Machine vision based tray sorting system and method

Abstract

This application relates to the field of machine vision technology and discloses a machine vision-based tray sorting system and method. The system includes an image acquisition module, a rule storage module, a matching analysis module, a confidence assessment module, a result judgment module, a rule adjustment module, a loop control module, and a sorting execution module. The method corresponds to this system. This application establishes an association assessment between recognition rules and conveying parameters through the matching analysis module, and combined with a confidence level iterative verification mechanism, achieves dynamic optimization of the rules. The core of this embodiment lies in incorporating real-time parameters such as conveying cycle time and conveying speed into the recognition logic, solving the problem that fixed rules are difficult to adapt to dynamic conveying scenarios, significantly improving the reliability of the recognition results and the system's adaptability, and reducing the need for manual intervention.

Description

Technical Field

[0001] This application relates to the field of machine vision technology, specifically a machine vision-based tray sorting system and method. Background Technology

[0002] In existing machine vision-based tray sorting systems, the recognition rules are often fixed and do not fully correlate with dynamic parameters such as cycle time and speed during the conveying process. This can easily lead to insufficient reliability of the recognition results due to parameter mismatch. Furthermore, there is a lack of a mechanism to dynamically optimize the rules based on the real-time matching status, requiring frequent manual intervention and adjustment, resulting in poor adaptability.

[0003] Chinese invention patent CN114219975B discloses a method for evaluating the confidence of artificial intelligence image recognition results. Although it discloses an iterative optimization method for image recognition confidence, it does not include the matching degree between recognition rules and parameters such as conveying cycle time and speed in the confidence evaluation dimension, making it difficult to cope with the impact of changes in conveying status on recognition accuracy. Furthermore, Chinese invention patent CN115569869B discloses an automatic detection and sorting device for optical device wire bonding. Although it discloses a tray sorting mechanism based on machine vision, its recognition rule adjustment does not form a closed-loop optimization logic based on parameter matching degree, making it unable to dynamically adapt to different conveying conditions and limiting recognition stability.

[0004] In conclusion, there is an urgent need for a new machine vision-based solution for tray sorting. Summary of the Invention

[0005] The purpose of this application is to provide a machine vision-based tray sorting system and method to solve the technical problems mentioned in the background art.

[0006] To achieve the above objectives, this application discloses the following technical solutions: In a first aspect, this application discloses a machine vision-based tray sorting system, which includes: The image acquisition module is configured to acquire the identification image of the material tray; The rule storage module is configured to store at least one set of identification rules; The matching analysis module is configured to analyze the matching degree between the current identification rule and the current conveying parameters, including the conveying cycle time and the conveying speed. The confidence assessment module is configured to: analyze the confidence of the initial recognition result based on the matching degree, wherein the initial recognition result is obtained by analyzing the identifier image based on the current recognition rules, and the confidence is determined based on the feature consistency analysis between the matching degree and the initial recognition result; The result determination module is configured to: determine whether to output the initial recognition result based on the confidence level; output the result when the confidence level meets the preset output conditions, otherwise do not output the result. The rule adjustment module is configured to: adjust the current recognition rule to obtain the new recognition rule when the initial recognition result is not output; The loop control module is configured to repeatedly analyze the identification image and evaluate the confidence level using the new recognition rule until the confidence level meets the preset output conditions and then output the corresponding final recognition result. The sorting execution module is configured to perform sorting operations on the pallets based on the output of the final identification result.

[0007] Preferably, the matching analysis module analyzes the matching degree by including the following steps: A1: Get the processing time of the current recognition rule per cycle; A2: Calculate the ratio of the single processing time to the conveying cycle time to obtain the time matching coefficient.

[0008] Preferably, when the matching analysis module analyzes the matching degree, it further includes the following steps: A3: Obtain the minimum image clarity requirement corresponding to the current recognition rule; A4: Calculate the difference between the minimum image sharpness requirement and the actual image blur amount associated with the transmission speed to obtain the sharpness matching coefficient; A5: The time matching coefficient and the sharpness matching coefficient are weighted and fused, and the weighted fused value is defined as the matching degree.

[0009] Preferably, the actual image blur is the product of the transport speed and the exposure time of the image acquisition module, and the minimum image clarity requirement is the maximum allowable blur preset by the current recognition rule.

