time Time The detection frequency of the defect type, is the historical maximum detection frequency of this defect type, It is with Nonlinear influencing factors related to defect types, The standardized testing frequency is increased to Second power, It is Severity ratings for each defect type, It is with Factors related to the severity of the defect types, is a positive smoothing parameter, The impact on the weight coefficient, is an exponential decay function, is the total number of defect types; It's in time Time The average color intensity of the key area image is the average color intensity of the corresponding key area image in the reference atlas. It is The standard deviation of the color intensity of the key area images, is the number of key region images, is the adjustment factor, It is a time series prediction model based on machine learning. It is an adaptive adjustment function based on Bayesian optimization.

[0136] Initialize the basic weight coefficient :

[0137] Basic weight coefficient

[0138] According to historical data and expert experience, the basic weight coefficient for each defect type is set.

[0139] Calculate the normalized detection frequency:

[0140] Detection frequency , the historical maximum detection frequency , nonlinear impact factor

[0141] Calculate the normalized detection frequency for each defect type And raise it to a nonlinear impact factor Power.

[0142] Calculate the impact of severity score:

[0143] Severity Rating , severity-related influencing factors

[0144] Calculate a severity score for each defect type .

[0145] Compute the exponential decay function:

[0146] Smoothing parameter , the change of weight coefficient

[0147] Calculates the exponential decay function , to take into account the impact of changes in weight coefficients.

[0148] Calculate the color intensity adjustment factor:

[0149] The average color intensity of the key area image , the average color intensity in the reference spectrum , standard deviation of color intensity , calculate the color intensity adjustment factor for each key region image .

[0150] Apply machine learning prediction models and adaptive tuning functions:

[0151] Time series prediction model based on machine learning ,Adaptive adjustment function based on Bayesian optimization

[0152] Leverage trained machine learning models and Bayesian optimization for adaptive adjustment functions Make forecasts and adjustments.

[0153] Interpretation and application of results:

[0154] Weight coefficient

[0155] The calculated new weight coefficients are applied to the multidimensional quality standard atlas library, and quality assessment and management decisions are made based on them.

[0156] In the embodiment of the present application, it is assumed that the weight coefficient of the type of surface defect of the goods needs to be dynamically updated. The following are the specific implementation steps:

[0157] Set basic weight coefficients based on historical data and expert experience =[0.2, 0.3, 0.1, 0.4], representing the initial weights of the four defect types A, B, C, and D respectively.

[0158] Assume that at time When , the detection frequencies of the four defect types are:

[0159] .

[0160] The historical maximum detection frequencies are:

[0161] .

[0162] The nonlinear impact factors are .

[0163] Calculate the normalized detection frequency:

[0164] ;

[0165] Assume that the severity scores of the four defect types are .

[0166] The severity-related influencing factors are .

[0167] Calculate the impact of severity score:

[0168] ;

[0169] Assuming a change in weight coefficients , smoothing parameter .

[0170] Compute the exponential decay function:

[0171] ;

[0172] Calculate color intensity adjustment factor

[0173] Suppose there are two key region images, whose average color intensities are .

[0174] The average color intensities in the reference spectra are .

[0175] The standard deviations of color intensity are .

[0176] Calculate the color intensity adjustment factor:

[0177] ;

[0178] Assume a time series prediction model based on machine learning The output is 0.7, an adaptive adjustment function based on Bayesian optimization The output is 0.6.

[0179] Assume that the adjustment factors are .

[0180] Calculation process:

[0181] ;

[0182] ;

[0183] New weight coefficient They are 0.89258, 1.06918, 0.65118, and 1.13208 respectively, indicating that defect types B and D have higher weights at the current time point.

[0184] Its quality control strategy can be adjusted according to these new weight coefficients to strengthen the detection and processing of high-weight defect types.

[0185] By dynamically updating the weight coefficients, it can better adapt to environmental changes and emerging problems.

