Port customs management and control method and system based on artificial intelligence

Abstract

The embodiment of the invention provides a port customs management and control method and system based on artificial intelligence. The method comprises the steps that key feature information is extracted from container cargo list data received by a port; shooting actual goods by using an image recognition algorithm and generating image feature data; performing automatic matching analysis based on the extracted key feature information and the image feature data; and according to a matching result, determining whether there is a bill inconsistency condition and outputting a checking conclusion. Through the scheme of the embodiment of the invention, the problem of how to automatically match the actual cargo image with the container cargo list data so as to solve the problem of high checking difficulty caused by the inconsistency of the cargo list can be solved.

Description

Technical Field

[0001] The present application relates to the field of smart port technology, and specifically to a port customs control method and system based on artificial intelligence. Background Art

[0002] Port customs control aims to improve the management and monitoring efficiency of port customs by introducing advanced intelligent technologies such as machine learning, image recognition and big data analysis. It processes massive import and export cargo information through automated data analysis and conducts rapid and accurate risk assessment of high-risk cargo flows, thereby improving inspection efficiency and security. However, this approach still faces some challenges, especially how to automatically match container cargo manifest data with actual cargo images to solve the problem of high inspection difficulty caused by inconsistent cargo manifests. Traditional manual comparison methods are not only time-consuming and labor-intensive, but also prone to errors. Summary of the invention

[0003] In view of this, the embodiments of the present disclosure provide a port customs control method based on artificial intelligence, which at least partially solves the problems existing in the prior art.

[0004] First, a port customs control method based on artificial intelligence is provided, including:

[0005] Extract key feature information from container cargo manifest data received at the port;

[0006] Use image recognition algorithms to photograph actual goods and generate image feature data;

[0007] Automatically match and analyze the extracted key feature information with the image feature data;

[0008] Determine whether there is any discrepancy between the cargo and the invoice based on the matching results and output the inspection conclusion.

[0009] In a specific embodiment, the automatic matching analysis based on the extracted key feature information and image feature data includes:

[0010] The extracted key feature information is normalized to eliminate the influence of different feature dimensions;

[0011] Calculate the similarity between the image feature data and the normalized key feature information;

[0012] Based on the similarity, determine whether a preset matching threshold condition is met. If so, it is considered that the key feature information and the image feature data are matched successfully, otherwise the match fails;

[0013] Mark the successfully matched data for subsequent verification of the conclusion output.

[0014] In a specific implementation, judging whether a preset matching threshold condition is met based on the similarity includes:

[0015] Get the preset matching threshold, which is dynamically adjusted based on historical matching data and cargo type;

[0016] Based on the comparison between the similarity and a preset matching threshold, if the similarity is greater than or equal to the threshold, the matching condition is met;

[0017] If the similarity is less than the threshold, further analyze whether there is feature deviation caused by abnormal feature points in the image feature data;

[0018] Based on the analysis results of abnormal feature points, determine whether it is necessary to temporarily adjust the matching threshold to adapt to the special circumstances of the current goods.

[0019] In a specific implementation, the determining whether it is necessary to temporarily adjust the matching threshold based on the analysis result of the abnormal feature point includes:

[0020] Count the number and distribution of abnormal feature points;

[0021] Based on the number and distribution of abnormal feature points, it is determined whether it falls within the acceptable deviation range. The acceptable deviation range formula is:

[0022] Acceptable deviation range = [average number of abnormal feature points − k × standard deviation, average number of abnormal feature points + k × standard deviation], where k is the confidence coefficient;

[0023] If it is within the acceptable deviation range, the matching threshold is not adjusted and the matching is directly judged as failed;

[0024] If the deviation exceeds the acceptable range, the matching threshold is temporarily lowered according to the influence of the abnormal feature points. The reduction range is:

[0025] Downward adjustment range = α×(number of abnormal feature points / total number of feature points), where α is the adjustment coefficient.

[0026] In a specific implementation, temporarily lowering the matching threshold according to the influence of the abnormal feature point includes:

[0027] Determine the impact level of abnormal feature points and classify them according to their importance in cargo identification;

[0028] Based on the impact level, find the corresponding downward adjustment coefficient table to obtain the corresponding downward adjustment coefficient;

[0029] The actual downward adjustment is calculated as follows:

[0030] Actual downward adjustment = downward adjustment coefficient × basic downward adjustment, where the basic downward adjustment is the pre-set minimum downward adjustment value;

[0031] Based on the calculated actual downward adjustment range, the matching threshold is adjusted and the adjusted threshold is recorded.

[0032] In a specific implementation, searching a corresponding downward adjustment coefficient table based on the impact level to obtain the corresponding downward adjustment coefficient includes:

[0033] Constructing a downward adjustment coefficient table, wherein the downward adjustment coefficient table maps the impact level to the downward adjustment coefficient based on historical data and expert experience;

[0034] Input the influence level of the current abnormal feature point, and search for the corresponding downward adjustment coefficient based on the downward adjustment coefficient table;

[0035] If the current impact level is not directly matched in the table, the downward adjustment coefficient of the current level is calculated by interpolation based on the downward adjustment coefficients of the adjacent levels;

[0036] Based on the obtained downward adjustment coefficient, a subsequent matching threshold adjustment operation is performed.

[0037] In a specific implementation, if the current impact level is not directly corresponding in the table, the downward adjustment coefficient of the current level is calculated by interpolation based on the downward adjustment coefficients of the adjacent levels, including:

[0038] Determine two known levels adjacent to the current impact level and the corresponding downward adjustment coefficients;

[0039] Based on the downward adjustment coefficients of adjacent levels, the change rate of the downward adjustment coefficients between adjacent levels is calculated;

[0040] Based on the change rate, calculating the downward adjustment coefficient increment of the current level;

[0041] The increment is added to the downward adjustment coefficient of the lower level of the two known levels to obtain the downward adjustment coefficient of the current level.

