Artificial Intelligence-based Risk Early Warning System and Method for Intelligent Electronic Products

Abstract

The invention relates to the technical field of risk warning, and discloses an intelligent electronic product risk warning system and method based on artificial intelligence. The system comprises: an acquisition module determines a target product image of an intelligent electronic product to be detected, an analysis module identifies a suspected compliance calibration, divides the target product image into a plurality of sub-quality detection areas, and analyzes each sub-quality detection area, deletes the sub-quality detection area according to the analysis result, a processing module extracts all quality pixel values ​​of the remaining sub-quality detection areas after deletion, and uses every two quality pixel values ​​on a first quality sequence or a second quality sequence as quality pixels to be associated, and verifies whether the quality pixels to be associated are associated based on a quality evaluation model, and a calculation module determines whether to issue a warning to the intelligent electronic product to be detected according to a comprehensive quality value. The invention issues a risk warning to the intelligent electronic product through an artificial intelligence learning model, thereby improving the accuracy of the warning.

Description

Technical Field

[0001] The present invention relates to the technical field of risk early warning, and in particular, to an intelligent electronic product risk early warning system and method based on artificial intelligence. Background Art

[0002] With the continuous progress of technology, artificial intelligence (AI) technology has been widely applied in various fields. The deep learning model and data analysis in artificial intelligence technology can effectively monitor and early warn electronic products in real time, so as to assist personnel in timely discovering potential problems and dealing with them.

[0003] Chinese Patent Publication No.: CN118966893A discloses an intelligent electronic product quality and safety risk monitoring and control system, including: a data acquisition unit for acquiring quality inspection data of electronic products; a data processing unit for calculating a quality inspection score according to the quality inspection data and determining a safety risk level according to the quality inspection score; a risk control unit for comparing the safety risk level with a preset maximum safety risk level threshold. When the safety risk level of the electronic product is higher than the maximum safety risk level threshold, the risk control unit determines that the electronic product is unqualified and gives an early warning. Thus, when acquiring the quality inspection data of electronic products, if the acquired quality inspection data is inaccurate, there will be a risk of incorrect early warning, and moreover, the correlation between data is not considered, resulting in the accumulation of early warning errors.

[0004] Therefore, it is necessary to design an intelligent electronic product risk early warning system and method based on artificial intelligence to solve the problems existing in the current technology. Summary of the Invention

[0005] In view of this, the present invention proposes an intelligent electronic product risk early warning system and method based on artificial intelligence, aiming to solve the problems that if the acquired detection data is inaccurate, there will be a risk of incorrect early warning, and moreover, the correlation between data is not considered, resulting in the accumulation of early warning errors.

[0006] On the one hand, the present invention proposes an intelligent electronic product risk early warning system based on artificial intelligence, including:

[0007] An acquisition module, configured to acquire an initial product image of the intelligent electronic product to be detected, and analyze the initial product image to determine a target product image of the intelligent electronic product to be detected;

[0008] An analysis module, configured to analyze the target product image, generate a quality calibration for the intelligent electronic product to be detected according to the analysis result, when a suspected compliance is recognized, divide the target product image into multiple sub-quality detection regions, and parse each sub-quality detection region, and delete the sub-quality detection regions according to the parsing result;

[0009] A processing module, configured to extract all the quality pixel values of the remaining sub-quality detection regions after deletion, compare the quality pixel values with the corresponding quality standard pixel values, divide all the quality pixel values according to the comparison result, construct a first quality sequence and a second quality sequence according to the division result, numerically sort all the quality pixel values in the first quality sequence and the second quality sequence, and use every two quality pixel values in the first quality sequence or the second quality sequence as the quality pixels to be associated, and verify whether the quality pixels to be associated are associated based on a quality evaluation model, and count the number of associations according to the verification result;

[0010] A calculation module, configured to determine the comprehensive quality value of the intelligent electronic product to be detected according to the number of associations and the quality pixel values that are not associated, and determine whether to give an early warning to the intelligent electronic product to be detected according to the comprehensive quality value.

[0011] Further, when parsing the initial product image to determine the target product image of the intelligent electronic product to be detected, it includes:

[0012] The acquisition module preprocesses the initial product image, and the preprocessing includes image denoising, contrast adjustment, and image edge sharpening;

[0013] Substitute the preprocessed initial product image into a convolutional neural network model, and output the image effective value of the preprocessed initial product image based on the convolutional neural network model;

[0014] When the image effective value is greater than or equal to a preset image effective value, the initial product image is determined as the target product image of the intelligent electronic product to be detected;

[0015] When the image effective value is less than the preset image effective value, the initial product image of the intelligent electronic product to be detected is re-acquired.

[0016] Further, when analyzing the target product image and generating a quality calibration for the intelligent electronic product to be detected according to the analysis result, it includes:

[0017] The analysis module extracts all target image pixel points of the target product image, extracts all standard target image pixel points corresponding to the target image pixel points in the standard target product image, determines the target pixel values corresponding to all target image pixel points, and determines the standard target pixel values corresponding to all standard target image pixel points;

[0018] When the target pixel values are all equal to the standard target pixel values, a quality compliance calibration is generated for the intelligent electronic product to be detected;

[0019] When the target pixel values are all not equal to the standard target pixel values, a quality non-compliance calibration is generated for the intelligent electronic product to be detected;

[0020] When there is one or more target pixel values equal to the standard target pixel values, and there is one or more target pixel values not equal to the standard target pixel values, a suspected compliance calibration is generated for the intelligent electronic product to be detected.

[0021] Further, when parsing each sub-quality detection area and deleting the sub-quality detection area according to the parsing result, it includes:

[0022] When the target pixel values in the sub-quality detection area are all equal to the standard target pixel values, the sub-quality detection area is deleted, and each target pixel value of the remaining sub-quality detection areas is used as a quality pixel value.