[0010] Preferably, the identification rules include preprocessing rules, feature extraction rules, and classification rules; The preprocessing rules include at least the filtering method and grayscale conversion parameters; The feature extraction rules include at least the type of feature to be extracted and the feature quantization dimension; The classification rules include at least the classification model type and the category determination threshold.

[0011] Preferably, the confidence assessment module analyzes the confidence level by including the following steps: B1: Obtain the feature consistency parameter of the initial recognition result. The feature consistency parameter is obtained by: processing the same identified image continuously at least three times using the current recognition rule, extracting the feature vector obtained from each processing, analyzing the corresponding similarity between all feature vectors, and defining the normalized value of the similarity as the feature consistency parameter. B2: The matching degree is weighted and fused with the feature consistency parameter of the initial recognition result, and the weighted fusion value is defined as the confidence degree.

[0012] Preferably, when the rule adjustment module adjusts the current recognition rule, it includes adjusting at least one of the preprocessing rule, adjusting the feature extraction rule, and adjusting the classification rule.

[0013] Preferably, the sorting execution module includes at least a gripping mechanism, a sorting conveying mechanism, and a sorting and storage mechanism; The gripping mechanism is equipped with an actuator that adapts to the material tray; The sorting and conveying mechanism includes at least two sorting channels, each corresponding to a different sorting category; The sorting and storage mechanism has storage areas that correspond one-to-one with each of the sorting channels.

[0014] Preferably, when adjusting the current recognition rules, the rule adjustment module first analyzes the specific influencing factors that cause the confidence level of the initial recognition result to not meet the output conditions, and then adjusts the corresponding type of recognition rules for the specific influencing factors, including: When the specific influencing factor is that the time matching coefficient is less than a preset time threshold, the feature quantization dimension in the feature extraction rule is adjusted. When the specific influencing factor is that the sharpness matching coefficient is less than the preset sharpness threshold, the filtering method in the preprocessing rule is adjusted. When the specific influencing factor is that the feature consistency parameter is less than the preset consistency threshold, the category determination threshold in the classification rule is adjusted.

[0015] Secondly, this application discloses a machine vision-based tray sorting method, applied to the machine vision-based tray sorting system described above. The method includes the following steps: S1: Collect the identification image of the material tray; S2: Store at least one set of identification rules; S3: Analyze the matching degree between the current identification rules and the current conveying parameters, including the conveying cycle time and the conveying speed; S4: Analyze the confidence level of the initial recognition result based on the matching degree. The initial recognition result is obtained by analyzing the identifier image based on the current recognition rules. The confidence level is determined based on the feature consistency analysis between the matching degree and the initial recognition result. S5: Determine whether to output the initial recognition result based on the confidence level. Output the result when the confidence level meets the preset output conditions, otherwise do not output the result. S6: Without outputting the initial recognition result, adjust the current recognition rule to obtain the new recognition rule; S7: Repeatedly analyze the identified image and evaluate the confidence level using the new recognition rule until the confidence level meets the preset output conditions, and then output the corresponding final recognition result; S8: Perform a sorting operation on the pallet based on the final identification result output.

[0016] Beneficial effects: The machine vision-based pallet sorting system and method of this application establishes a dynamic correlation between recognition rules and conveying parameters through a matching analysis module, and achieves multi-dimensional verification of recognition results by combining confidence assessment, thus solving the problem that fixed rules in the prior art are difficult to adapt to changes in conveying status. Adaptability is quantified by time matching coefficient and sharpness matching coefficient, and confidence assessment is optimized by integrating feature consistency parameters, making the recognition results more reliable. The rule adjustment module accurately optimizes the corresponding rules based on specific influencing factors, reducing the number of iterations and improving adaptive efficiency. The structured design of the sorting execution module ensures efficient implementation of recognition results, which improves the overall stability, recognition accuracy and dynamic adaptability of the sorting system, reduces manual intervention, and is suitable for multi-condition pallet sorting scenarios. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A structural block diagram of a machine vision-based tray sorting system provided in an embodiment of this application; Figure 2 A flowchart illustrating a machine vision-based tray sorting method provided in an embodiment of this application. Detailed Implementation