[0042] In a specific implementation, the step of adding the increment to the downward adjustment coefficient of the lower level of the two known levels to obtain the downward adjustment coefficient of the current level includes:

[0043] Obtaining the downward adjustment coefficient of the lower level of the two known levels as the base value;

[0044] Add the calculated increment to the base value to get the downward adjustment coefficient of the current level;

[0045] The rationality of the calculated downward adjustment coefficient of the current level is checked. If the current downward adjustment coefficient is between 0 and 1, it is reasonable, otherwise it is unreasonable;

[0046] If the test result is reasonable, the downward adjustment coefficient is used to adjust the matching threshold; if it is unreasonable, the selection or interpolation method of adjacent levels is readjusted to ensure the rationality of the downward adjustment coefficient.

[0047] In a specific implementation, if the test result is reasonable, the matching threshold is adjusted using the downward adjustment coefficient; if it is unreasonable, the selection or interpolation method of the adjacent levels is readjusted, including:

[0048] If the test result is reasonable, the downward adjustment coefficient of the current level is directly applied to the matching threshold adjustment formula to adjust the matching threshold;

[0049] If the test result is unreasonable, analyze the reason for the unreasonableness, whether it is due to improper selection of adjacent levels or inappropriate interpolation method;

[0050] Based on the analysis results, if the adjacent level is not properly selected, reselect other adjacent levels; if the interpolation method is not applicable, try other interpolation methods, such as spline interpolation;

[0051] Based on the reselected adjacent levels or the new interpolation method, the downward adjustment coefficient of the current level is recalculated and the rationality check is performed again until a reasonable downward adjustment coefficient is obtained.

[0052] Secondly, an artificial intelligence-based port customs control system is provided, including:

[0053] A key feature information extraction module extracts key feature information from the container cargo manifest data received at the port;

[0054] An image feature data generation module uses an image recognition algorithm to photograph actual goods and generate image feature data;

[0055] Matching analysis module, which performs automatic matching analysis based on the extracted key feature information and image feature data;

[0056] The judgment module determines whether there is any discrepancy between the goods and the invoice based on the matching results and outputs the inspection conclusion.

[0057] The disclosed embodiment provides a port customs control method based on artificial intelligence, including: extracting key feature information from the container cargo manifest data received at the port; photographing the actual cargo using an image recognition algorithm and generating image feature data; automatically matching and analyzing the extracted key feature information with the image feature data; determining whether there is a cargo manifest discrepancy based on the matching result and outputting an inspection conclusion. Through the solution of the disclosed embodiment, it is possible to solve the problem of how to automatically match the container cargo manifest data with the actual cargo image to solve the high inspection difficulty caused by the cargo manifest discrepancy. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] In order to more clearly illustrate the technical solutions of the exemplary implementation methods of the embodiments of the present disclosure, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the embodiments of the present disclosure and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0059] Figure 1 It is a flow chart of a port customs control method based on artificial intelligence;

[0060] Figure 2 It is a flow chart for automatic matching analysis based on the extracted key feature information and image feature data;

[0061] Figure 3 It is a flow chart for judging whether a preset matching threshold condition is met based on similarity;

[0062] Figure 4 It is a flow chart for determining whether a temporary adjustment of the matching threshold is required based on the analysis results of abnormal feature points;

[0063] Figure 5 It is a flow chart for temporarily lowering the matching threshold according to the influence of abnormal feature points;

[0064] Figure 6 Based on the impact level, the corresponding downward adjustment coefficient table is searched to obtain the corresponding downward adjustment coefficient flow chart;

[0065] Figure 7 It is a flowchart of calculating the downward adjustment coefficient of the current level by interpolation method based on the downward adjustment coefficients of the adjacent levels if the current impact level is not directly corresponding in the table;

[0066] Figure 8 A flow chart of adding the increment to the downward adjustment coefficient of the lower level of the two known levels to obtain the downward adjustment coefficient of the current level;

[0067] Fig. 9 If the test result is reasonable, the matching threshold is adjusted using the downward adjustment coefficient; if it is unreasonable, the flow chart of the selection or interpolation method of the adjacent levels is readjusted;

[0068] Fig.10 It is a flowchart for recalculating the downward adjustment coefficient of the current level based on the reselected adjacent level or new interpolation method and performing rationality check again. DETAILED DESCRIPTION

[0069] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0070] Next, refer to Figure 1 , describing the steps of an artificial intelligence-based port customs control method of the present invention. The method aims to improve the efficiency of cargo inspection, reduce the risk caused by cargo and bill discrepancies, and ultimately enhance the security and reliability of the entire customs clearance process.

[0071] S101: Extract key feature information from the container cargo manifest data received at the port. Specifically, when the goods enter the port, they will carry detailed commodity description information - these paper or electronic lists cover the basic attributes of each independent packaging unit, such as product category, quantity specifications and even brand batch and other identification elements. The above structured document content is screened and summarized through a preset text parsing algorithm to obtain a representative label vector for subsequent data processing and analysis. For example, in this embodiment, natural language processing technology in deep learning can be used to identify and classify different types of goods to obtain more accurate feature information. These features are not limited to text descriptions but also include corresponding numerical values ​​or codes to make the meaning of each entry clear and easy to compare and calculate.

[0072] In a specific embodiment, the container cargo manifest data may first be preprocessed to remove noise and redundant information in the data; then, based on the preprocessed data, a feature extraction algorithm, such as a principal component analysis (PCA) algorithm, may be used to extract key feature information; then, the integrity of the extracted key feature information may be calculated, and the integrity calculation formula is: integrity = number of extracted key features / preset total number of key features × 100%; then, based on the integrity, it is determined whether a preset integrity threshold is met, and if so, the extracted key feature information is considered complete, otherwise, feature extraction is performed again to ensure the accuracy and integrity of the key feature information.

[0073] S102: Use image recognition algorithms to photograph actual goods and generate image feature data. Specifically, camera devices can be deployed at strategic points in the terminal to cover the entire loading and unloading operation to obtain the most realistic on-site perspective image resources. These high-definition images are instantly transmitted to the cloud center via a specific computer program to wait for further processing and transformation to become a knowledge carrier that can be interpreted and used by the machine. In order to ensure matching accuracy, strict standardized measures are taken during the shooting stage, such as using multi-angle continuous shooting under fixed light conditions to ensure that the clarity of the obtained material meets the established requirements to avoid misjudgment due to external environmental factors; at the same time, advanced target tracking and grasping methods are used to accurately capture the state of moving objects in order to fully record every change.

