A dish visual processing detection system

Abstract

The present application relates to the field of image processing, and more particularly to a dish visual processing detection system, comprising: a collection and analysis module, configured to perform quality detection on a randomly selected target detection product and key label image collection, and to detect a collection quality coefficient of each character image in the key label image, and determine the category of the character image according to the collection quality coefficient; a data analysis module, configured to determine a processing mode according to an abnormal character proportion; a secondary processing module, configured to determine a to-be-adjusted character region and perform region adjustment, and determine an image processing mode according to a contrast reference value and a proportion of a reflective region; and a character analysis module, configured to determine a character recognition mode according to an abnormal key coefficient, so that the accuracy of the information recognition result of the key label image with abnormal character images is improved, and the degree of dependence on personnel recognition is reduced.

Description

Technical Field

[0001] This invention relates to the field of image processing, and more particularly to a visual processing and detection system for dishes. Background Technology

[0002] To ensure the quality requirements of pre-prepared food products, it is necessary to conduct multi-faceted quality inspections on the prepared products and trace the production process based on the label information on their packaging. By acquiring images of the label information on the packaging and completing information recognition, the consumption of human resources can be effectively reduced. However, in actual working scenarios, the image quality and information recognition difficulty of the acquired label images vary greatly, which can easily affect the information recognition process. Therefore, how to determine a targeted analysis method based on the actual acquired label images of pre-prepared food product packaging to further reduce the reliance on human resources in actual working scenarios is a problem that urgently needs to be solved by those skilled in the art.

[0003] Chinese Patent Publication No. CN116863483A discloses a text recognition method and an intelligent review method and system for pre-packaged food labels. This method can perform semantic analysis on complexly formatted image documents and, combined with a pre-packaged label standard information database, can achieve label review. The present invention addresses the shortcomings of existing OCR text recognition algorithms in recognizing extremely large images by first cutting the large image into smaller images, performing text detection on the smaller images, merging the detection results, and finally using the merged detection results to crop and recognize the text in the original image. However, the above solution has the following problems: it fails to determine a targeted analysis and processing method based on the actual situation of the acquired product label image, resulting in low accuracy and efficiency of information recognition, and consequently, a waste of human resources in actual work scenarios. Summary of the Invention

[0004] To address this issue, the present invention provides a visual processing and inspection system for dishes, which overcomes the problem that existing technologies fail to determine targeted analysis and processing methods based on the actual situation of the acquired product label images, resulting in low accuracy and efficiency of information recognition, and consequently, a waste of human resources in actual work scenarios.

[0005] To achieve the above objectives, the present invention provides a visual processing and inspection system for dishes, comprising:

[0006] The data acquisition and analysis module is used to perform quality inspection and key label image acquisition on randomly selected target products, and to detect the acquisition quality coefficient of each character image within the key label image, and determine the category of the character image based on the acquisition quality coefficient.

[0007] The key analysis module, which is connected to the acquisition and analysis module, is used to respond to whether there are abnormal character images in the key label image, so as to determine whether to perform abnormal key coefficient analysis on each character block in the key label.

[0008] The data analysis module, which is connected to the acquisition and analysis module, is used to determine the processing method based on the proportion of abnormal characters corresponding to the processing conditions. This method involves detecting the contrast reference value and the proportion of reflective areas of the key label image, or detecting the difference in the range of ink marks in each abnormal character area.

[0009] A secondary processing module, which is connected to the acquisition and analysis module and the data analysis module respectively, is used to determine the character area to be adjusted and perform area adjustment, as well as the contrast reference value and the proportion of reflective area corresponding to the processing conditions, so as to determine the image processing method accordingly.

[0010] The character analysis module is connected to the acquisition and analysis module, the secondary processing module, and the data analysis module, respectively. It is used to respond to the abnormal key coefficients corresponding to the set conditions, and to determine the character recognition method accordingly, either by extracting features from each abnormal character image within the character block, or by determining the character information of the abnormal character image based on the character information of character blocks of the same information category.

[0011] Furthermore, the data acquisition and analysis module performs quality testing on randomly selected target products to obtain quality testing results;

[0012] For a single target detection product,

[0013] If the acquisition and analysis module responds to the acquisition condition that the detection result of any type of detection is not within the preset result range, then it is determined that key label image acquisition will be performed for the target detection product.

[0014] Furthermore, the acquisition and analysis module detects the acquisition quality coefficient of each character image within the key label image of the target product and determines the category of the corresponding character image based on the acquisition quality coefficient;

[0015] For a single character image, the image analysis module responds with the classification condition that if the acquisition quality coefficient is less than or equal to the preset acquisition quality coefficient, then the character image is determined to be an abnormal character image.