[0023] Further, when extracting all quality pixel values of the remaining sub-quality detection areas after deletion, comparing the quality pixel values with the corresponding quality standard pixel values, dividing all quality pixel values according to the comparison result, and constructing a first quality sequence and a second quality sequence, it includes:

[0024] The processing module extracts all quality pixel values of the remaining sub-quality detection areas after deletion, and compares each quality pixel value with the corresponding quality standard pixel value;

[0025] When the quality pixel value is greater than or equal to the quality standard pixel value, a first pixel difference between the quality pixel value and the quality standard pixel value is obtained, and the first quality sequence is constructed;

[0026] When the quality pixel value is less than the quality standard pixel value, a second pixel difference between the quality pixel value and the quality standard pixel value is obtained, and the second quality sequence is constructed;

[0027] The first pixel differences in the same sub-quality detection area in the first quality sequence are divided into a first sub-quality sequence, and the remaining first pixel differences are divided into a second sub-quality sequence. The first quality sequence includes the second sub-quality sequence and several first sub-quality sequences;

[0028] Divide the second pixel differences in the same sub-quality detection regions of the second quality sequence into a third sub-quality sequence, and divide the remaining second pixel differences into a fourth sub-quality sequence. The second quality sequence includes the fourth sub-quality sequence and several of the third sub-quality sequences.

[0029] Further, when numerically sorting all the quality pixel values in the first quality sequence and the second quality sequence, and taking every two quality pixel values on the first quality sequence or the second quality sequence as the quality pixels to be associated, it includes:

[0030] The processing module numerically sorts the first pixel differences in each of the first sub-quality sequences, and takes adjacent first pixel differences as the quality pixels to be associated with the first quality;

[0031] The processing module numerically sorts the second pixel differences in each of the third sub-quality sequences, and takes adjacent second pixel differences as the quality pixels to be associated with the second quality.

[0032] Further, when verifying whether the quality pixels to be associated are associated based on the quality evaluation model and counting the number of associations according to the verification result, it includes:

[0033] The processing module obtains the historical quality pixel values of the historical target product images, the quality standard pixel values of the standard target product images, and the historical pixel differences, and constructs a pixel data set. The pixel data set is sampled according to a preset ratio to obtain a pixel training set and a pixel test set;

[0034] The processing module pre-selects a siamese neural network model, iteratively trains the siamese neural network model according to the pixel training set, and evaluates the iteratively trained siamese neural network model according to the test subset;

[0035] If the triplet loss of the siamese neural network model after the current iterative training is greater than or equal to the triplet loss of the siamese neural network model after the previous iterative training, then the learning rate is adjusted by piecewise decay and the iterative training continues until the preset number of iterations is reached;

[0036] If the triplet loss of the siamese neural network model after the current iterative training is less than the triplet loss of the siamese neural network model after the previous iterative training, then the iterative training is stopped, and the siamese neural network model after the current iterative training is used as the quality evaluation model. The quality pixels to be associated with the first quality and the quality pixels to be associated with the second quality are respectively substituted into the quality evaluation model to output the model verification result;

[0037] When the model verification result meets the preset model verification result, it is determined that there is pixel association for the to-be-associated first quality pixel or the to-be-associated second quality pixel, and the number of associations with pixel association is counted.

[0038] Further, when determining the comprehensive quality value of the to-be-detected intelligent electronic product according to the number of associations and the quality pixel values that have not been associated, it includes:

[0039] The calculation module determines the comprehensive quality value by the following formula:

[0040] ;

[0041] ;

[0042] ;

[0043] Wherein, represents the associated quality, represents the number of associations, represents the number of the first sub-quality sequences, represents the largest first pixel difference in the th first sub-quality sequence, represents the smallest first pixel difference in the th first sub-quality sequence, represents the number of the third sub-quality sequences, represents the largest second pixel difference in the th third sub-quality sequence, represents the smallest second pixel difference in the th third sub-quality sequence, represents the number of the first pixel differences in the second sub-quality sequence, represents the largest first pixel difference in the second sub-quality sequence, represents the th first pixel difference in the second sub-quality sequence, represents the number of the second pixel differences in the fourth sub-quality sequence, represents the largest second pixel difference in the fourth sub-quality sequence, represents the th second pixel difference in the fourth sub-quality sequence, represents the comprehensive quality value, represents the weight coefficient, and the weight coefficient is determined according to the group of the to-be-detected intelligent electronic product.

[0044] Further, when determining whether to give an early warning to the to-be-detected intelligent electronic product according to the comprehensive quality value, it includes:

[0045] The comprehensive quality value and a pre-set comprehensive quality value threshold Performing a comparison, and determining whether to issue an early warning for the intelligent electronic product to be detected according to the comparison result;

[0046] when Greater than or equal to When the detected intelligent electronic product is detected, it is determined to issue an early warning;

[0047] when Less than When , it is determined that no warning is issued for the intelligent electronic product to be detected.

[0048] Compared with the prior art, the beneficial effects of the present invention are as follows: by dividing the target product image into multiple sub-quality detection areas and performing analysis, comparison and division of quality pixel values, the quality problems of the intelligent electronic products to be detected are effectively identified, and the accuracy of risk identification is ensured; by extracting and screening out the sub-quality detection areas with problems, manual intervention is reduced, and at the same time, the intelligence and automation of risk warnings are improved, and the focus is effectively on the sub-quality detection areas with problems, and the analysis and processing processes are optimized; and, by utilizing the quality evaluation model in artificial intelligence to analyze the associated quality pixels, the comprehensive quality value can be accurately calculated, thereby issuing warnings in a timely manner, reducing the risk of non-compliant intelligent electronic products to be detected entering the market, and improving the accuracy and reliability of risk warnings for intelligent electronic products.