[0074] In a specific embodiment, the actual goods may be photographed at multiple angles to obtain multiple goods images; then the obtained goods images may be preprocessed, including image denoising, enhancement and other operations; then based on the preprocessed images, an image recognition algorithm, such as a convolutional neural network (CNN) algorithm, may be used to extract image feature data; finally, the clarity of the extracted image feature data may be calculated, and the clarity calculation formula is: , where N is the dimension of the image feature data, is the maximum value of the image feature data; finally, based on the clarity, it is determined whether the preset clarity threshold is met. If so, the generated image feature data is considered valid, otherwise, the image is re-shot and feature extraction is performed to ensure the quality of the image feature data.

[0075] S103: Automatic matching analysis based on the extracted key feature information and image feature data. The system relies on a large-scale sample learning model to perform this complex task, that is, to build a mapping relationship between the two dimensional features to achieve a fast and accurate map-like registration and positioning function - every time a photo is uploaded, the algorithm will quickly search for matching commodity items based on its internal embedded attributes, and then assess the similarity score range between the two to determine whether there is a consistency relationship between them. In one example, if the declared item is a mechanical part, the image database should be able to find items with typical structural graphics such as bolts and bearings for comparative evaluation. This approach greatly shortens the time required for verification, improves work efficiency, and reduces the probability of omissions that may occur due to manual work.

[0076] S104: Determine whether there is any discrepancy between the cargo list and the inspection conclusion based on the matching results. When an abnormal signal warning is found, the customs staff will be notified immediately to start a more detailed and thorough manual review and verification to ensure that there is no possible risk of any illegal or smuggled goods being mixed in. Under normal circumstances, normal release instructions will also be issued on time to maintain the normal rhythm of inbound and outbound logistics, ensuring that the port transportation order operates well and is not affected.

[0077] This method effectively solves the problem of how to compare and verify the actual cargo image in the container with the data information in its customs declaration documents, effectively reduces the burden of cumbersome procedures caused by unnecessary physical inspections, greatly saves time and energy cost investment, and optimizes the entire import and export business chain.

[0078] Next, refer to Figure 2 , describing the specific steps of the present invention for automatic matching analysis based on the extracted key feature information and image feature data.

[0079] S201: Normalize the extracted key feature information to eliminate the impact of different feature dimensions. This step aims to eliminate the impact of different feature dimensions on the analysis results. For example, in the port customs scenario, a variety of feature information (such as shape, size, color, etc.) is extracted from the input container cargo list data. These features have different dimensions and numerical ranges. If they are not unified, errors will be introduced into subsequent calculations. During the normalization process, all key features will be mapped to the [0,1] interval, and the optimal parameters are the boundary values ​​of this interval. Through the formula, This ensures that each feature is converted into a representation of the same scale.

[0080] S202: Calculate the similarity between the image feature data and the normalized key feature information. Specifically, the distance measurement method or cosine similarity is used to evaluate the closeness between the two, which reflects how close the corresponding relationship of the feature information is to the actual situation. In one embodiment, suppose that a batch of auto parts unloaded from a cargo ship needs to be inspected, and the system has acquired and learned the ideal shape and identification of these parts in advance. At this time, the similarity between various visual representations in the cargo picture to be inspected and the normalized key feature information is calculated.

[0081] In one embodiment, for example, the similarity between the image feature data and the normalized key feature information can be calculated using the following formula: , where n is the feature dimension.

[0082] S203: Based on the similarity, determine whether the preset matching threshold condition is met. If so, it is considered that the key feature information and the image feature data are successfully matched, otherwise the match fails. This step is used to ensure that the matching accuracy is sufficient to meet the needs of the application scenario. If the similarity is higher than a certain threshold (for example, 85%-95%, the specific value can be determined according to the port control requirements), it is considered a successful match, which means that the current detection object is highly consistent with the known pattern; on the contrary, if it is lower than the threshold, it is determined to be a mismatch. For the automotive parts in the above example, when the similarity score exceeds the predetermined limit, such as more than 93%, it is considered a match; on the contrary, if the score is low, there may be risks and further verification of authenticity is required.

[0083] S204: Mark the successfully matched data for the subsequent inspection conclusion output. This marking is convenient for subsequent personnel to view and quickly make corresponding decisions. For example, in the above-mentioned AI-based port inspection procedure, once a certain imported item is determined to be safe and compliant, a confirmation mark is immediately given to facilitate the smooth flow of other links in the logistics chain and also to generate a complete cargo passage log file for tracking and supervision purposes. Such a series of rigorous operations ensures the automation of the entire process while greatly improving management efficiency and accuracy.

[0084] Next, refer to Figure 3 , describing the steps of the present invention for determining whether a preset matching threshold condition is met based on similarity.

[0085] S301: Obtain a preset matching threshold, which is dynamically adjusted according to historical matching data and cargo type. Specifically, the preset matching threshold is not fixed, but is derived based on previously accumulated matching cases and data on different types of cargo. For example, in port customs control, if the similarity between cargo images and cargo manifest data involving electronic products in the past is mostly 0.8 to 0.9, then the preset matching threshold for this category of cargo may be set to a number slightly lower than the average value of the range (such as 0.75). This is intended to ensure that new samples pass when they are more likely to be true examples, rather than treating all inaccurate matches as potential risks.

[0086] S302: Based on the similarity, the preset matching threshold is compared. If the similarity is greater than or equal to the threshold, the matching condition is met. Once the calculation result shows that the similarity between the two is greater than or equal to the pre-given limit, the goods in this inspection process are considered to meet the standards. For example, in one embodiment, for the inspection of textiles in a container, the appearance characteristics and material composition obtained by the system calculation are highly consistent with the expected model, and the similarity reaches or exceeds 0.9 (assuming that the above steps have obtained the preset threshold for fabric materials as 0.8), then it is directly recognized. This process effectively improves the customs clearance efficiency under normal trade activities and reduces the probability of misjudgment.