[0019] The technical solutions in the embodiments of this application will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0020] In this document, the term "comprising" is intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0021] Example 1 In existing machine vision-based pallet sorting systems, the recognition rules are mostly fixed presets, lacking correlation with dynamic parameters during the conveying process, such as conveying cycle time and speed. When the conveying cycle time increases or the conveying speed changes, the fixed rules often lead to recognition errors due to insufficient processing efficiency or mismatched image quality. Furthermore, there is a lack of a mechanism for dynamically optimizing rules based on real-time conditions, requiring repeated manual adjustments and exhibiting poor adaptability. Therefore, this embodiment provides a machine vision-based pallet sorting system that can dynamically adapt recognition rules to conveying parameters.

[0022] like Figure 1 As shown, this embodiment discloses a machine vision-based tray sorting system, which includes: The image acquisition module is configured to acquire images of the markings on the material tray. In this embodiment, existing image acquisition equipment, such as an industrial camera, is used to acquire images of the markings placed on the material tray. These markings may be, but are not limited to, barcode stickers.

[0023] The rule storage module is configured to store at least one set of recognition rules, including preprocessing rules, feature extraction rules, and classification rules.

[0024] The matching analysis module is configured to analyze the matching degree between the current identification rules and the current conveying parameters, including the conveying cycle time and the conveying speed.

[0025] The confidence assessment module is configured to analyze the confidence of the initial recognition result based on the matching degree. The initial recognition result is obtained by analyzing the labeled image based on the current recognition rules.

[0026] The result determination module is configured to determine whether to output the initial recognition result based on the confidence level. The result is output when the confidence level meets the preset output conditions, and not output otherwise.

[0027] The rule adjustment module is configured to adjust the current recognition rules to obtain new recognition rules when the initial recognition result is not output.

[0028] The loop control module is configured to repeatedly analyze the labeled image and evaluate the confidence level using new recognition rules until the confidence level meets the preset output conditions, and then output the corresponding final recognition result.

[0029] The sorting execution module is configured to perform sorting operations on the pallets based on the final identification results output.

[0030] It should be noted that in this embodiment, the multiple sets of identification rules are determined based on the actual needs of tray identification and its corresponding conveying parameters; furthermore, the correspondence between the identification rules and the conveying parameters can be obtained based on historical identification data.

[0031] Based on the above, the association evaluation between the identification rules and the transportation parameters is established through the matching analysis module. Combined with the confidence iteration verification mechanism, the dynamic optimization of the rules is realized. The core of this embodiment is to incorporate real-time parameters such as transportation cycle time and transportation speed into the identification logic, which solves the problem that fixed rules are difficult to adapt to dynamic transportation scenarios, significantly improves the reliability of the identification results and the system's adaptability, and reduces the need for manual intervention.

[0032] In the matching analysis of identification rules and conveying parameters, the adaptability of the time dimension is crucial. If the processing time of the identification rules exceeds the conveying cycle time, the tray may have moved before identification is completed, leading to missed detections. Conversely, if the processing time is too short, the accuracy may be reduced due to oversimplification of the identification logic. Existing technologies do not clearly quantify the time matching relationship between the two, resulting in ambiguous adaptability assessments. Therefore, this embodiment proposes to further improve the adaptability of the time dimension by calculating a time matching coefficient.

[0033] Specifically, when analyzing the matching degree, the matching analysis module includes the following steps: A1: Get the processing time of the current recognition rule per cycle; A2: Calculate the ratio of single processing time to conveying cycle time to obtain the time matching coefficient.

[0034] Based on the above, by calculating the ratio of single processing time to conveying cycle time, time adaptability is quantified into a time matching coefficient, allowing the matching relationship between the processing efficiency of the identification rules and the conveying rhythm to be directly measured. This design avoids missed detections or accuracy loss due to time mismatch, provides a clear time dimension basis for subsequent confidence assessment, and ensures that the identification process is synchronized with the conveying rhythm.

[0035] In practical applications, evaluating adaptability solely based on time matching coefficients is insufficient. Image sharpness is fundamental to machine vision recognition, and transmission speed directly impacts image blurriness; faster speeds result in more blurred images. If the sharpness requirements of the recognition rules do not match the actual amount of blur, feature extraction errors will occur. Existing technologies do not correlate transmission speed with the sharpness requirements of the recognition rules, leading to a one-sided matching evaluation. This embodiment proposes further improving image sharpness adaptability by calculating the sharpness dimension.