[0016] If the classification condition of the image analysis module is that the acquisition quality coefficient is greater than the preset acquisition quality coefficient, then the character image is determined to be a standard character image.

[0017] The acquisition quality coefficient is determined based on the similarity between the character image and each reference character image.

[0018] Furthermore, if the analysis condition of the key analysis module is that there are abnormal character images within the key label image of the target detection product, then it determines to perform abnormal key coefficient analysis on each character block within the key label, including:

[0019] According to preset rules, the character information within the key tag is divided into character blocks, and the proportion of abnormal character images in each character block and the key coefficient of each abnormal character in each character block are detected. The abnormal key coefficient of the corresponding character block is determined based on the proportion of abnormal character images in each character block and the average key coefficient.

[0020] The abnormal key coefficient is positively correlated with the proportion of abnormal character images and the average key coefficient.

[0021] Furthermore, the key analysis module determines the key coefficient of the abnormal character at the corresponding character position based on the character repetition coefficient of each character position within the character block of the same information category in the current production cycle, and records the key coefficient of each abnormal character within the character block as the average key coefficient of the corresponding character block.

[0022] The critical coefficient of the abnormal character is negatively correlated with the character repetition coefficient at the corresponding character position.

[0023] Furthermore, if the processing condition of the data analysis module is that the proportion of abnormal characters in the key label image is greater than the preset proportion of abnormal characters, then the secondary processing module will detect the contrast reference value and the proportion of reflective area of ​​the key label image.

[0024] The image processing condition responded by the secondary processing module is that the contrast reference value of the key label image is greater than the preset contrast reference value and the proportion of reflective area is less than or equal to the preset proportion of reflective area. Then, it is determined to perform grayscale transformation on the key label image, divide the key label image into several sub-image regions, and detect the optimal threshold for each sub-image region.

[0025] If the image processing condition of the secondary processing module is that the contrast reference value is less than or equal to the preset contrast reference value or the proportion of reflective area is greater than the preset proportion of reflective area, then it is determined to perform similarity analysis on each pixel in the key label image.

[0026] Furthermore, the processing condition for the data analysis module to respond is that the proportion of abnormal characters in the key label image is less than or equal to the preset proportion of abnormal characters, and then it is determined to detect the difference value of the character ink range for each abnormal character region.

[0027] For a single abnormal character region, the adjustment condition of the secondary processing module is that the difference value of the character ink range is greater than the preset difference value of the character ink range, and then the adjacent character regions of the abnormal character region are determined to be the character regions to be adjusted.

[0028] For a single character region to be adjusted, the border and adjustment distance of the region to be adjusted are determined based on the adjacent directions of its neighboring abnormal character regions and the difference in the character ink range.

[0029] Furthermore, in response to the recognition conditions, the character analysis module determines to perform character recognition for each abnormal character image;

[0030] For a single character block containing an abnormal character image,

[0031] The character analysis module responds when the abnormal key coefficient of the character block is greater than the preset abnormal key coefficient. Then, it determines that the first analysis unit performs feature extraction on each abnormal character image within the character block.

[0032] The character analysis module responds when the abnormal key coefficient of the character block is less than or equal to the preset abnormal key coefficient. Then, the second analysis unit determines the character information of the abnormal character image in the character block based on the character information of character blocks of the same information category in the current production cycle.

[0033] The recognition condition is that all regions of characters to be adjusted have completed the region adjustment.

[0034] Furthermore, the first analysis unit responds to the feature extraction conditions and performs feature extraction on the images of each abnormal character within the character block;

[0035] For a single abnormal character image, the first analysis unit compares the extracted character features with the character features of each benchmark character image to determine the feature overlap, and determines the suspected character based on the feature overlap.

[0036] The feature extraction condition is determined by the character analysis module, which determines that the first analysis unit performs feature extraction on each abnormal character image within the character block.

[0037] Furthermore, the first analysis unit responds to the matching condition and determines to perform product quality matching for the corresponding character block. The product quality matching process includes:

[0038] Based on each suspected character, several tag information to be determined were identified.

[0039] For a single undetermined label information, the difference in quality parameters between the target test product to which the undetermined label information belongs and the associated test product is detected.

[0040] Priority coefficients are set for the label information to be determined based on the degree of difference in quality parameters;

[0041] The matching condition is that all abnormal character images within the character block have been identified as suspected characters.

[0042] Compared with the prior art, the beneficial effect of the present invention is that the technical solution of the present invention makes targeted adjustments to the subsequent image processing and information recognition process based on the actual situation of the key label images of abnormal products collected, so that the selected image processing method and information recognition process are more in line with the actual situation, thereby improving the accuracy of information recognition results for key label images with abnormal character images, and thus reducing the reliance on human identification.