[0049] On the other hand, the present application also provides an intelligent electronic product risk warning method based on artificial intelligence, which is applied to the above-mentioned intelligent electronic product risk warning system based on artificial intelligence, including:

[0050] Acquire an initial product image of the intelligent electronic product to be detected, and analyze the initial product image to determine a target product image of the intelligent electronic product to be detected;

[0051] Analyze the target product image, generate a quality calibration for the intelligent electronic product to be detected according to the analysis result, divide the target product image into a plurality of sub-quality detection areas when a suspected product that meets the calibration is identified, analyze each sub-quality detection area, and delete the sub-quality detection area according to the analysis result;

[0052] Extract all the quality pixel values in the remaining sub-quality detection area after extraction and deletion, compare the quality pixel values with the corresponding quality standard pixel values, divide all the quality pixel values according to the comparison results, construct a first quality sequence and a second quality sequence according to the division results, numerically sort all the quality pixel values in the first quality sequence and the second quality sequence, and use every two quality pixel values on the first quality sequence or the second quality sequence as the quality pixels to be associated, and verify whether the quality pixels to be associated are associated based on the quality evaluation model, and count the number of associations according to the verification results;

[0053] Determine the comprehensive quality value of the intelligent electronic product to be detected according to the number of quality pixels to be associated and the quality pixel values that have not been associated, and determine whether to give an early warning to the intelligent electronic product to be detected according to the comprehensive quality value.

[0054] It can be understood that the above-mentioned risk warning system and method for intelligent electronic products based on artificial intelligence have the same beneficial effects, which will not be elaborated here. Brief Description of the Drawings

[0055] By reading the following detailed description of the preferred embodiments, various other advantages and benefits will become clear to those of ordinary skill in the art. The drawings are only for the purpose of showing the preferred embodiments and are not considered to be a limitation of the present invention. Moreover, throughout the drawings, the same reference numerals are used to represent the same components. In the drawings:

[0056] Figure 1 It is a functional block diagram of a risk warning system for intelligent electronic products based on artificial intelligence provided by an embodiment of the present invention;

[0057] Figure 2 It is a flowchart of a risk warning method for intelligent electronic products based on artificial intelligence provided by an embodiment of the present invention. Detailed Embodiments

[0058] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the drawings. Although the exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided so that the present disclosure can be more thoroughly understood and the scope of the present disclosure can be fully conveyed to those skilled in the art. It should be noted that, without conflict, the embodiments in the present invention and the features in the embodiments can be combined with each other. The present invention will be described in detail below with reference to the drawings and in combination with the embodiments.

[0059] In some embodiments of the present application, referring to Figure 1 as shown, a risk warning system for intelligent electronic products based on artificial intelligence includes:

[0060] A collection module, configured to obtain an initial product image of a smart electronic product to be detected, and analyze the initial product image to determine a target product image of the smart electronic product to be detected.

[0061] An analysis module, configured to analyze the target product image, generate a quality calibration for the smart electronic product to be detected according to the analysis result. When a suspected compliance is recognized, the target product image is divided into multiple sub-quality detection regions, and each sub-quality detection region is analyzed. According to the analysis result, the sub-quality detection regions are deleted.

[0062] A processing module, configured to extract all quality pixel values of the remaining sub-quality detection regions after deletion, compare the quality pixel values with corresponding quality standard pixel values, divide all quality pixel values according to the comparison result, construct a first quality sequence and a second quality sequence according to the division result, numerically sort all quality pixel values in the first quality sequence and the second quality sequence, and use every two quality pixel values in the first quality sequence or the second quality sequence as quality pixels to be associated, and verify whether the quality pixels to be associated are associated based on a quality evaluation model, and count the number of associations according to the verification result.

[0063] A calculation module, configured to determine a comprehensive quality value of the smart electronic product to be detected according to the number of associations and the quality pixel values that are not associated, and determine whether to give an early warning to the smart electronic product to be detected according to the comprehensive quality value.

[0064] Specifically, the acquisition module obtains the initial product image of the intelligent electronic product to be detected through an image acquisition device, and uses image algorithms and deep learning models for analysis to obtain the target product image. This avoids the accumulation of errors in subsequent analysis caused by incorrect target product images, and ensures the accuracy and effectiveness of the target product image. The analysis module is responsible for analyzing the target product image to generate a quality calibration and determine whether there are risks in the intelligent electronic product to be detected. Risks include cracks or fissures on the surface of the intelligent electronic product to be detected. If there is a suspected compliance with the calibration, it is considered that there may be a risk in the intelligent electronic product to be detected and further analysis is required. The target product image is divided into multiple sub-quality detection regions. The sub-quality detection regions are preferably 20, and can be specifically divided according to the actual size of the intelligent electronic product to be detected. The intelligent division and deletion of the sub-quality detection regions helps to optimize computing resources, and each sub-quality detection region is analyzed. According to the analysis results, the sub-quality detection regions are deleted, avoiding the influence of irrelevant regions on the results of risk warning and further improving the detection accuracy. The processing module extracts all the quality pixel values in the remaining sub-quality detection regions after deletion and compares them with the corresponding standard quality pixel values, avoiding errors caused by human factors to the intelligent electronic product to be detected and improving the accuracy of risk warning. All the quality pixel values are divided to respectively construct a first quality sequence and a second quality sequence, which are hierarchically divided and numerically sorted, making the distribution of quality pixel values in the first quality sequence and the second quality sequence clearer and laying a foundation for capturing the correlation between quality pixel values subsequently. Usually, when there are cracks or fissures on the surface of the intelligent electronic product to be detected, they may not be completely independent random ruptures, but expand along the direction of stress concentration of the intelligent electronic product to be detected. Therefore, the quality pixel values at the cracks or fissures will maintain a certain correlation. The quality evaluation model is used to verify whether there is a correlation between the quality pixels to be associated, improving the intelligence level and accuracy of the risk judgment of the intelligent electronic product to be detected. The calculation module calculates the comprehensive quality value of the intelligent electronic product to be detected based on the number of associations and the unassociated quality pixel values, thereby determining whether there are risks in the intelligent electronic product to be detected. It not only pays attention to individual possible defects but also considers the overall defects of the intelligent electronic product to be detected, ensuring the accuracy and comprehensiveness of risk warning.