[0036] Specifically, when analyzing the matching degree, the matching analysis module also includes the following steps: A3: Obtain the minimum image clarity requirement corresponding to the current recognition rule; A4: Calculate the difference between the minimum image sharpness requirement and the actual image blur amount related to the transmission speed to obtain the sharpness matching coefficient; A5: Perform a weighted fusion of the time matching coefficient and the sharpness matching coefficient, and define the weighted fusion value as the matching degree.

[0037] By introducing a sharpness matching coefficient and weighting it with a time matching coefficient, a multi-dimensional evaluation of the recognition rules and transmission parameters is achieved. The time dimension ensures that processing efficiency is adapted, while the sharpness dimension ensures that image quality meets recognition requirements. The combination of the two makes the matching degree more consistent with the actual sorting scenario, providing a more comprehensive and accurate basis for confidence calculation and reducing misjudgments caused by single-dimensional evaluation.

[0038] The calculation of the sharpness matching coefficient depends on the accurate definition of the actual image blur amount and the minimum sharpness requirement of the recognition rule. If the meanings of these two are ambiguous or the calculation methods are unclear, the matching degree analysis results will be unreliable. For example, the definition of image blur in existing technologies often lacks quantitative standards, resulting in a lack of comparability of analysis results in different scenarios. This embodiment clarifies the definitions and calculation methods of both.

[0039] Specifically, the actual image blur is the product of the transmission speed and the exposure time of the image acquisition module, and the minimum image clarity requirement is the maximum allowable blur preset by the current recognition rules.

[0040] Based on the above, the actual image blur is defined as the product of the transmission speed and the exposure time, allowing the degree of blur to be directly quantified through physical parameters. The minimum sharpness requirement is defined as the maximum allowable blur, clarifying the bottom line for image quality in the recognition rules. This design ensures the consistency and repeatability of the sharpness matching coefficient calculation, improves the rigor of the matching degree analysis, and provides a unified standard for cross-scenario applications.

[0041] Recognition rules are the core logic of machine vision recognition, but if their composition is not clearly defined, the storage, retrieval, and adjustment of rules will lack specificity. In existing technologies, recognition rules are often general concepts, failing to distinguish key steps such as preprocessing and feature extraction. When rule adjustments are needed, it is difficult to pinpoint the specific optimization target, leading to low iteration efficiency. This embodiment provides a foundation for adjusting recognition rules by refining their composition and parameters.

[0042] Specifically, the identification rules include preprocessing rules, feature extraction rules, and classification rules; Preprocessing rules should include at least the filtering method and grayscale conversion parameters.

[0043] Feature extraction rules should include at least the type of feature to be extracted and the feature quantization dimension.

[0044] The classification rules should include at least the classification model type and the category determination threshold.

[0045] It should be noted that the preprocessing rules, feature extraction rules, and classification rules in this embodiment can be set based on the actual needs of tray sorting and in combination with existing identification rules.

[0046] Based on the above, this embodiment breaks down the recognition rules into preprocessing, feature extraction, and classification rules, and clarifies the core parameters of each part, such as filtering methods, feature types, and classification thresholds, making the structure of the recognition rules clearer. This structured design facilitates the system's accurate invocation of corresponding rules, and when adjustments are needed, it can directly target specific steps, such as optimizing only the feature extraction rules, avoiding blind modifications to the overall rules and significantly improving the efficiency and relevance of rule iteration.

[0047] Confidence level is a core indicator for determining the reliability of recognition results. If only the matching degree is considered in the evaluation without taking into account the stability of the recognition results themselves, recognitions with high matching degrees but large fluctuations in results may be judged as reliable, leading to missorting. Existing technologies often neglect the consistency of recognition results in confidence level evaluation, and the evaluation dimension is singular. This embodiment clarifies the specific analysis method of confidence level, combining matching degree and result stability to improve the quality of recognition results.