[0043] Furthermore, in the present invention, when there are abnormal character images in the key label image, the character information in the key label is divided into character blocks, and the abnormal key coefficient of the corresponding character block is determined according to the proportion of abnormal characters in each character block and the average key coefficient. Since some characters in the character information have fixed information in the same production cycle, adding this factor to the determination process of the abnormal key coefficient can effectively determine the degree of information loss and the difficulty of information recognition of each character block, thereby making the determination of the subsequent character recognition method more in line with the actual situation.

[0044] Furthermore, in this invention, if the proportion of abnormal characters in the key label image is greater than the preset proportion of abnormal characters, the contrast reference value and the proportion of reflective area of ​​the key label image are detected, and the image processing method is determined based on the contrast reference value and the proportion of reflective area. If the proportion of abnormal characters is too large, it is more likely that the image acquisition quality is poor, resulting in most of the characters being unrecognizable. In addition, since there are differences in the position of the key label and the packaging background of different target detection products, the image processing method is determined based on the contrast reference value and the proportion of reflective area. This invention improves the effectiveness of the image processing process.

[0045] Furthermore, in this invention, if the proportion of abnormal characters in the key label image is less than or equal to the preset proportion of abnormal characters, the associated influence parameters are detected for each abnormal character region. When the difference value of the character ink range is greater than the preset difference value of the character ink range, the region to be adjusted is adjusted. Since printing defects can easily occur during the printing process of key labels on product packaging, resulting in the expansion of the ink range, which in turn affects adjacent character regions, adjusting the character region according to the adjacent direction of each adjustable character region and its abnormal character region and the difference value of the character ink range can effectively reduce the impact of abnormal characters on the information recognition process of adjacent characters, thereby improving the efficiency of the information recognition process.

[0046] Furthermore, in this invention, for character blocks containing abnormal character images, the character recognition method is determined based on the abnormality key coefficient of the character block. For character blocks with a small abnormality key coefficient, the character information of the abnormal character can be determined based on the character information in character blocks of the same information category within the current production cycle. For character blocks with a large abnormality key coefficient, suspected characters are obtained by comparing the acquired character features, and the character information can be further determined through product quality matching. This invention improves the accuracy of information recognition results. Attached Figure Description

[0047] Figure 1 This is a module connection diagram of the visual processing and detection system for dishes according to the present invention;

[0048] Figure 2 This is a block diagram of the module structure of the character analysis module of the present invention;

[0049] Figure 3 This is a flowchart illustrating how the present invention determines the category of a corresponding character image based on the acquisition quality coefficient;

[0050] Figure 4 This is a flowchart illustrating how the processing method is determined based on the proportion of abnormal characters in this invention. Detailed Implementation

[0051] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0052] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0053] It should be noted that in the description of this invention, the terms "upper", "lower", "left", "right", "inner", "outer", etc., which indicate directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.

[0054] Furthermore, it should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0055] Please see Figures 1 to 4 As shown, the present invention provides a visual processing and inspection system for dishes, comprising:

[0056] The data acquisition and analysis module is used to perform quality inspection and key label image acquisition on randomly selected target products, and to detect the acquisition quality coefficient of each character image within the key label image, and determine the category of the character image based on the acquisition quality coefficient.

[0057] The key analysis module, which is connected to the acquisition and analysis module, is used to respond to whether there are abnormal character images in the key label image, so as to determine whether to perform abnormal key coefficient analysis on each character block in the key label.

[0058] The data analysis module, which is connected to the acquisition and analysis module, is used to determine the processing method based on the proportion of abnormal characters corresponding to the processing conditions. This method involves detecting the contrast reference value and the proportion of reflective areas of the key label image, or detecting the difference in the range of ink marks in each abnormal character area.

[0059] A secondary processing module, which is connected to the acquisition and analysis module and the data analysis module respectively, is used to determine the character area to be adjusted and perform area adjustment, as well as the contrast reference value and the proportion of reflective area corresponding to the processing conditions, so as to determine the image processing method accordingly.

[0060] The character analysis module is connected to the acquisition and analysis module, the secondary processing module, and the data analysis module, respectively. It is used to respond to the abnormal key coefficients corresponding to the set conditions, and to determine the character recognition method accordingly, either by extracting features from each abnormal character image within the character block, or by determining the character information of the abnormal character image based on the character information of character blocks of the same information category.