[0087] S303: If the similarity is less than the threshold, further analyze whether there is feature deviation caused by abnormal feature points in the image feature data. Taking the observation of the shape structure of the goods under X-ray scanning as an example, the discovery of feature deviation phenomena such as obvious bulges or other morphological changes in a certain place is a suspected cause, which is not necessarily a dangerous factor. In another embodiment, feature deviation may also be caused by damaged packaging of the goods. Such problems are sometimes caused by changes in perspective or unique packaging methods, and do not mean that the goods themselves are suspicious.

[0088] In a specific embodiment, when the similarity value is less than a threshold, the matching process is suspended, and the image feature data is deeply analyzed; key feature points in the image feature data are identified, such as shape contour feature points, color distribution feature points, texture feature points, etc. of the goods; key feature points are checked for abnormalities, such as whether the shape contour feature points are distorted, whether the color distribution feature points have color deviation, whether the texture feature points have texture blur, etc.; if abnormal feature points are found, the specific location, type and degree of the abnormal feature points are recorded, and the possible causes of the abnormal feature points are analyzed, such as damaged packaging of the goods, discoloration of the goods due to moisture, failure of the image acquisition device, etc. It is understandable that further analysis of whether there is a feature deviation caused by abnormal feature points in the image feature data can be obtained by comparing with the images of the abnormal feature points stored in the database.

[0089] S304: Based on the analysis results of abnormal feature points, determine whether it is necessary to temporarily adjust the matching threshold to adapt to the special situation of the current goods. If there are visual differences in the container due to special packing as mentioned above, the matching requirements can be appropriately relaxed after evaluation. For example, the preset parameters (such as 0.8) previously determined for the ordinary fabric recognition process can be reduced to a position closer to the current actual situation but not too loose, such as around 0.7, as a new inspection basis to ensure the best balance between safety and accuracy.

[0090] In one embodiment, whether the matching threshold needs to be adjusted can be evaluated based on the degree of abnormality of the abnormal feature points and the impact on the matching results; if the abnormal feature points have a small impact on the matching results and there is a reasonable explanation, such as slight packaging damage of the goods but it does not affect the essential properties of the goods, the original matching threshold can be maintained unchanged and a special situation description can be recorded; if the abnormal feature points have a large impact on the matching results, such as the goods are damp, resulting in significant changes in key features such as color and texture, and cannot be accurately identified by conventional matching algorithms, the matching threshold is temporarily lowered based on the impact of the abnormal feature points, and the reduction range is determined based on similar historical situations and customs risk assessment; the matching results after the temporary adjustment of the matching threshold are compared with the results of the customs manual review. If the two are consistent, it is confirmed that the temporarily adjusted matching threshold is valid. If not, the cause is further analyzed to optimize the matching algorithm and threshold adjustment strategy.

[0091] Next, refer to Figure 4 , describing the steps of determining whether a temporary adjustment of a matching threshold is required based on the analysis results of abnormal feature points of the present invention.

[0092] S401: Count the number and distribution of abnormal feature points. This step aims to identify the recorded data points that do not conform to the conventional characteristics, and quantify and group these abnormal points and analyze their regional locations. For example, in a port customs control method based on artificial intelligence, image comparison of goods is a common scenario. Suppose that the scan results of the goods in a container show several inconsistencies with the pre-trained model, which may be caused by loading differences. Therefore, this step is to count and analyze the location of all such data. In another embodiment, the image data can first be pre-processed by an image recognition algorithm to extract the feature points therein, and then the extracted feature points can be classified to distinguish between normal feature points and abnormal feature points, and finally the number of abnormal feature points in different location areas can be calculated.

[0093] S402: Based on the number and distribution of abnormal feature points, determine whether it is within an acceptable deviation range. The acceptable deviation range formula is:

[0094] Acceptable deviation range = [average number of abnormal feature points − k × standard deviation, average number of abnormal feature points + k × standard deviation]

[0095] Among them, k is the confidence coefficient, the average number of abnormal feature points refers to the average of abnormal points found within the normal range accumulated when scanning the same type of objects multiple times in the past; the standard deviation represents the degree of variability between different measurements; k is a confidence coefficient, usually between 1 and 3, used to adjust the acceptance range, and the effect is better when it is close to 2 because it can ensure that more than 95% of the samples fall within this range; this setting is to reflect the random fluctuations of the real world as accurately as possible. If the distribution of abnormal points falls within this range, it means that the system operation is still within expectations.

[0096] S403: If it is within the acceptable deviation range, the matching threshold is not adjusted and the matching is directly determined to be failed. For example, if a container is found to have an abnormality within the aforementioned acceptable range on a certain day, this means that the abnormality may be caused by a reasonable error and does not represent a real risk. Therefore, in an example embodiment, customs staff can directly follow the normal process without further modifying the rules.

[0097] S404: If the deviation exceeds the acceptable range, the matching threshold is temporarily lowered according to the influence of the abnormal feature point, and the reduction range is:

[0098] Downward adjustment range = α × (number of abnormal feature points / total number of feature points).

[0099] Where α is the adjustment coefficient. α should be between 0 and 1, and ideally close to but not exceeding 0.7. The specific choice depends on the balance between the desired tolerance after adjustment and the desired rigor. This adjustment method is intended to prevent potential problems from being completely ignored due to interference from some extreme abnormal events, while at the same time not causing the alert line to be too tight to affect the smooth operation of the business. For example, in the case of encountering new unknown materials or abnormal placement of items resulting in a large number of feature points deviating from the normal level, fine-tuning can be used to ensure that subsequent shipments of the same type can be successfully checked without excessive manual intervention in calibration parameters.

[0100] Next, refer to Figure 5 , describing the specific steps of temporarily lowering the matching threshold according to the influence degree of abnormal feature points of the present invention.