[0048] Specifically, the confidence assessment module analyzes confidence levels by including the following steps: B1: Obtain the feature consistency parameter of the initial recognition result. The feature consistency parameter is obtained by processing the same labeled image continuously at least three times using the current recognition rule, extracting the feature vector obtained from each processing, analyzing the corresponding similarity between all feature vectors, and defining the normalized value of the similarity as the feature consistency parameter. In this embodiment, the number of consecutive processing operations performed on the same identified image using the current recognition rules is determined by existing computing power allocation methods, based on actual computing power, ensuring that this parameter is allocated no less than three times. Furthermore, existing similarity analysis and normalization techniques are used to perform similarity analysis and normalization among all feature vectors.

[0049] B2: Weight the matching degree with the feature consistency parameter of the initial recognition result, calculate the weighted fusion value and define it as the confidence level.

[0050] By introducing a feature consistency parameter to quantify the stability of multiple recognitions of the same image, and then weighting and fusing it with the matching degree to calculate the confidence level, the evaluation reflects both the fit between the rules and the input parameters and the stability of the recognition results. This multi-dimensional evaluation avoids the limitations of a single indicator, significantly improves the accuracy of result judgment, and reduces erroneous outputs caused by accidental matching.

[0051] When the confidence level of the recognition result does not meet the output conditions, the recognition rules need to be adjusted. However, in existing technologies, rule adjustments often lack a clear target, which may lead to the blind modification of irrelevant parameters, resulting in ineffective optimization or over-adjustment. For example, instead of optimizing feature extraction, the classification rules are modified, reducing iteration efficiency. This embodiment further defines the specific target of rule adjustment.

[0052] Specifically, when the rule adjustment module adjusts the current recognition rules, it includes at least one or more of the following: adjusting preprocessing rules, adjusting feature extraction rules, and adjusting classification rules.

[0053] Based on the above, it is clear that the objects of rule adjustment are preprocessing rules, feature extraction rules, or classification rules, making the adjustment direction more focused. For example, when image noise affects recognition, the filtering method of the preprocessing rules can be directly optimized; when feature discrimination is insufficient, the feature extraction rules can be adjusted accordingly. This design avoids blind optimization, improves the efficiency of rule iteration, and ensures that each adjustment accurately targets the problem area.

[0054] Furthermore, rule adjustments must be based on the specific reasons for insufficient confidence levels; otherwise, the adjustments may be ineffective in improving recognition results. For example, low confidence levels due to insufficient time matching could lead to ineffective iterations if classification rules are adjusted incorrectly. Existing technologies often employ general strategies for rule adjustments without considering specific influencing factors, resulting in numerous iterations and low efficiency. This embodiment optimizes the adjustment logic to achieve targeted improvements.

[0055] Specifically, when the rule adjustment module adjusts the current recognition rules, it first analyzes the specific influencing factors that cause the initial recognition result confidence level to fail to meet the output conditions, and then adjusts the corresponding type of recognition rules for the specific influencing factors, including: When the specific influencing factor is that the time matching coefficient is less than the preset time threshold, the feature quantization dimension in the feature extraction rule is adjusted. When the specific influencing factor is that the sharpness matching coefficient is less than the preset sharpness threshold, the filtering method in the preprocessing rules is adjusted. When the specific influencing factor is that the feature consistency parameter is less than the preset consistency threshold, the category determination threshold in the classification rule is adjusted.

[0056] Based on the above, by first identifying the specific factors affecting the failure of location reliability, such as time matching, clarity matching, or feature consistency, and then adjusting the corresponding rules accordingly, such as optimizing the feature extraction dimension if time matching is insufficient, each adjustment directly targets the root cause of the problem. This logic of precise adjustment based on problem location significantly shortens the number of iterations required to achieve the required confidence level and improves the system's adaptive speed, making it particularly suitable for scenarios where transmission parameters change frequently.

[0057] In practical applications, the sorting execution module is crucial for translating identification results into physical actions. If its composition or function is unclear, the identification results cannot be accurately implemented. For example, a mismatch between the gripping mechanism and the pallet will lead to gripping failure, and insufficient sorting channels will prevent the differentiation of different types of pallets. This embodiment details the composition and function of the sorting execution module.