[0065] It can be understood that through the division of quality pixel values and the verification of the quality evaluation model, the automation level of risk warning is improved. By analyzing the initial product image, dividing the sub-quality detection regions, and verifying the quality evaluation model, the accuracy of risk warning is improved, avoiding the accumulation of errors caused by human factors or incorrect data, and thus improving the intelligence level of risk warning.

[0066] In some embodiments of the present application, when parsing the initial product image to determine the target product image of the intelligent electronic product to be detected, it includes: the acquisition module preprocesses the initial product image, and the preprocessing includes image denoising, contrast adjustment, and edge sharpening of the image. The preprocessed initial product image is input into the convolutional neural network model, and based on the convolutional neural network model, the image effective value of the preprocessed initial product image is output. When the image effective value is greater than or equal to the preset image effective value, the initial product image is determined as the target product image of the intelligent electronic product to be detected. When the image effective value is less than the preset image effective value, the initial product image of the intelligent electronic product to be detected is re-acquired.

[0067] Specifically, the acquisition module performs image denoising, contrast adjustment, and edge sharpening on the initial product image, reducing the interference of noise on the subsequent determination of the target product image, enhancing the details of the initial product image, enabling the convolutional neural network (CNN) model to extract image features, and improving the classification accuracy. The preset image effective value is preferably 0.7. The convolutional neural network model is pre-trained. During the training process, a large number of preprocessed initial product images are used for training. The convolutional neural network model includes multiple convolutional layers, pooling layers, and fully connected layers. By training the convolutional neural network model, it can learn and capture the image features of the initial product image, and based on the softmax function of the convolutional neural network model, the probability of the effectiveness or ineffectiveness of the image, that is, the image effective value, is output, laying the foundation for subsequent analysis, avoiding deviations in risk warnings caused by incorrect target product images, and improving the accuracy of the target product image.

[0068] In some embodiments of the present application, when analyzing the target product image and generating a quality calibration for the intelligent electronic product to be detected according to the analysis result, it includes: the analysis module extracts all the target image pixel points of the target product image, and extracts all the standard target image pixel points corresponding to the target image pixel points in the standard target product image, determines the target pixel values corresponding to all the target image pixel points, determines the standard target pixel values corresponding to all the standard target image pixel points. When all the target pixel values are equal to the standard target pixel values, a quality compliance calibration is generated for the intelligent electronic product to be detected. When all the target pixel values are not equal to the standard target pixel values, a quality non-compliance calibration is generated for the intelligent electronic product to be detected. When there is one or more target pixel values equal to the standard target pixel values, and there is one or more target pixel values not equal to the standard target pixel values, a suspected compliance calibration is generated for the intelligent electronic product to be detected.

[0069] Specifically, the standard target product image is obtained based on the standard intelligent electronic product. Moreover, the target image pixel points and the standard target image pixel points correspond one by one, ensuring the corresponding relationship between the target pixel values and the standard target pixel values. That the quality conforms to the calibration indicates that the intelligent electronic product to be detected meets the standard, and that the quality does not conform to the calibration indicates that the intelligent electronic product to be detected does not meet the standard, then a warning is directly issued. When it is generated that it is suspected to conform to the calibration, it indicates that the intelligent electronic product to be detected may or may not meet the standard, and further analysis and processing are required to avoid missed judgments.

[0070] In some embodiments of the present application, when parsing each sub-quality detection region and deleting the sub-quality detection region according to the parsing result, it includes: when all the target pixel values in a sub-quality detection region are equal to the standard target pixel values, then delete the sub-quality detection region, and use each target pixel value of the remaining sub-quality detection regions as the quality pixel value.

[0071] Specifically, when all the target pixel values in a sub-quality detection region are consistent with the standard target pixel values, it means that the sub-quality detection region conforms to the standard in the standard target product image, and no further analysis is required. The sub-quality detection region is directly deleted, and only the sub-quality detection regions with differences are retained for subsequent processing, reducing the impact of regional redundancy on subsequent analysis.

[0072] In some embodiments of the present application, when extracting all the quality pixel values of the remaining sub-quality detection regions after deletion, comparing the quality pixel values with the corresponding quality standard pixel values, and dividing all the quality pixel values according to the comparison result, and constructing the first quality sequence and the second quality sequence according to the division result, it includes: the processing module extracts all the quality pixel values of the remaining sub-quality detection regions after deletion, and compares each quality pixel value with the corresponding quality standard pixel value. When the quality pixel value is greater than or equal to the quality standard pixel value, then obtain the first pixel difference between the quality pixel value and the quality standard pixel value, and construct the first quality sequence. When the quality pixel value is less than the quality standard pixel value, then obtain the second pixel difference between the quality pixel value and the quality standard pixel value, and construct the second quality sequence. Divide the first pixel differences in the same sub-quality detection region in the first quality sequence into the first sub-quality sequence, and divide the remaining first pixel differences into the second sub-quality sequence. The first quality sequence includes the second sub-quality sequence and several first sub-quality sequences. Divide the second pixel differences in the same sub-quality detection region in the second quality sequence into the third sub-quality sequence, and divide the remaining second pixel differences into the fourth sub-quality sequence. The second quality sequence includes the fourth sub-quality sequence and several third sub-quality sequences.