[0101] S501: Determine the impact level of abnormal feature points and classify them according to their importance in cargo identification. Specifically, by analyzing historical data to analyze the possible impact of each abnormality on customs clearance security risks or logistics efficiency, the corresponding level is assigned to each feature to ensure that key issues are given priority. For example, in the port customs monitoring scenario, the presence of undeclared large amounts of cash in the container is a high-risk factor, and this type of abnormality is classified as the most serious impact level; while relatively minor issues such as packaging damage are at a lower level.

[0102] S502: Based on the impact level, find the corresponding downward adjustment coefficient table and obtain the corresponding downward adjustment coefficient. This step involves comparing the pre-defined standards and tables to find the specific adjustment parameter values ​​that meet the corresponding positions of the evaluated values. In an artificial intelligence system, suppose there is a coefficient table that stipulates different levels of violations. For example, the downward adjustment corresponding to the above-mentioned major financial violations may be 0, which means that this situation should be checked more strictly or handled flexibly than the normal situation.

[0103] S503: Calculate the actual downward adjustment, the formula is: actual downward adjustment = downward adjustment coefficient × basic downward adjustment, where the basic downward adjustment is the preset minimum downward adjustment value. In this expression, the downward adjustment coefficient is a dynamically adjusted variable, the range depends on the specific implementation rules and environmental requirements, and the optimal value setting needs to comprehensively consider the balance between detection accuracy and timeliness; the basic downward adjustment represents the minimum change range under the default conditions, which is generally set to a value that can cause appropriate sensitivity but will not cause excessive fluctuations. Such a setting allows the algorithm to capture truly noteworthy change points while maintaining a certain sensitivity, without causing the system to be too aggressive and produce unnecessary error rate increases. In an actual application scenario, when faced with high-risk situations such as suspected possession of hazardous substances (assuming that the impact is serious), the coefficient selection will be more conservative. If the basic value is set to 5%, the final matching condition may be reduced from 80% to about 74%, which can allow some samples close to the limit to enter a more detailed manual review link.

[0104] S504: Based on the calculated actual downward adjustment, the matching threshold is adjusted and the adjusted threshold is recorded. This adjustment is intended to enable the automatic identification system's evaluation criteria to respond to emerging uncertainties and risk factors in a timely manner, and retaining logs will help future audit tracking and experience summary. Taking the port as an example, if the initial criteria for judging whether the cargo identification is qualified or not is set to a similarity of 90%, but due to the damage of the cargo packaging, the identified similarity is 88%. Through searching, it is found that the abnormal feature is the damaged packaging. According to the level of impact, the threshold can be lowered by 5%, that is, 85%. This means that even if there is a slight difference between the images, it can be further reviewed, which increases the probability of eliminating safety hazards while avoiding excessive obstruction to normal cargo transportation.

[0105] Next, refer to Figure 6 , describing the specific steps of the present invention for searching the corresponding downward adjustment coefficient table based on the impact level and obtaining the corresponding downward adjustment coefficient.

[0106] S601: Construct a downward adjustment coefficient table, which maps the impact level to the downward adjustment coefficient based on historical data and expert experience. The specific operation is to formulate a mapping table based on a large number of past abnormal records and opinions of professionals in related fields. The impact of each level is associated with a corresponding value to indicate how much the risk needs to be reduced or the threshold needs to be adjusted. For example, in one embodiment, various types of threat behaviors and severity levels encountered in the port container inspection process are quantified and compiled into a table.

[0107] S602: Input the impact level of the current abnormal feature point, and search for the corresponding downward adjustment coefficient based on the downward adjustment coefficient table. For each determined level value, it is only necessary to quickly retrieve the preset proportional factor in the table according to the set rules. In a specific example, if a certain item in a cargo inspection is determined to be a level 2 threat (a total of five levels from 1 to 5 with increasing danger), the system directly refers to the past probability of occurrence of such events and their consequences to decide which existing k=0.7 should be used as its downward adjustment parameter.

[0108] S603: If the current impact level is not directly corresponding in the table, the downward adjustment coefficient of the current level is calculated by interpolation based on the downward adjustment coefficient of the adjacent level. If no exact match is found in the above query process, the difference between the two adjacent levels is used for interpolation estimation to obtain the required proportion. For example, if the impact intensity score in a certain situation is 3.8 points but there are only integer level regulations in the table, a more accurate result Knew = a × Klower + b × Klupper can be linearly calculated through the data points of the adjacent levels. In this formula, a and b represent distance weight factors, which refer to the proportion closest to the lower and higher ends of the target level respectively; the range is limited to [0, 1], and the condition a + b = 1 is satisfied to maintain the consistency and continuity of the results.

[0109] S604: Based on the obtained downward adjustment coefficient, subsequent matching threshold adjustment operations are performed. For example, when the artificial intelligence model analyzes cargo information, if a risk signal of suspected prohibited items is encountered, and the evaluation result of the signal indicates that there is a high risk, then the downward adjustment can be set to 0, or even the matching threshold can be increased, so that the customs control system can respond accordingly to ensure the safety and efficiency of port transportation activities.

[0110] Next, refer to Figure 7 , describing the steps of the present invention of calculating the downward adjustment coefficient of the current level by interpolation method based on the downward adjustment coefficients of adjacent levels if the current impact level is not directly corresponding in the table.

[0111] S701: Determine two known levels adjacent to the current impact level and the corresponding downward adjustment coefficients. This step ensures that the nearest neighbor data point is found within the existing data range. Specifically, assuming that the current impact level is within a certain interval but there is no directly corresponding value, select data one level higher (recorded as level 1) and one level lower (recorded as level 2) than the current level for comparative analysis. For example, if the impact level is above medium, medium and high downward adjustment coefficients are selected for calculation.

[0112] S702: Based on the downward adjustment coefficients of adjacent levels, calculate the rate of change of the downward adjustment coefficients between adjacent levels. This rate of change refers to the change of the downward adjustment coefficient as the level increases. It is expressed as ΔC = (C1 -C2) / (L1 - L2), where ΔC represents the rate of change, C1 is the downward adjustment coefficient of the higher level, C2 is the downward adjustment coefficient of the lower level; and L1 and L2 are the identification numbers of the two levels respectively. The best choice should be a fixed ratio value that makes ΔC as stable as possible and accurately reflects the relationship between the downward adjustment amplitudes of two adjacent levels.