[0186] Combining information from multiple dimensions such as detection frequency, severity, and image features provides a more comprehensive quality assessment.

[0187] The calculation accuracy of weight coefficients is improved by using machine learning and Bayesian optimization methods.

[0188] Based on more accurate weight coefficients, management departments can make more scientific and reasonable decisions.

[0189] Through continuous feedback and adjustment, quality control strategies can be gradually optimized and cargo quality can be continuously improved.

[0190] The present application also takes into account that in existing image similarity detection technologies, the preset similarity threshold is usually fixed and set based on historical data and expert experience. This method has the following problems: the fixed threshold cannot adapt to environmental changes and emerging problems, which may lead to misjudgment or missed judgment; the fixed threshold may not be able to reflect the changes in the current image features in a timely manner, resulting in inaccurate similarity judgment; and only relying on data of a certain dimension (such as color intensity) for similarity judgment ignores other important information (such as feature vectors, time periodicity, etc.). In order to solve the above problems, the present application proposes a method for dynamically updating the preset similarity threshold through a calculation formula. Through this method, the current image features can be reflected more comprehensively and flexibly, thereby providing more accurate similarity judgments and management recommendations.

[0191] The options are as follows:

[0192] Optionally, it also includes:

[0193] The preset similarity threshold is dynamically updated by the following calculation formula :

[0194] ;

[0195] in, is the basic similarity threshold, is the preset adjustment factor, It is in time The average color intensity of the key area images, is the average color intensity of the corresponding key area image in the reference atlas, It is The root mean square error of the color intensity of the key area image, is the number of key region images, It's in time Time The Euclidean distance between the feature vector of a sub-region and the feature vector of the corresponding sub-region in the reference atlas, is the ideal Euclidean distance, It is The standard deviation of the Euclidean distances of the sub-region feature vectors, is the number of sub-regions, is a periodic adjustment factor, is the current time point, is an adaptive adjustment function based on machine learning. is an adaptive adjustment function based on reinforcement learning. is the color intensity difference term, is the eigenvector difference term, It is a periodic adjustment item.

[0196] Initialize the basic similarity threshold :

[0197] Basic similarity threshold

[0198] Set the initial similarity threshold based on historical data and expert experience .

[0199] Compute the color intensity difference term:

[0200] The average color intensity of the key area image , the average color intensity in the reference spectrum , the root mean square error of color intensity , calculate the color intensity difference of each key region image , and then sum and take the logarithm to get the color intensity difference term .

[0201] Compute the eigenvector difference term:

[0202] The Euclidean distance between the feature vector of the subregion and the feature vector of the corresponding subregion in the reference atlas , ideal Euclidean distance , the standard deviation of the Euclidean distance

[0203] Calculate the difference of eigenvectors for each sub-region , and then sum and square root to get the eigenvector difference term .

[0204] Calculate periodic adjustments:

[0205] Periodic Adjustment Factor , the current time point

[0206] Calculating periodic adjustments , to take into account the effect of time periodicity.

[0207] Apply the adaptive adjustment function:

[0208] Adaptive adjustment function based on machine learning ,Adaptive adjustment function based on reinforcement learning

[0209] Leverage trained machine learning models and reinforcement learning models Make adaptive adjustments.

[0210] Interpretation and application of results:

[0211] Similarity Threshold

[0212] The calculated new similarity threshold is applied to image similarity detection, and quality assessment and management decisions are made based on it.

[0213] In the embodiment of the present application, it is assumed that the similarity threshold of the product surface defect detection needs to be dynamically updated. The following are the specific implementation steps:

[0214] Set the basic similarity threshold based on historical data and expert experience .

[0215] Suppose there are two key region images, whose average color intensities are .

[0216] The average color intensities in the reference spectra are .

[0217] The root mean square errors of color intensity are .

[0218] Compute the color intensity difference term:

[0219] ;

[0220] Assume that there are three sub-regions, and the Euclidean distances between their feature vectors and the feature vectors of the corresponding sub-regions in the reference atlas are .