[0073] Specifically, the target pixel values corresponding to the remaining sub-quality detection regions are used as quality pixel values, and the standard target pixel values corresponding in the standard target product image are used as quality standard pixel values. The first pixel difference indicates being greater than or equal to the quality standard pixel value, and the second pixel difference indicates being less than the quality standard pixel value. They are separated and further refined for the same sub-quality detection region, laying a foundation for capturing the correlation between quality pixel values for the same sub-quality detection region, and dividing the quality pixel values of the remaining different sub-quality detection regions, avoiding misjudgment caused by ignoring the whole due to a single difference and improving the accuracy of risk warning.

[0074] In some embodiments of the present application, when numerically sorting all quality pixel values in the first quality sequence and the second quality sequence, and taking every two quality pixel values in the first quality sequence or the second quality sequence as quality pixels to be associated, it includes: the processing module numerically sorts the first pixel differences in each first sub-quality sequence, and takes adjacent first pixel differences as the first quality pixels to be associated, and the processing module numerically sorts the second pixel differences in each third sub-quality sequence, and takes adjacent second pixel differences as the second quality pixels to be associated.

[0075] Specifically, according to the first pixel differences of the same sub-quality detection region, they are divided into first sub-quality sequences, and there may be multiple first sub-quality sequences. The same is true for the third sub-quality sequences. Therefore, numerically sorting the first pixel differences in each first sub-quality sequence and the second pixel differences in each third sub-quality sequence, when there are cracks or fissures on the surface of the intelligent electronic product to be detected, the cracks or fissures will expand along the direction of stress concentration of the intelligent electronic product to be detected, resulting in a certain correlation between the first pixel differences or the second pixel differences. Through numerical sorting, the first pixel differences or the second pixel differences can be aggregated, avoiding the influence of disordered data on the analysis result and laying a foundation for the subsequent quality evaluation model.

[0076] In some embodiments of the present application, when verifying whether the quality pixels to be associated are associated based on the quality evaluation model and counting the number of associations according to the verification result, it includes: The processing module obtains the historical quality pixel values of the historical target product images, the quality standard pixel values of the standard target product images, and the historical pixel differences, and constructs a pixel data set. The pixel data set is sampled according to a preset ratio to obtain a pixel training set and a pixel test set. The processing module pre-selects a siamese neural network model, iteratively trains the siamese neural network model according to the pixel training set, and evaluates the iteratively trained siamese neural network model according to the test subset. If the triplet loss of the siamese neural network model after the current iterative training is greater than or equal to the triplet loss of the siamese neural network model after the previous iterative training, the learning rate is adjusted by piecewise decay and the iterative training continues until the preset number of iterations is reached. If the triplet loss of the siamese neural network model after the current iterative training is less than the triplet loss of the siamese neural network model after the previous iterative training, the iterative training is stopped, and the siamese neural network model after the current iterative training is used as the quality evaluation model. The first quality pixel to be associated and the second quality pixel to be associated are respectively input into the quality evaluation model to output the model verification result. When the model verification result meets the preset model verification result, it is determined that there is a pixel association between the first quality pixel to be associated or the second quality pixel to be associated, and the number of associations with pixel associations is counted.

[0077] Specifically, the historical quality pixel values of the historical target product images, the quality standard pixel values of the standard target product images, and the historical pixel differences record all the data of the intelligent electronic products in different periods. By constructing a pixel data set and sampling it according to a preset ratio to obtain a pixel training set and a pixel test set, the preset ratio is preferably a sampling of 3:2, so that the pixel training set and the pixel test set contain various data in different periods. The pixel training set is used to train the siamese neural network model, and the pixel test set is used to evaluate the performance of the trained model. The siamese neural network model contains multiple layers, and each layer is responsible for extracting different levels of features. In the siamese neural network model, the same hierarchical structure is shared to continuously extract features to capture the similarity between the input data. The activation function is used for non-linear transformation to help the siamese neural network model capture complex patterns in the data. By training the siamese neural network model, it can accurately judge whether there is an association between the first quality pixel to be associated or the second quality pixel to be associated, and thus output the verification result, avoiding judgment errors caused by simple calculation.

[0078] It can be understood that the triplet loss enables the siamese neural network to pull similar data closer and push dissimilar data farther apart. If the triplet loss of the siamese neural network model after the current iterative training is less than that after the previous iterative training, piecewise decay is used to adjust the learning rate of the model, and iterative training continues. In each iteration, the model tries to learn the patterns and relationships in the input data to improve its prediction or verification ability until the preset number of iterations is reached. If the triplet loss of the siamese neural network model after the current iterative training is less than that after the previous iterative training, iterative training is stopped. At this time, the siamese neural network model has tended to be stable and can approach the global optimal solution. Taking it as the quality evaluation model can effectively determine whether there is pixel association for the to-be-associated first quality pixel or the to-be-associated second quality pixel, and thus obtain the number of associations with pixel association based on all the to-be-associated first quality pixels and all the to-be-associated second quality pixels.