[0061] In this invention, the application scenario is the inspection of pre-prepared food products. The target inspection product is the completed pre-prepared food product. The key label is the information printed on the packaging of the pre-prepared food product. The information on the key label represents the production batch of the corresponding pre-prepared food product and other relevant information. The user randomly selects the target inspection product and performs quality inspection based on the selected target inspection product, while simultaneously acquiring the key label image. The key label image is an image of the complete printed area of ​​the key label.

[0062] Specifically, the data acquisition and analysis module performs quality testing on randomly selected target products to obtain quality testing results.

[0063] For a single target detection product,

[0064] If the acquisition and analysis module responds to the acquisition condition that the detection result of any type of detection is not within the preset result range, then it is determined that key label image acquisition will be performed for the target detection product.

[0065] The quality testing process for the target product includes several types of testing, including but not limited to testing the packaging sealing of the target product and testing the contents for microorganisms. For a single type of test, if the value of the test result is less than or equal to the maximum value of the preset result range and greater than the minimum value of the preset result range, then the test result of that type of test is determined to be within the preset result range. If the value of the test result is greater than the maximum value of the preset result range or less than or equal to the minimum value of the preset result range, then the test result of that type of test is determined to be outside the preset result range. How to set the preset result range for the test results of each type of test is a topic easily understood by those skilled in the art and will not be elaborated here.

[0066] Specifically, the acquisition and analysis module detects the acquisition quality coefficient of each character image within the key label image of the target product and determines the category of the corresponding character image based on the acquisition quality coefficient.

[0067] For a single character image, the image analysis module responds with the classification condition that if the acquisition quality coefficient is less than or equal to the preset acquisition quality coefficient, then the character image is determined to be an abnormal character image.

[0068] If the classification condition of the image analysis module is that the acquisition quality coefficient is greater than the preset acquisition quality coefficient, then the character image is determined to be a standard character image.

[0069] The acquisition quality coefficient is determined based on the similarity between the character image and each reference character image.

[0070] The character images are categorized into abnormal character images and standard character images, with characters in abnormal character images being designated as abnormal characters and characters in standard character images being designated as standard characters.

[0071] This invention divides key tag images into character regions, with each region containing only a single character. The image corresponding to each character region is recorded as the character image for that character. For a single character image, the acquisition quality coefficient is positively correlated with the baseline similarity of that character image. The baseline similarity is the maximum similarity between the character image and all baseline character images. Users set a baseline character image for each possible character within the key tag. Users can record images from historical records of various character information that are uninterrupted or missing, with clear character edges, and without blurring or distortion, as the baseline character image for the corresponding character. The characters in this invention may be numbers or letters. The types include, but are not limited to, 1, 2, 3, A, and B. Users can set the value of the preset acquisition quality coefficient according to actual needs and historical records. The higher the accuracy requirement of the user for the information recognition results of the key label image, the larger the value of the preset acquisition quality coefficient. A method for setting the value of the preset acquisition quality coefficient is provided, which takes the minimum value of the acquisition quality coefficient of each character image that can be directly identified in the historical records that meets the user's accuracy requirements for the information recognition results of the key label image as the preset acquisition quality coefficient. How to detect the similarity between the character image and each benchmark character image is a content that is easy for those skilled in the art to understand, and will not be elaborated here.

[0072] Specifically, if the analysis condition of the key analysis module is that there are abnormal character images within the key label image of the target detection product, then it will determine to perform abnormal key coefficient analysis on each character block within the key label, including:

[0073] According to preset rules, the character information within the key tag is divided into character blocks, and the proportion of abnormal character images in each character block and the key coefficient of each abnormal character in each character block are detected. The abnormal key coefficient of the corresponding character block is determined based on the proportion of abnormal character images in each character block and the average key coefficient.

[0074] The abnormal key coefficient is positively correlated with the proportion of abnormal character images and the average key coefficient.

[0075] Specifically, the key analysis module determines the key coefficient of the abnormal character at the corresponding character position based on the character repetition coefficient of each character position in the character block of the same information category within the current production cycle, and records the key coefficient of each abnormal character in the character block as the average key coefficient of the corresponding character block.

[0076] The critical coefficient of the abnormal character is negatively correlated with the character repetition coefficient at the corresponding character position.

[0077] The preset rules are adaptively set according to the printing rules of key labels in the actual work scenario. The preset rules can clearly indicate the setting order of character blocks of each information category. The character blocks obtained according to the preset rules can only completely represent one type of information. The information categories of the character blocks include, but are not limited to, year, month and date. For example, if the printing rule of the key label is "year, month and date", and a key label is "2024 09 30", then the key label can be divided into three character blocks: "2024", "09" and "30". For a single character block, the percentage of abnormal character images = the number of abnormal character images / the number of character images in the character block.