[0221] The ideal Euclidean distances are .

[0222] The standard deviations of the Euclidean distances are .

[0223] Compute the eigenvector difference term:

[0224] ;

[0225] Assume the current time is =10, periodic adjustment factor =20.

[0226] Calculate periodic adjustments:

[0227] ;

[0228] Assume that the adaptive adjustment function based on machine learning The output is 0.1, an adaptive adjustment function based on reinforcement learning The output is 0.2. Comprehensive calculation of similarity threshold ;

[0229] Assume that the adjustment factors are .

[0230] Calculation process:

[0231] ;

[0232] New similarity threshold It is 1.50759, indicating that the similarity threshold at the current time point has increased.

[0233] Enterprises can adjust their image similarity detection algorithms based on this new threshold to improve detection accuracy and reduce false positives and missed positives.

[0234] By dynamically updating the similarity threshold, it can better adapt to environmental changes and emerging problems.

[0235] Combining information from multiple dimensions such as color intensity, feature vectors, and time periodicity provides a more comprehensive similarity judgment.

[0236] The calculation accuracy of the similarity threshold is improved by using machine learning and reinforcement learning methods.

[0237] Based on more accurate similarity thresholds, management departments can make more scientific and reasonable decisions.

[0238] Through continuous feedback and adjustment, the similarity detection strategy can be gradually optimized and the quality of goods can be continuously improved.

[0239] Figure 2 A structural diagram of a cargo quality monitoring and analysis system based on image recognition is provided for an embodiment of the present application, such as Figure 2 As shown, the device comprises:

[0240] An acquisition module 21 is used to acquire a real-time video stream of the goods to be identified;

[0241] A construction module 22 is used to construct a multi-dimensional quality standard atlas library related to the goods to be identified, wherein the multi-dimensional quality standard atlas library includes defect type atlas, qualified product atlas and reference atlas under different lighting conditions;

[0242] A matching module 23 is used to match the real-time video stream with the atlas in the multi-dimensional quality standard atlas library through an adaptive comparison algorithm to determine whether there is an area in the real-time video stream that matches the preset quality defect, and generate a matching result;

[0243] A trigger module 24 is used to automatically trigger a response strategy for the defect type corresponding to the area if it is determined that there is an area matching the preset quality defect in the real-time video stream, and the response strategy includes generating alarm information;

[0244] The updating module 25 is used to dynamically update the multi-dimensional quality standard atlas library according to the matching results and the execution of the response strategy to optimize the monitoring accuracy of the goods quality.

[0245] Figure 2 The cargo quality monitoring and analysis system based on image recognition can be performed Figure 1 The implementation principle and technical effect of the cargo quality monitoring and analysis method based on image recognition described in the illustrated embodiment will not be described in detail. The specific manner in which each module and unit performs operations in the cargo quality monitoring and analysis system based on image recognition in the above embodiment has been described in detail in the embodiment of the method, and will not be described in detail here.

[0246] In one possible design, Figure 2 The cargo quality monitoring and analysis system based on image recognition of the embodiment shown can be implemented as a computing device, such as Figure 3 As shown, the computing device may include a storage component 31 and a processing component 32;

[0247] The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are called and executed by the processing component 32 .

[0248] The processing component 32 is used to: obtain a real-time video stream of the goods to be identified; construct a multi-dimensional quality standard atlas library related to the products to be monitored; match the real-time video stream with the atlas in the multi-dimensional quality standard atlas library through an adaptive comparison algorithm to determine whether there is an area in the real-time video stream that matches the preset quality defects, and generate a matching result; if it is determined that there is an area in the real-time video stream that matches the preset quality defects, automatically trigger a response strategy for the defect type corresponding to the area, and the response strategy includes generating alarm information; dynamically update the multi-dimensional quality standard atlas library to optimize the monitoring accuracy of the goods quality.

[0249] The processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above method. Of course, the processing component may also be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors or other electronic components to perform the above method.