[0079] In some embodiments of the present application, when determining the comprehensive quality value of the to-be-detected intelligent electronic product according to the number of associations and the quality pixel values that have not been associated, it includes:

[0080] The calculation module determines the comprehensive quality value by the following formula:

[0081] ;

[0082] ;

[0083] ;

[0084] Wherein, represents the associated quality, represents the number of associations, represents the number of the first sub-quality sequences, represents the largest first pixel difference in the th first sub-quality sequence, represents the smallest first pixel difference in the th first sub-quality sequence, represents the number of the third sub-quality sequences, represents the largest second pixel difference in the th third sub-quality sequence, represents the smallest second pixel difference in the th third sub-quality sequence, represents the number of first pixel differences in the second sub-quality sequence, represents the largest first pixel difference in the second sub-quality sequence, Indicates the first pixel difference in the second sub-quality sequence, represents the number of second pixel differences in the fourth sub-quality sequence, represents the largest second pixel difference in the fourth sub-quality sequence, Indicates the second pixel difference in the fourth sub-quality sequence, represents the comprehensive quality value, represents the weight coefficient, and the weight coefficient is determined according to the group of the intelligent electronic product to be detected.

[0085] Specifically, the weight coefficient is determined according to the group of the intelligent electronic product to be detected. When the intelligent electronic product to be detected is used by general consumers (excluding the elderly and young children), the weight coefficient is selected as 1.2. When the intelligent electronic product to be detected is used by the elderly and young children (children over 36 months old to 14 years old), the weight coefficient is 1.6. When the intelligent electronic product to be detected is used by infants and toddlers (0 to 36 months old), the weight coefficient is 1.8. When the intelligent electronic product to be detected is used by all age groups, the weight coefficient is 1.4. Calculate according to the audience group, related quality and non-related quality to determine the comprehensive quality value, so as to comprehensively measure the risk of the intelligent electronic product to be detected and improve the flexibility and adaptability of the risk warning for the intelligent electronic product to be detected.

[0086] In some embodiments of the present application, when determining whether to give a warning to the intelligent electronic product to be detected according to the comprehensive quality value, it includes: comparing the comprehensive quality value with a preset comprehensive quality value threshold and determining whether to give a warning to the intelligent electronic product to be detected according to the comparison result. When is greater than or equal to , it is determined to give a warning to the intelligent electronic product to be detected. When is less than , it is determined not to give a warning to the intelligent electronic product to be detected.

[0087] Specifically, by comparing the comprehensive quality value of the intelligent electronic product to be detected with the preset comprehensive quality value threshold, the state of the intelligent electronic product to be detected can be accurately judged. When the comprehensive quality value is greater than or equal to the comprehensive quality value threshold, the non-compliant intelligent electronic product can be accurately identified, so as to achieve effective warning. Analyze and verify according to the convolutional neural network model and the quality evaluation model to ensure the accuracy of the analysis process, thereby improving the accuracy of the risk warning.

[0088] In summary, the beneficial effects of the present invention are as follows: By dividing the target product image into multiple sub-quality detection regions, analyzing, comparing, and dividing the quality pixel values, the quality problems of the intelligent electronic products to be detected are effectively identified, ensuring the accuracy of risk identification. By extracting and screening the problematic sub-quality detection regions, the manual intervention is reduced, and at the same time, the intelligence and automation of risk warning are improved. The analysis and processing process is optimized by effectively focusing on the problematic sub-quality detection regions. Moreover, by using the quality evaluation model in artificial intelligence to analyze the quality pixels to be associated, the comprehensive quality value can be accurately calculated, so as to issue a warning in a timely manner, reducing the risk of non-compliant intelligent electronic products to be detected flowing into the market, and improving the accuracy and reliability of the risk warning for intelligent electronic products.

[0089] In another preferred embodiment based on the above embodiments, refer to Figure 2 As shown, this embodiment provides an artificial intelligence-based risk warning method for intelligent electronic products, which is applied to the above-mentioned artificial intelligence-based risk warning system for intelligent electronic products, and includes:

[0090] S100: Obtain the initial product image of the intelligent electronic product to be detected, and analyze the initial product image to determine the target product image of the intelligent electronic product to be detected.

[0091] S200: Analyze the target product image, generate a quality calibration for the intelligent electronic product to be detected according to the analysis result. When it is recognized that it is suspected to meet the standard, divide the target product image into multiple sub-quality detection regions, and analyze each sub-quality detection region, and delete the sub-quality detection regions according to the analysis result.

[0092] S300: Extract all the quality pixel values of the remaining sub-quality detection regions after deletion, compare the quality pixel values with the corresponding quality standard pixel values, divide all the quality pixel values according to the comparison result, construct the first quality sequence and the second quality sequence according to the division result, numerically sort all the quality pixel values in the first quality sequence and the second quality sequence, and use every two quality pixel values in the first quality sequence or the second quality sequence as the quality pixels to be associated, and verify whether the quality pixels to be associated are associated based on the quality evaluation model, and count the number of associations according to the verification result.

[0093] S400: Determine the comprehensive quality value of the intelligent electronic product to be detected according to the number of quality pixels to be associated and the quality pixel values that are not associated, and determine whether to issue a warning for the intelligent electronic product to be detected according to the comprehensive quality value.

[0094] Specifically, the initial product image of the intelligent electronic product to be detected is obtained, and it is analyzed to determine the target product image. By processing the initial product image, the key features of the intelligent electronic product to be detected can be effectively extracted, laying the foundation for the analysis of subsequent risk warnings. By analyzing the target product image of the intelligent electronic product to be detected and generating a quality calibration, it is determined whether there is a suspected compliance calibration, the target product image is divided into multiple sub-quality detection areas, and each sub-quality detection area is analyzed. According to the analysis results, the sub-quality detection areas are deleted to remove irrelevant or standard sub-quality detection areas, thereby focusing on the sub-quality detection areas with problems, reducing the subsequent calculation process, thereby improving the accuracy of risk warnings. The quality pixel value is compared with the corresponding quality standard pixel value, and all the quality pixel values ​​are divided according to the comparison results, and every two quality pixel values ​​on the first quality sequence or the second quality sequence are used as quality pixels to be associated. Based on the quality evaluation model, it is verified whether the quality pixels to be associated are associated, and the deep learning model is used for training and the quality evaluation model is obtained, which further improves the accuracy of risk warnings. The comprehensive quality value of the smart electronic product to be tested is determined based on the number of quality pixels to be associated and not associated. The comprehensive quality value reflects the overall quality of the smart electronic product to be tested. Whether to issue an early warning for the smart electronic product to be tested is determined based on the comprehensive quality value. This achieves accurate evaluation of the smart electronic products to be tested, thereby identifying risks and issuing early warnings in real time, thereby improving the accuracy and reliability of risk early warnings.