[0078] Users can set the production cycle duration according to actual needs. The higher the accuracy requirement of the key tag image information recognition result, the shorter the production cycle duration. One production cycle value is provided, which is 24 hours. The information categories of the character block include, but are not limited to, year, month, and date. For any character position of a single character block, the character repetition coefficient is the maximum number of character repetitions at that position / the number of different key tags in the production cycle. The number of character repetitions is the number of times each type of character information appears at that character position in the current cycle.

[0079] Specifically, the processing condition for the data analysis module to respond is that the proportion of abnormal characters in the key label image is greater than the preset proportion of abnormal characters. Then, the secondary processing module will detect the contrast reference value and the proportion of reflective area of ​​the key label image.

[0080] The image processing condition responded by the secondary processing module is that the contrast reference value of the key label image is greater than the preset contrast reference value and the proportion of reflective area is less than or equal to the preset proportion of reflective area. Then, it is determined to perform grayscale transformation on the key label image, divide the key label image into several sub-image regions, and detect the optimal threshold for each sub-image region.

[0081] If the image processing condition of the secondary processing module is that the contrast reference value is less than or equal to the preset contrast reference value or the proportion of reflective area is greater than the preset proportion of reflective area, then it is determined to perform similarity analysis on each pixel in the key label image.

[0082] Wherein, the abnormal character ratio of the key label image = the number of abnormal character images in the key label image / the number of character images in the key label image. Users can set the value of the preset abnormal character ratio according to actual needs and historical records. The higher the user's requirement for the accuracy of the information recognition result of the key label image, the smaller the value of the preset abnormal character ratio. A method for setting the value of the preset abnormal character ratio is provided, in which the historical records of image processing for the key label image are recorded as the ratio reference records, and the minimum value of the abnormal character ratio of each key label image in the ratio reference records that meets the user's requirement for the accuracy of the information recognition result of the key label image is recorded as the preset abnormal character ratio. A value of the preset abnormal character ratio is provided, and the value of the preset abnormal character ratio is 0.6.

[0083] The contrast reference value is the absolute value of the difference between the maximum and minimum grayscale values ​​of the key label image. The reflective area ratio is the sum of the areas of all reflective areas / the area of ​​the key label image. The reflective area is the area with a high brightness value formed by light reflection in the image. How to determine the contrast reference value and the reflective area is easy for those skilled in the art to understand and will not be elaborated here.

[0084] The preset contrast reference value and the preset reflective area ratio can be set by the user according to actual needs and historical records. The higher the user's requirement for the accuracy of the information recognition results of the key label image, the larger the preset contrast reference value and the smaller the preset reflective area ratio. A method for setting the preset contrast reference value is provided, in which the historical records of determining the filter value based on the similarity between each pixel are recorded as image reference records, and the maximum value of the contrast reference value of each key label image in the image reference records that meets the user's requirement for the accuracy of the information recognition results of the key label image is recorded as the preset contrast reference value. A method for setting the preset reflective area ratio is provided, in which the minimum value of the reflective area ratio of each key label image in the image reference records that meets the user's requirement for the accuracy of the information recognition results of the key label image is recorded as the preset reflective area ratio.

[0085] If the contrast reference value of the key label image is greater than the preset contrast reference value and the proportion of the reflective area is less than or equal to the preset proportion of the reflective area, grayscale transformation is performed on the key label image. The grayscale transformation process includes transforming the grayscale value of each pixel in the input image. For a single pixel, the transformed grayscale value is the product of the original grayscale value raised to the power of γ and the preset transformation coefficient. γ is the transformation factor, which controls the degree of adjustment during the transformation process. Users can set the value of the transformation factor according to the original grayscale value of the pixel. By performing grayscale transformation on each pixel in the key label image, the clarity of the key label image can be effectively improved to solve the problems of reflective areas and low image contrast. After completing the grayscale transformation, local threshold analysis is performed on the key label image. In this invention, an adaptive binary method is used for local threshold analysis to segment the key label graphic. How to set the transformation factor according to the original grayscale value of the pixel, how to set the preset transformation coefficient, and how to use the adaptive binary method for local threshold analysis are all content that is easy for those skilled in the art to understand and will not be elaborated here.