[0250] The storage component 31 is configured to store various types of data to support operations at the terminal. The storage component can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0251] Of course, the computing device may also include other components, such as input / output interfaces, display components, communication components, etc.

[0252] The input / output interface provides an interface between the processing component and the peripheral interface module, which may be an output device, an input device, etc.

[0253] The communication component is configured to facilitate, among other things, wired or wireless communications between the computing device and other devices.

[0254] Among them, the computing device can be a physical device or an elastic computing host provided by a cloud computing platform, etc. In this case, the computing device can refer to a cloud server, and the above-mentioned processing components, storage components, etc. can be basic server resources rented or purchased from the cloud computing platform.

[0255] The present application also provides a computer storage medium storing a computer program, wherein the computer program can achieve the above-mentioned Figure 1 The illustrated embodiment is a method for monitoring and analyzing cargo quality based on image recognition.

[0256] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0257] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0258] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0259] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A cargo quality monitoring and analysis method based on image recognition, characterized in that: include: Obtain real-time video stream of the goods to be identified; Constructing a multi-dimensional quality standard atlas library related to the goods to be identified, wherein the multi-dimensional quality standard atlas library includes defect type atlas, qualified product atlas, and reference atlas under different lighting conditions; Matching the real-time video stream with the atlas in the multi-dimensional quality standard atlas library by an adaptive comparison algorithm to determine whether there is an area in the real-time video stream that matches the preset quality defect, and generating a matching result; If it is determined that there is an area in the real-time video stream that matches the preset quality defect, a response strategy for the defect type corresponding to the area is automatically triggered, and the response strategy includes generating alarm information; Dynamically updating the multi-dimensional quality standard atlas library according to the matching results and the execution of the response strategy to optimize the monitoring accuracy of the goods quality; The method of matching the real-time video stream with the atlas in the multi-dimensional quality standard atlas library by an adaptive comparison algorithm to determine whether there is an area in the real-time video stream that matches the preset quality defect, and generating a matching result, includes: Extracting the key area image of the goods to be identified from the real-time video stream; Normalizing the key area image, and segmenting the normalized key area image into multiple sub-areas using image segmentation technology; For each of the sub-regions, applying a feature descriptor extraction method to obtain a visual feature vector of the sub-region; Selecting a corresponding reference atlas in the multi-dimensional quality standard atlas library, and obtaining a visual feature vector of the reference atlas using a feature descriptor extraction method; Calculating a similarity score between a visual feature vector of each sub-region of the key region image and a visual feature vector of a corresponding sub-region of the reference atlas; The similarity score is judged according to a preset similarity threshold to determine whether there is an area in the real-time video stream that matches the preset quality defect; if so, the specific position of the area and the defect type corresponding to the area are located to generate a matching result; The dynamically updating the multi-dimensional quality standard atlas library according to the matching result and the execution of the response strategy to optimize the accuracy of goods quality monitoring includes: Collecting the execution status of the response strategy after execution, wherein the execution status includes the accuracy of the alarm information; Determining the implementation effect of the response strategy according to the implementation of the response strategy; Analyze the defect types in the matching results and the defect features corresponding to the defect types to determine whether there are defect types that are not fully covered, and add the defect types that are not fully covered as defect samples to the defect type atlas to update the multi-dimensional quality standard atlas library; In combination with the updated multi-dimensional quality standard atlas library and the execution effect of the response strategy, dynamically update the relevant parameters of the multi-dimensional quality standard atlas library to optimize the accuracy of goods quality monitoring, wherein the relevant parameters at least include a defect type weight coefficient; The method further comprises: Dynamically update the defect type weight coefficient in the multi-dimensional quality standard atlas library through the following calculation formula : ； in, is the preset adjustment factor, is the basic weight coefficient, It's in time Time The detection frequency of the defect type, is the historical maximum detection frequency of this defect type, It is with Nonlinear influencing factors related to defect types, The standardization test frequency is increased to Second power, It is Severity ratings for each defect type, It is with Factors related to the severity of the defect types, is a positive smoothing parameter, The impact on the weight coefficient, is an exponential decay function, is the total number of defect types; It's in time Time The average color intensity of the key area images, is the average color intensity of the corresponding key area image in the reference atlas, It is The standard deviation of the color intensity of the key area images, is the number of key region images, is the adjustment factor, It is a time series prediction model based on machine learning. It is an adaptive adjustment function based on Bayesian optimization; The method further comprises: The preset similarity threshold is dynamically updated by the following calculation formula : ； in, is the basic similarity threshold, is the preset adjustment factor, It is in time The average color intensity of the key area images, is the average color intensity of the corresponding key area image in the reference atlas, It is The root mean square error of the color intensity of the key area image, is the number of key region images, It's in time Time The Euclidean distance between the feature vector of a sub-region and the feature vector of the corresponding sub-region in the reference atlas, is the ideal Euclidean distance, It is The standard deviation of the Euclidean distances of the sub-region feature vectors, is the number of sub-regions, is a periodic adjustment factor, is the current time point, is an adaptive adjustment function based on machine learning. is an adaptive adjustment function based on reinforcement learning. is the color intensity difference term, is the eigenvector difference term, It is a periodic adjustment item.