[0095] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0096] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0097] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to operate in a particular manner, such that the instructions stored in the computer-readable memory produce a manufacture including an instruction device that implements the functions specified in one or more of the processes and / or blocks Figure 1 in one or more of the processes and / or blocks Figure 1 specified in one or more of the processes and / or blocks.

[0098] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, whereby the instructions executed on the computer or other programmable apparatus provide steps for implementing the functions specified in one or more of the processes and / or blocks Figure 1 in one or more of the processes and / or blocks Figure 1 specified in one or more of the processes and / or blocks.

[0099] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: modifications or equivalent substitutions can still be made to the specific embodiments of the present invention, and any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered by the protection scope of the claims of the present invention.

Claims

1. An intelligent electronic product risk warning system based on artificial intelligence, characterized in that: include: An acquisition module is configured to acquire an initial product image of the intelligent electronic product to be detected, and analyze the initial product image to determine a target product image of the intelligent electronic product to be detected; an analysis module configured to analyze the target product image, generate a quality calibration for the intelligent electronic product to be detected according to the analysis result, divide the target product image into a plurality of sub-quality detection areas when a suspected conformity with the calibration is identified, analyze each sub-quality detection area, and delete the sub-quality detection area according to the analysis result; The processing module is configured to extract all quality pixel values ​​of the remaining sub-quality detection area after deletion, and compare the quality pixel values ​​with the corresponding quality standard pixel values, divide all the quality pixel values ​​according to the comparison result, construct a first quality number series and a second quality number series according to the division result, numerically sort all the quality pixel values ​​in the first quality number series and the second quality number series, and use every two quality pixel values ​​in the first quality number series or the second quality number series as quality pixels to be associated, and verify whether the quality pixels to be associated are associated based on the quality evaluation model, and count the number of associations according to the verification result; The calculation module is configured to determine the comprehensive quality value of the intelligent electronic product to be detected according to the associated quantity and the quality pixel value without association, and determine whether to issue an early warning for the intelligent electronic product to be detected according to the comprehensive quality value.

2. The intelligent electronic product risk early warning system based on artificial intelligence according to claim 1 is characterized in that: When the initial product image is parsed to determine the target product image of the intelligent electronic product to be detected, it includes: The acquisition module preprocesses the initial product image, wherein the preprocessing includes image denoising, contrast adjustment, and image edge sharpening; Substituting the preprocessed initial product image into the convolutional neural network model, and outputting the image effective value of the preprocessed initial product image based on the convolutional neural network model; When the image validity value is greater than or equal to a preset image validity value, the initial product image is determined as the target product image of the intelligent electronic product to be detected; When the image validity value is less than the preset image validity value, the initial product image of the to-be-detected intelligent electronic product is reacquired.

3. The risk early warning system for intelligent electronic products based on artificial intelligence according to claim 2 is characterized in that: When analyzing the target product image and generating a quality calibration for the intelligent electronic product to be inspected according to the analysis result, it includes: The analysis module extracts all target image pixels of the target product image, extracts all standard target image pixels corresponding to the target image pixels in the standard target product image, determines the target pixel values ​​corresponding to all target image pixels, and determines the standard target pixel values ​​corresponding to all standard target image pixels; When the target pixel values ​​are all equal to the standard target pixel values, a quality compliance calibration is generated for the intelligent electronic product to be inspected; When the target pixel values ​​are not equal to the standard target pixel values, a quality non-conformity calibration is generated for the intelligent electronic product to be detected; When one or more target pixel values ​​are equal to the standard target pixel value, and one or more target pixel values ​​are not equal to the standard target pixel value, a suspected compliance calibration is generated for the intelligent electronic product to be detected.

4. The risk early warning system for intelligent electronic products based on artificial intelligence according to claim 3 is characterized in that: When parsing each sub-quality detection area and deleting the sub-quality detection area according to the parsing result, it includes: When there is a sub-quality detection region whose target pixel values ​​are all equal to the standard target pixel value, the sub-quality detection region is deleted, and each target pixel value of the remaining sub-quality detection region is used as a quality pixel value.

5. The risk early warning system for intelligent electronic products based on artificial intelligence according to claim 4 is characterized in that: When all quality pixel values ​​of the remaining sub-quality detection area after deletion are extracted, and the quality pixel values ​​are compared with corresponding quality standard pixel values, all quality pixel values ​​are divided according to the comparison result, and the first quality number series and the second quality number series are constructed according to the division result, it includes: The processing module extracts all quality pixel values ​​of the remaining sub-quality detection areas after deletion, and compares each quality pixel value with the corresponding quality standard pixel value; When the quality pixel value is greater than or equal to the quality standard pixel value, a first pixel difference between the quality pixel value and the quality standard pixel value is obtained, and the first quality sequence is constructed; When the quality pixel value is less than the quality standard pixel value, a second pixel difference between the quality pixel value and the quality standard pixel value is obtained, and the second quality sequence is constructed; Dividing first pixel differences of the same sub-mass detection area in the first mass number sequence into a first sub-mass number sequence, and dividing the remaining first pixel differences into a second sub-mass number sequence, wherein the first mass number sequence includes the second sub-mass number sequence and a plurality of the first sub-mass number sequences; The second pixel differences of the same sub-mass detection area in the second mass number sequence are divided into a third sub-mass number sequence, and the remaining second pixel differences are divided into a fourth sub-mass number sequence, wherein the second mass number sequence includes the fourth sub-mass number sequence and a plurality of the third sub-mass number sequences.