[0086] If the contrast reference value is less than or equal to the preset contrast reference value, or the proportion of reflective area is greater than the preset proportion of reflective area, the similarity of each pixel in the key label image is calculated. The search window and matching window are set according to the preset search window side length and the preset matching window side length. The similarity analysis process for a single pixel includes: extracting a matching window centered on that pixel and extracting matching windows of the same size within the key label image based on the search window; calculating the similarity between the two matching windows based on Euclidean distance; and determining weight coefficients based on the obtained similarity. The magnitude of the weight coefficients is positively correlated with the similarity. All weight coefficients are then assigned... The process involves normalization to ensure that the sum of the weight coefficients is 1. The weighted average of all pixels within the search window is calculated based on the normalized weight coefficients, and this calculated weighted average is assigned to the currently processed pixel. Users can adaptively set the preset search window side length and preset matching window side length according to their actual working conditions. In this invention, the weight coefficients are determined based on similarity using a Gaussian function. How to set the preset search window side length and preset matching window side length, how to calculate the similarity between two matching windows based on Euclidean distance, and how to determine the weight coefficients based on a Gaussian function are all easily understood by those skilled in the art and will not be elaborated upon here.

[0087] Specifically, the processing condition of the data analysis module is that if the proportion of abnormal characters in the key label image is less than or equal to the preset proportion of abnormal characters, then it is determined to detect the difference in the range of character ink marks in each abnormal character region.

[0088] For a single abnormal character region, the adjustment condition of the secondary processing module is that the difference value of the character ink range is greater than the preset difference value of the character ink range, and then the adjacent character regions of the abnormal character region are determined to be the character regions to be adjusted.

[0089] For a single character region to be adjusted, the border and adjustment distance of the region to be adjusted are determined based on the adjacent directions of its neighboring abnormal character regions and the difference in the character ink range.

[0090] Specifically, for a single abnormal character region, the character ink range difference value is the difference between the character ink width of the abnormal character region and the baseline character ink width. In this invention, several detection points are randomly selected for characters within the abnormal character region, and the character line width at each detection point is used for detection. The character ink width is the average of the character line widths obtained from different detection points, and the baseline character ink width is the average of the character ink widths of characters in each baseline character image. Users can set the preset character ink range difference value according to actual needs and historical records. The higher the accuracy requirement of the user for the information recognition result of the key label image, the smaller the preset character ink range difference value. A method for setting the preset character ink range difference value is provided, in which records of adjacent character regions of the abnormal character region that have not undergone character region adjustment are recorded as ink reference records, and the minimum value of the character ink range difference value of each abnormal character region in the ink reference records that meet the user's accuracy requirement for the information recognition result of the key label image is recorded as the preset character ink range difference value. For a single abnormal character region, there is no character region between the abnormal character region and any of its adjacent character regions.

[0091] For a single character region to be adjusted, the border and adjustment distance of the region to be adjusted are determined based on the adjacent direction of its neighboring abnormal character regions and the difference value of the character ink range. The adjacent direction is the direction of the character region to be adjusted relative to the neighboring abnormal character region. If the character region to be adjusted is located to the right of its neighboring abnormal character region, the left border of the character region to be adjusted is adjusted, that is, the left border is adjusted to the right. If the character region to be adjusted is located above its neighboring abnormal character region, the lower border of the character region to be adjusted is adjusted, that is, the lower border is adjusted upward. The adjustment distance is positively correlated with the difference value of the character ink range.

[0092] Specifically, the character analysis module responds to the recognition conditions and determines to perform character recognition for each abnormal character image;

[0093] For a single character block containing an abnormal character image,

[0094] The character analysis module responds when the abnormal key coefficient of the character block is greater than the preset abnormal key coefficient. Then, it determines that the first analysis unit performs feature extraction on each abnormal character image within the character block.

[0095] The character analysis module responds when the abnormal key coefficient of the character block is less than or equal to the preset abnormal key coefficient. Then, the second analysis unit determines the character information of the abnormal character image in the character block based on the character information of character blocks of the same information category in the current production cycle.

[0096] The recognition condition is that all regions of characters to be adjusted have completed the region adjustment.

[0097] The second analysis unit determines the character information of the abnormal character image in the character block based on the character information of the character block of the same information category in the current production cycle. For a single abnormal character, the character information corresponding to the maximum number of character repetitions at the character position of the character is recorded as the character information of the abnormal character.

[0098] Specifically, the first analysis unit responds to the feature extraction conditions and performs feature extraction on the images of each abnormal character within the character block;

[0099] For a single abnormal character image, the first analysis unit compares the extracted character features with the character features of each benchmark character image to determine the feature overlap, and determines the suspected character based on the feature overlap.

[0100] The feature extraction condition is determined by the character analysis module, which determines that the first analysis unit performs feature extraction on each abnormal character image within the character block.