[0113] S703: Based on the change rate, calculate the downward adjustment coefficient increment of the current level. Then, based on this change rate, calculate the downward adjustment coefficient increment Δx = ΔC × (Lv - L2) of the current impact level relative to the lower level. Here, Lv represents the impact level number of the current assessment, and Lv should be within the closed interval [L2, L1]. The purpose of this formula is to quantify the additional adjustment caused by increasing Lv relative to the lower bound (L2).

[0114] S704: Add the increment to the downward adjustment coefficient of the lower level of the two known levels to obtain the downward adjustment coefficient of the current level. Finally, add the increment calculated above to the downward adjustment coefficient of the bottom base level to determine the final downward adjustment coefficient: Final_C = C2 + Δx. This approach ensures that reasonably accurate results can be obtained for non-directly corresponding impact levels, and maintains the consistency and reliability of the algorithm.

[0115] Through such a detailed and systematic approach, the control strategies for cargo transportation and related operations can be adjusted more flexibly and effectively during the joint prevention and control process of port customs, thereby improving the safety and accuracy of the overall operation.

[0116] Next, refer to Figure 8 , describing the step of adding the increment to the downward adjustment coefficient of the lower level of the two known levels to obtain the downward adjustment coefficient of the current level.

[0117] S801: Obtain the downward adjustment coefficient of the lower level of the two known levels as the base value. In this step, the system determines the downward adjustment coefficient of the lower level as the starting point. The downward adjustment coefficient is used to determine the degree of reduction of a certain cargo risk after risk assessment. The value range is usually between 0 and 1. The optimal value is close to the middle of the range, indicating a risk attitude that is both moderately conservative and flexible.

[0118] S802: Add the calculated increment to the base value to obtain the downward adjustment coefficient of the current level. On this basis, add the calculated increment to this base value to calculate the downward adjustment coefficient of the current higher level. The increment is determined based on the actual environment of the port customs and specific situation analysis. For example, the value adjusted based on the risk assessment of import and export goods based on artificial intelligence algorithms. The increment may be positive or negative. This method can quantify the difference between the levels.

[0119] S803: Check the rationality of the calculated downward adjustment coefficient of the current level. If the current downward adjustment coefficient is between 0 and 1, it is reasonable, otherwise it is unreasonable. Next, check the rationality of the calculated downward adjustment coefficient of the current level. If the value falls within the interval [0, 1], it is considered valid and reasonable; otherwise, it is illegal and measures need to be taken to correct it. The establishment of this standard is based on the actual application requirements of the downward adjustment coefficient - it cannot exceed the maximum limit (set to 1) to avoid unnecessary excessive reduction in safety precautions, nor can it be less than the minimum limit (ie 0) to prevent invalid risk downgrade measures.

[0120] For example, in one embodiment, if the initial downward adjustment coefficient corresponding to low-risk goods is set to 0.3, and an increment of +0.25 is calculated based on a certain evaluation result, then the downward adjustment coefficient for the newly obtained higher-level goods is 0.55. At this time, according to the above test conditions, 0.55 is indeed within the allowable range, indicating that this downward adjustment factor is reasonable and applicable.

[0121] S804: If the inspection result is reasonable, the matching threshold is adjusted using the downward adjustment coefficient; if it is unreasonable, the selection or interpolation method of the adjacent level is readjusted to ensure the rationality of the downward adjustment coefficient. If the result obtained through the above process does not meet the requirements, specifically, after the calculation, it is found that the downward adjustment factor is not within the specified range, it is necessary to consider reselecting the adjacent level, or adjusting the method originally selected to generate the difference between these values ​​to redefine the interval between the two levels. This ensures that any level set in any case is within the appropriate limit, which neither affects the normal flow speed of goods nor endangers the overall safety. Specifically, in one embodiment, the adjacent level can be selected to minimize the downward adjustment to prevent the regulation from being too loose. For example, if the current level is 2.4, if the difference is based on level 2 and level 3, the downward adjustment is 20%, and if it is based on level 2 and level 4, the reduction is 8% (the smallest downward adjustment) and the coefficient after the reduction is between 0 and 1, then the adjacent level 2 and the adjacent level 4 can be selected as the adjacent level. As for the interpolation method, you can choose, for example, Lagrange interpolation, Newton interpolation, linear interpolation, etc.

[0122] Finally, when a reasonable current level reduction factor is obtained, the corresponding threshold can be set accordingly. This can ensure that each processing flow can correctly distinguish and treat items of different threat levels, and realize the effective operation of the intelligent and efficient port customs management mechanism.

[0123] Next, refer to Fig. 9, describing the steps of the present invention, if the test result is reasonable, then using the downward adjustment amplitude coefficient to adjust the matching threshold; if it is unreasonable, then readjusting the adjacent level selection or interpolation method.

[0124] S901: If the inspection result shows that it is reasonable, the system will directly apply the current level of downward adjustment coefficient to the matching threshold adjustment formula. For example, suppose that in a certain cargo risk assessment model, the matching threshold T is used to determine whether further inspection of the cargo is required, and the initial matching threshold is set through historical data analysis and model training. When the system's verification mechanism confirms that the downward adjustment coefficient a is reasonable, the system will use the newly adjusted threshold calculation formula T = T - ΔT, where ΔT represents the change calculated by the difference between the downward adjustment coefficient a and the original threshold. This design is intended to improve matching efficiency, optimize the inspection process, reduce false alarm rates, and ensure that high-risk items can be properly inspected.

[0125] S902: If the test result is unreasonable, analyze the reason for the unreasonableness, whether it is due to improper selection of adjacent levels or inappropriate interpolation method. The reason may be that the distance between two different risk levels is too small due to wrong selection or the interpolation method is not suitable, thus affecting the final judgment result. In this case, instead of immediately using the existing downward adjustment coefficient for calculation, it is necessary to deeply analyze the existing data and parameter selection to find out the problem so as to adjust the algorithm performance more finely.