6. The intelligent electronic product risk early warning system based on artificial intelligence according to claim 5 is characterized in that: When all the quality pixel values ​​in the first quality number sequence and the second quality number sequence are numerically sorted, and every two quality pixel values ​​in the first quality number sequence or the second quality number sequence are used as quality pixels to be associated, the method includes: The processing module sorts the first pixel difference values ​​in each of the first sub-mass sequence by numerical value, and uses adjacent first pixel difference values ​​as first mass pixels to be associated; The processing module sorts the second pixel differences in each of the third sub-quality series by numerical value, and uses adjacent second pixel differences as second quality pixels to be associated.

7. The intelligent electronic product risk early warning system based on artificial intelligence according to claim 6 is characterized in that: When verifying whether the to-be-associated quality pixels are associated based on the quality evaluation model and counting the number of associations according to the verification result, it includes: The processing module obtains historical quality pixel values ​​of historical target product images, quality standard pixel values ​​of standard target product images and historical pixel difference values, and constructs a pixel data set, and samples the pixel data set according to a preset ratio to obtain a pixel training set and a pixel test set; The processing module preselects a twin neural network model, iteratively trains the twin neural network model according to the pixel training set, and evaluates the iteratively trained twin neural network model according to the pixel test set; If the ternary loss of the twin neural network model after the current iterative training is greater than or equal to the ternary loss of the twin neural network model after the previous iterative training, the learning rate is adjusted using piecewise decay, and iterative training continues until the preset number of iterations is reached; If the ternary loss of the twin neural network model after the current iterative training is less than the ternary loss of the twin neural network model after the previous iterative training, the iterative training is stopped, and the twin neural network model after the current iterative training is used as the quality evaluation model, and the first quality pixel to be associated and the second quality pixel to be associated are respectively substituted into the quality evaluation model, and the model verification result is output; When the model verification result meets the preset model verification result, it is determined that the to-be-associated first quality pixels or the to-be-associated second quality pixels have pixel associations, and the number of associations with pixel associations is counted.

8. The risk early warning system for intelligent electronic products based on artificial intelligence according to claim 7 is characterized in that: When determining the comprehensive quality value of the intelligent electronic product to be detected according to the associated quantity and the unassociated quality pixel value, it includes: The calculation module determines the comprehensive quality value by the following formula: ； ； ； in, represents the quality of association, Indicates the number of associations, represents the number of the first sub-mass series, Indicates The largest first pixel difference in the first sub-mass sequence, Indicates The smallest first pixel difference in the first sub-mass sequence, represents the number of the third sub-mass series, Indicates The largest second pixel difference in the third sub-mass sequence, Indicates The smallest second pixel difference in the third sub-mass sequence, represents non-associated quality, represents the number of first pixel difference values ​​in the second sub-mass sequence, represents the maximum first pixel difference in the second sub-mass sequence, Indicates the number of the second sub-mass sequence The first pixel difference, represents the number of second pixel differences in the fourth sub-mass sequence, represents the maximum second pixel difference in the fourth sub-mass sequence, Indicates the fourth sub-mass sequence The second pixel difference, Represents the comprehensive quality value, Represents a weight coefficient, and the weight coefficient is determined according to the group of smart electronic products to be detected.

9. The intelligent electronic product risk early warning system based on artificial intelligence according to claim 8 is characterized in that: When determining whether to issue an early warning to the intelligent electronic product to be detected according to the comprehensive quality value, it includes: The comprehensive quality value and a pre-set comprehensive quality value threshold Performing a comparison, and determining whether to issue an early warning for the intelligent electronic product to be detected according to the comparison result; when Greater than or equal to When the detected intelligent electronic product is detected, it is determined to issue an early warning; when Less than When , it is determined that no warning is issued for the intelligent electronic product to be detected.

10. An artificial intelligence-based smart electronic product risk warning method, applied to the artificial intelligence-based smart electronic product risk warning system as claimed in any one of claims 1 to 9, characterized in that: include: Acquire an initial product image of the intelligent electronic product to be detected, and analyze the initial product image to determine a target product image of the intelligent electronic product to be detected; Analyze the target product image, generate a quality calibration for the intelligent electronic product to be detected according to the analysis result, divide the target product image into a plurality of sub-quality detection areas when a suspected product that meets the calibration is identified, analyze each sub-quality detection area, and delete the sub-quality detection area according to the analysis result; Extracting all quality pixel values ​​of the remaining sub-quality detection area after deletion, and comparing the quality pixel values ​​with the corresponding quality standard pixel values, dividing all quality pixel values ​​according to the comparison result, constructing a first quality number series and a second quality number series according to the division result, numerically sorting all quality pixel values ​​in the first quality number series and the second quality number series, and taking every two quality pixel values ​​in the first quality number series or the second quality number series as quality pixels to be associated, and verifying whether the quality pixels to be associated are associated based on the quality evaluation model, and counting the number of associations according to the verification result; The comprehensive quality value of the intelligent electronic product to be detected is determined according to the number to be associated and the quality pixel value that has not been associated, and whether to issue an early warning to the intelligent electronic product to be detected is determined according to the comprehensive quality value.

Images