[0101] In this invention, several character features are set for the reference character images corresponding to various types of character information. The feature overlap degree between a single abnormal character image and any reference character image is equal to the number of identical character features in the two character images divided by the number of character features in the reference character image. Characters corresponding to reference character images with a feature overlap degree greater than a preset feature overlap degree are recorded as suspected characters. Users can set the preset feature overlap degree value according to actual needs and historical records. The higher the accuracy requirement of the user for the information recognition results of the key label image, the larger the preset feature overlap degree value. A method for setting the preset feature overlap degree is provided, which records the average feature overlap degree of the reference character images corresponding to each suspected character in the historical records that meets the user's accuracy requirements for the information recognition results of the key label image as the preset feature overlap degree. How to set character features for the reference character images corresponding to various types of character information and how to compare character features are contents that are easy for those skilled in the art to understand, and will not be elaborated here.

[0102] Specifically, if the first analysis unit responds to the matching condition, it determines to perform product quality matching on the corresponding character block. The product quality matching process includes:

[0103] Based on each suspected character, several tag information to be determined were identified.

[0104] For a single undetermined label information, the difference in quality parameters between the target test product to which the undetermined label information belongs and the associated test product is detected.

[0105] Priority coefficients are set for the label information to be determined based on the degree of difference in quality parameters;

[0106] The matching condition is that all abnormal character images within the character block have been identified as suspected characters.

[0107] Specifically, for character blocks for which suspected characters have been identified in all existing abnormal character images, the suspected characters of each abnormal character image are combined to obtain the tag information to be determined. For example, there is a character block containing abnormal character images, the character block is "2xxx", and the second, third and fourth character images are abnormal character images. The suspected characters of the abnormal character image in the second position are 3 and 6, the suspected characters of the abnormal character image in the third position are 1 and 7, and the suspected characters of the abnormal character image in the fourth position are 9 and 8. Therefore, the tag information to be determined that can be obtained from this character block includes, but is not limited to, "2319", "2318" and "2378".

[0108] For a single undetermined label information, it is combined with other character blocks to obtain the undetermined key label information. The associated detection product is the target detection product whose production time is closest to that represented by the undetermined key label information and which has quality detection results. The quality parameter difference degree is the sum of the products of the result difference value of each type of detection and the influence coefficient corresponding to each type of detection. Users can set the value of the influence coefficient corresponding to the result difference value of each type of detection according to actual needs. If the quality detection process for the target detection product includes Salmonella content detection and nitrite content detection, the quality parameter difference degree = Salmonella content difference value × influence coefficient corresponding to the Salmonella content difference value + The influence coefficient corresponding to the difference in nitrite content is calculated by multiplying the difference in nitrite content by the difference in Salmonella content. The difference in Salmonella content is the absolute value of the difference in Salmonella content detected by two target detection products. The difference in nitrite content is the absolute value of the difference in nitrite content detected by two target detection products. A value of 0.5 is provided for the influence coefficients corresponding to the difference in Salmonella content and the difference in nitrite content. The priority coefficient is negatively correlated with the difference in quality parameters, and the label information to be determined with a higher priority coefficient is set as the character information of the corresponding character block.

[0109] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

[0110] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A visual processing and inspection system for dishes, characterized in that, include: The data acquisition and analysis module is used to perform quality inspection and key label image acquisition on randomly selected target products, and to detect the acquisition quality coefficient of each character image within the key label image, and determine the category of the character image based on the acquisition quality coefficient. The key analysis module, which is connected to the acquisition and analysis module, is used to respond to whether there are abnormal character images in the key label image, so as to determine whether to perform abnormal key coefficient analysis on each character block in the key label. The data analysis module, which is connected to the acquisition and analysis module, is used to determine the processing method based on the proportion of abnormal characters corresponding to the processing conditions. This method involves detecting the contrast reference value and the proportion of reflective areas of the key label image, or detecting the difference in the range of ink marks in each abnormal character area. A secondary processing module, which is connected to the acquisition and analysis module and the data analysis module respectively, is used to determine the character area to be adjusted and perform area adjustment, as well as the contrast reference value and the proportion of reflective area corresponding to the processing conditions, so as to determine the image processing method accordingly. The processing condition for the data analysis module to respond is that the proportion of abnormal characters in the key label image is greater than the preset proportion of abnormal characters. Then, the secondary processing module will detect the contrast reference value and the proportion of reflective area of ​​the key label image. The image processing condition responded by the secondary processing module is that the contrast reference value of the key label image is greater than the preset contrast reference value and the proportion of reflective area is less than or equal to the preset proportion of reflective area. Then, it is determined to perform grayscale transformation on the key label image, divide the key label image into several sub-image regions, and detect the optimal threshold for each sub-image region. The image processing condition responded by the secondary processing module is that the contrast reference value is less than or equal to the preset contrast reference value or the proportion of reflective area is greater than the preset proportion of reflective area. Then, it is determined to perform similarity analysis on each pixel in the key label image. The processing condition for the data analysis module to respond is that the proportion of abnormal characters in the key label image is less than or equal to the preset proportion of abnormal characters, and then it is determined to detect the difference in the range of character ink marks in each abnormal character region. For a single abnormal character region, the adjustment condition of the secondary processing module is that the difference value of the character ink range is greater than the preset difference value of the character ink range, and then the adjacent character regions of the abnormal character region are determined to be the character regions to be adjusted. For a single character region to be adjusted, the border and adjustment distance of the region to be adjusted are determined based on the adjacent directions of its neighboring abnormal character regions and the difference in the character ink range. The character analysis module, connected to the acquisition and analysis module, the secondary processing module, and the data analysis module, responds to anomaly key coefficients corresponding to set conditions to determine the character recognition method: either extracting features from each abnormal character image within a character block, or determining the character information of the abnormal character image based on the character information of character blocks of the same information category. Specifically, The character analysis module responds when there is an abnormal key coefficient of any character block that is greater than the preset abnormal key coefficient. Then, the first analysis unit is determined to extract features from each abnormal character image within that character block. The character analysis module responds under the condition that there exists an abnormal key coefficient of any character block that is less than or equal to the preset abnormal key coefficient. Then, the second analysis unit determines the character information of the abnormal character image in the character block based on the character information of character blocks of the same information category in the current production cycle.