[0126] S903: Based on the analysis results, if the adjacent level is not properly selected, reselect other adjacent levels; if the interpolation method is not applicable, try other interpolation methods, such as spline interpolation, etc. For problems caused by improper selection of adjacent levels, for example, in the customs risk control strategy, it is found that the classification standards for high-risk goods may be too general or insufficiently detailed, which makes some low-risk goods often regarded as key monitoring objects. At this time, the relevant classification standards will be revised according to the new situation, and a data set that is closer to the characteristics of the target group will be selected as a reference point to build a more scientific risk assessment system, and then redefine the risk level relationship suitable for these goods.

[0127] If the interpolation method is not suitable, other alternatives will be explored, such as spline interpolation, which can better adapt to those characteristics that fluctuate greatly over time or are nonlinear. Other advanced algorithms such as spline interpolation can establish smooth transition curves between different stages, effectively filling the data gaps caused by the lack of actual measurement points, and making the model more adaptable to the changing trends of the diverse logistics environment while ensuring stability and accuracy. In one embodiment, for some special types of goods, the original simple interpolation method may lead to increased estimation errors. After the introduction of cubic spline interpolation with moderate complexity, it has been improved, which can improve the prediction accuracy while keeping the computing resource overhead low.

[0128] S904: Based on the reselected adjacent levels or new interpolation methods, recalculate the downward adjustment coefficient of the current level, and conduct a rationality check again until a reasonable downward adjustment coefficient is obtained. After determining the new selection or interpolation method, the downward adjustment coefficient of the current level will be calculated again, and the verification process will be repeated many times until the most reasonable solution is found. Specifically, it is to use the new adjacent level combination formed after the re-selection and the updated processing mode to regenerate a set of new parameters under the corresponding conditions, and conduct comprehensive testing and review based on a strict rule system to prove the correctness and reliability of the setting, so as to ensure the smooth and efficient operation of the entire artificial intelligence-based port customs control method.

[0129] For example, in this scenario: every time the system automatically adjusts the matching threshold based on the historical data of cargo transportation, it will conduct a strict assessment of the actual operating effect after the adjustment. If the result of the adjustment is considered to be problematic, the technical team responsible for this link will conduct a comprehensive review of the previously used algorithm and adjust the algorithm parameters based on the actual situation. Once the correct adjustment path is found, it will be used to guide the formulation of future cargo management strategies, continuously iterate and optimize, and ultimately achieve precise and intelligent customs control.

[0130] On this basis, if the adjacent level is reselected, the change rate and increment will be adjusted according to the newly selected adjacent level. In this step, the originally set change rate is calculated based on the difference between the two specific risk levels. As the new level is selected, its change rate will also be corrected accordingly; similarly, the increment is the total amount of variables required to accumulate from the initial level to the newly selected level. After the two are adjusted, the updated current level downgrade coefficient can be calculated.

[0131] If a new interpolation method is introduced, then the corresponding formula needs to be applied, which involves the current level and the downward adjustment coefficients of the adjacent related levels as input values. At this stage, the formula can integrate and smooth the numerical differences between the levels. The parameters included in the formula have their own definitions and reasonable change ranges. For example, the parameter x represents the weight corresponding to a specific risk scenario, and the range is set in (0,1). The optimal value is usually close to 0.6 to balance the error minimization and system stability requirements. This setting is intended to ensure a good fit even in the case of multi-dimensional data fluctuations.

[0132] In addition, if Fig.10 As shown, the embodiment of the present application also provides a port customs control system 1000 based on artificial intelligence, including:

[0133] The key feature information extraction module 1001 extracts key feature information from the container cargo manifest data received at the port;

[0134] The image feature data generation module 1002 uses an image recognition algorithm to photograph the actual goods and generate image feature data;

[0135] A matching analysis module 1003 performs automatic matching analysis based on the extracted key feature information and image feature data;

[0136] The determination module 1004 determines whether there is any discrepancy between the goods and the invoice according to the matching result and outputs the inspection conclusion.

[0137] The functions of each module of the artificial intelligence-based port customs control system 1000 have been described above with reference to the accompanying drawings and will not be repeated here.

[0138] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function according to the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium.

[0139] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine different embodiments or examples described in this specification and the features of different embodiments or examples, unless they are contradictory.

[0140] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of the features. In the description of this application, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0141] Any process or method description in the flow chart or otherwise described herein can be understood to represent a module, fragment or portion of a code including one or more executable instructions for implementing the steps of a specific logical function or process. And the scope of the preferred embodiment of the present application includes other implementations, in which the functions may not be performed in the order shown or discussed, including in a substantially simultaneous manner or in a reverse order according to the functions involved.

[0142] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, which can be embodied in any computer-readable medium for use by an instruction execution system, apparatus or device (such as a computer-based system, a system including a processor or other system that can fetch instructions from an instruction execution system, apparatus or device and execute instructions), or used in combination with these instruction execution systems, apparatuses or devices.

[0143] It should be understood that the various parts of the present application can be implemented with hardware, software, firmware or a combination thereof. In the above embodiments, multiple steps or methods can be implemented with software or firmware stored in a memory and executed by a suitable instruction execution system. All or part of the steps of the above embodiment method can be completed by instructing the relevant hardware through a program, which can be stored in a computer-readable storage medium, and when the program is executed, it includes one of the steps of the method embodiment or a combination thereof.

[0144] In addition, each functional unit in each embodiment of the present application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into one module. The above-mentioned integrated module can be implemented in the form of hardware or in the form of a software functional module. If the above-mentioned integrated module is implemented in the form of a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. The storage medium can be a read-only memory, a disk or an optical disk, etc.

[0145] The above are only specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of various changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A port customs control method based on artificial intelligence, characterized in that: include: Extract key feature information from the container cargo manifest data received at the port; Use image recognition algorithms to photograph actual goods and generate image feature data; Automatically match and analyze the extracted key feature information with the image feature data; Determine whether there is any discrepancy between the cargo and the invoice based on the matching results and output the inspection conclusion.