2. The method according to claim 1, characterized in that The constructing of a multi-dimensional quality standard atlas library related to the goods to be identified includes: Obtain sample images containing various preset quality defects, classify and annotate each defect, and form a defect type map; Selecting samples of goods that meet quality standards from warehouses or logistics centers, taking images of qualified products at different angles and in different states, and classifying and marking the images of qualified products to generate a qualified product map; Take images of the same product under different lighting conditions to ensure that all lighting environments encountered by the product during the actual production process are covered, and classify the images of the same product according to the lighting conditions to form a reference map under different lighting conditions; The defect type map, the qualified product map and the reference map are integrated into a multi-dimensional quality standard map library.

3. The method according to claim 1, characterized in that If it is determined that there is an area in the real-time video stream that matches the preset quality defect, a response strategy for the defect type corresponding to the area is automatically triggered, including: Generate an alarm message containing detailed information of the defect according to the specific location of the area and the defect type corresponding to the area; Based on the alarm information, a matching response strategy is selected according to a preset rule base, and after the response strategy is implemented, defect detection is re-performed using an adaptive comparison algorithm to verify whether the occurrence of defects has been reduced or eliminated; wherein the preset rule base stores response strategies corresponding to different defect types.

4. A cargo quality monitoring and analysis system based on image recognition, used to execute the cargo quality monitoring and analysis method based on image recognition as claimed in any one of claims 1 to 3, characterized in that: include: An acquisition module, used to acquire a real-time video stream of the goods to be identified; A construction module, used to construct a multi-dimensional quality standard atlas library related to the goods to be identified, wherein the multi-dimensional quality standard atlas library includes defect type atlas, qualified product atlas and reference atlas under different lighting conditions; A matching module, used to match the real-time video stream with the atlas in the multi-dimensional quality standard atlas library through an adaptive comparison algorithm to determine whether there is an area in the real-time video stream that matches the preset quality defect, and generate a matching result; A trigger module, configured to automatically trigger a response strategy for the defect type corresponding to the area if it is determined that there is an area matching a preset quality defect in the real-time video stream, wherein the response strategy includes generating alarm information; An updating module is used to dynamically update the multi-dimensional quality standard atlas library according to the matching results and the execution of the response strategy to optimize the monitoring accuracy of the goods quality.

5. A computing device, characterized in that: It comprises a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement a cargo quality monitoring and analysis method and system based on image recognition as described in any one of claims 1 to 3.

6. A computer storage medium, characterized in that: A computer program is stored, and when the computer program is executed by a computer, a cargo quality monitoring and analysis method and system based on image recognition as described in any one of claims 1 to 3 is implemented.

Images