2. The food visualization processing and inspection system according to claim 1, characterized in that, The data acquisition and analysis module performs quality testing on randomly selected target products to obtain quality testing results. For a single target detection product, If the acquisition and analysis module responds to the acquisition condition that the detection result of any type of detection is not within the preset result range, then it is determined that key label image acquisition will be performed for the target detection product.

3. The food visualization processing and inspection system according to claim 2, characterized in that, The acquisition and analysis module detects the acquisition quality coefficient of each character image within the key label image of the target product and determines the category of the corresponding character image based on the acquisition quality coefficient. For a single character image, the image analysis module responds with the classification condition that if the acquisition quality coefficient is less than or equal to the preset acquisition quality coefficient, then the character image is determined to be an abnormal character image. If the classification condition of the image analysis module is that the acquisition quality coefficient is greater than the preset acquisition quality coefficient, then the character image is determined to be a standard character image. The acquisition quality coefficient is determined based on the similarity between the character image and each reference character image.

4. The food visualization processing and inspection system according to claim 3, characterized in that, The critical analysis module responds when the analysis condition is that there are abnormal character images within the critical label image of the target product. Then, it determines to perform abnormal critical coefficient analysis on each character block within the critical label, including: According to preset rules, the character information within the key tag is divided into character blocks, and the proportion of abnormal character images in each character block and the key coefficient of each abnormal character in each character block are detected. The abnormal key coefficient of the corresponding character block is determined based on the proportion of abnormal character images in each character block and the average key coefficient. The abnormal key coefficient is positively correlated with the proportion of abnormal character images and the average key coefficient.

5. The food visualization processing and inspection system according to claim 4, characterized in that, The key analysis module determines the key coefficient of the abnormal character at the corresponding character position based on the character repetition coefficient of each character position in the character block of the same information category within the current production cycle, and records the key coefficient of each abnormal character in the character block as the average key coefficient of the corresponding character block. The key coefficient of the abnormal character is negatively correlated with the character repetition coefficient at the corresponding character position.

6. The food visualization processing and inspection system according to claim 5, characterized in that, The character analysis module responds to the recognition conditions and then determines to perform character recognition for each abnormal character image; The recognition condition is that all regions of characters to be adjusted have completed the region adjustment.

7. The food visualization processing and inspection system according to claim 6, characterized in that, The first analysis unit responds to the feature extraction conditions and extracts features from the images of each abnormal character within the character block; For a single abnormal character image, the first analysis unit compares the extracted character features with the character features of each benchmark character image to determine the feature overlap, and determines the suspected character based on the feature overlap. The feature extraction condition is determined by the character analysis module, which determines that the first analysis unit performs feature extraction on each abnormal character image within the character block.

8. The food visualization processing and inspection system according to claim 7, characterized in that, If the first analysis unit responds to the matching condition, it determines to perform product quality matching on the corresponding character block. The product quality matching process includes: Based on each suspected character, several tag information to be determined were identified. For a single undetermined label information, the difference in quality parameters between the target test product to which the undetermined label information belongs and the associated test product is detected. Priority coefficients are set for the label information to be determined based on the degree of difference in quality parameters; The matching condition is that all abnormal character images within the character block have been identified as suspected characters.

Images