2. The port customs control method based on artificial intelligence according to claim 1 is characterized in that: The automatic matching analysis based on the extracted key feature information and image feature data includes: The extracted key feature information is normalized to eliminate the influence of different feature dimensions; Calculate the similarity between the image feature data and the normalized key feature information; Based on the similarity, determine whether a preset matching threshold condition is met. If so, it is considered that the key feature information and the image feature data are matched successfully, otherwise the match fails; Mark the successfully matched data for subsequent verification of the conclusion output.

3. The port customs control method based on artificial intelligence according to claim 2 is characterized in that: The determining, based on the similarity, whether a preset matching threshold condition is met includes: Get the preset matching threshold, which is dynamically adjusted based on historical matching data and cargo type; Based on the comparison between the similarity and a preset matching threshold, if the similarity is greater than or equal to the threshold, the matching condition is met; If the similarity is less than the threshold, further analyze whether there is feature deviation caused by abnormal feature points in the image feature data; Based on the analysis results of abnormal feature points, determine whether it is necessary to temporarily adjust the matching threshold to adapt to the special circumstances of the current goods.

4. The port customs control method based on artificial intelligence according to claim 3 is characterized in that: The determining whether it is necessary to temporarily adjust the matching threshold based on the analysis result of the abnormal feature point includes: Count the number and distribution of abnormal feature points; Based on the number and distribution of abnormal feature points, it is determined whether it falls within the acceptable deviation range. The acceptable deviation range formula is: Acceptable deviation range = [average number of abnormal feature points − k × standard deviation, average number of abnormal feature points + k × standard deviation], where k is the confidence coefficient; If it is within the acceptable deviation range, the matching threshold is not adjusted and the matching is directly judged as failed; If the deviation exceeds the acceptable range, the matching threshold is temporarily lowered according to the influence of the abnormal feature points. The reduction range is: Downward adjustment range = α×(number of abnormal feature points / total number of feature points), where α is the adjustment coefficient.

5. The artificial intelligence-based port customs control method according to claim 4 is characterized in that: The temporarily lowering of the matching threshold according to the influence of the abnormal feature points includes: Determine the impact level of abnormal feature points and classify them according to their importance in cargo identification; Based on the impact level, find the corresponding downward adjustment coefficient table to obtain the corresponding downward adjustment coefficient; The actual downward adjustment is calculated as follows: Actual downward adjustment = downward adjustment coefficient × basic downward adjustment, where the basic downward adjustment is the pre-set minimum downward adjustment value; Based on the calculated actual downward adjustment range, the matching threshold is adjusted and the adjusted threshold is recorded.

6. The artificial intelligence-based port customs control method according to claim 5 is characterized in that: The step of searching the corresponding downward adjustment coefficient table based on the impact level to obtain the corresponding downward adjustment coefficient includes: Constructing a downward adjustment coefficient table, wherein the downward adjustment coefficient table maps the impact level to the downward adjustment coefficient based on historical data and expert experience; Input the influence level of the current abnormal feature point, and search for the corresponding downward adjustment coefficient based on the downward adjustment coefficient table; If the current impact level is not directly matched in the table, the downward adjustment coefficient of the current level is calculated by interpolation based on the downward adjustment coefficients of the adjacent levels; Based on the obtained downward adjustment coefficient, a subsequent matching threshold adjustment operation is performed.

7. The artificial intelligence-based port customs control method according to claim 6 is characterized in that: If the current impact level is not directly corresponding in the table, the downward adjustment coefficient of the current level is calculated by interpolation based on the downward adjustment coefficient of the adjacent level, including: Determine two known levels adjacent to the current impact level and the corresponding downward adjustment coefficients; Based on the downward adjustment coefficients of adjacent levels, the change rate of the downward adjustment coefficients between adjacent levels is calculated; Based on the change rate, calculating the downward adjustment coefficient increment of the current level; The increment is added to the downward adjustment coefficient of the lower level of the two known levels to obtain the downward adjustment coefficient of the current level.

8. The port customs control method based on artificial intelligence according to claim 7 is characterized in that: The step of adding the increment to the downward adjustment coefficient of the lower level of the two known levels to obtain the downward adjustment coefficient of the current level includes: Obtaining the downward adjustment coefficient of the lower level of the two known levels as the base value; Add the calculated increment to the base value to get the downward adjustment coefficient of the current level; The rationality of the calculated downward adjustment coefficient of the current level is checked. If the current downward adjustment coefficient is between 0 and 1, it is reasonable, otherwise it is unreasonable; If the test result is reasonable, the downward adjustment coefficient is used to adjust the matching threshold; if it is unreasonable, the selection or interpolation method of adjacent levels is readjusted to ensure the rationality of the downward adjustment coefficient.

9. The artificial intelligence-based port customs control method according to claim 8 is characterized in that: If the test result is reasonable, the matching threshold is adjusted using the downward adjustment coefficient; if it is unreasonable, the selection or interpolation method of the adjacent levels is readjusted, including: If the test result is reasonable, the downward adjustment coefficient of the current level is directly applied to the matching threshold adjustment formula to adjust the matching threshold; If the test result is unreasonable, analyze the reason for the unreasonableness, whether it is due to improper selection of adjacent levels or inappropriate interpolation method; Based on the analysis results, if the adjacent level is not selected properly, reselect other adjacent levels; if the interpolation method is not applicable, try other interpolation methods; Based on the reselected adjacent levels or the new interpolation method, the downward adjustment coefficient of the current level is recalculated and the rationality check is performed again until a reasonable downward adjustment coefficient is obtained.

10. A port customs control system based on artificial intelligence, characterized in that: include: A key feature information extraction module extracts key feature information from the container cargo manifest data received at the port; An image feature data generation module uses an image recognition algorithm to photograph actual goods and generate image feature data; Matching analysis module, which performs automatic matching analysis based on the extracted key feature information and image feature data; The judgment module determines whether there is any discrepancy between the goods and the invoice based on the matching results and outputs the inspection conclusion.

Images