[0058] Specifically, the sorting execution module includes at least a gripping mechanism, a sorting conveying mechanism, and a sorting and storage mechanism; The gripping mechanism is equipped with an actuator adapted to the material tray. In this embodiment, the gripping mechanism can be, but is limited to, existing sorting robots.

[0059] The sorting and conveying mechanism includes at least two sorting channels, each corresponding to a different sorting category.

[0060] The sorting and storage mechanism has storage areas corresponding to each sorting channel. Correspondingly, the sorting and storage mechanism in this embodiment can be multiple collection boxes to collect different categories of sorted trays separately.

[0061] Through the above, the gripping mechanism (an actuator with an adaptable tray), the sorting and conveying mechanism (multiple channels corresponding to multiple categories), and the classification and storage mechanism (storage areas corresponding to each channel) ensure that the identification results are accurately executed. The adaptability of the gripping mechanism prevents trays from falling off, the multi-channel design meets the sorting needs of multiple categories, and the correspondence of the storage areas ensures orderly classification. The synergy of these three components makes the closed loop from identification to execution more reliable, improving the sorting accuracy and efficiency of the entire system.

[0062] Example 2 like Figure 2As shown, this embodiment discloses a machine vision-based tray sorting method, applied to the machine vision-based tray sorting system described above. The method includes the following steps: S1: Collect the identification image of the material tray; S2: Store at least one set of identification rules; S3: Analyze the matching degree between the current identification rules and the current conveying parameters, including conveying cycle time and conveying speed; S4: Analyze the confidence level of the initial recognition result based on the matching degree. The initial recognition result is obtained by analyzing the labeled image based on the current recognition rules. S5: Determine whether to output the initial recognition result based on the confidence level; S6: Without outputting the initial recognition result, adjust the current recognition rule to obtain a new recognition rule; S7: Repeatedly analyze the labeled image and evaluate the confidence level using the new recognition rules until the confidence level meets the preset output conditions, and then output the corresponding final recognition result; S8: Perform sorting operations on the pallets based on the final identification results output.

[0063] It should be noted that the machine vision-based tray sorting method in this embodiment corresponds to the aforementioned machine vision-based tray sorting system. Therefore, any content not specifically described in the machine vision-based tray sorting method of this embodiment, including but not limited to functional definitions, working principles, and technical effects, can be referred to the description in the aforementioned machine vision-based tray sorting system, and will not be repeated here.

[0064] In summary, the machine vision-based pallet sorting system and method of this embodiment establishes a dynamic correlation between recognition rules and conveying parameters through a matching analysis module, and achieves multi-dimensional verification of recognition results by combining confidence assessment, thus solving the problem that fixed rules in the prior art are difficult to adapt to changes in conveying status. Adaptability is quantified by using time matching coefficients and sharpness matching coefficients, and confidence assessment is optimized by integrating feature consistency parameters, making the recognition results more reliable. The rule adjustment module accurately optimizes the corresponding rules based on specific influencing factors, reducing the number of iterations and improving adaptive efficiency. The structured design of the sorting execution module ensures efficient implementation of recognition results, thereby improving the overall stability, recognition accuracy, and dynamic adaptability of the sorting system, reducing manual intervention, and making it suitable for multi-condition pallet sorting scenarios.

[0065] In the embodiments provided in this application, it should be understood that the embodiments described herein can be implemented in hardware, software, firmware, middleware, code, or any suitable combination thereof. For hardware implementation, the processor may be implemented in one or more of the following: application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), processors, controllers, microcontrollers, microprocessors, other electronic units designed to implement the functions described herein, or combinations thereof. For software implementation, some or all of the processes of the embodiments may be performed by a computer program instructing the associated hardware. During implementation, the program may be stored in a computer-readable storage medium or transmitted as one or more instructions or code on a computer-readable storage medium. Computer-readable storage media include computer storage media and communication media, wherein communication media include any medium that facilitates the transmission of a computer program from one place to another. Storage media may be any available medium accessible to a computer. Computer-readable storage media may include, but are not limited to, RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code having the form of instructions or data structures and accessible to a computer.

[0066] Finally, it should be noted that the above description is only a preferred embodiment of this application and is not intended to limit this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A machine vision-based tray sorting system, characterized in that, The system includes: The image acquisition module is configured to acquire the identification image of the material tray; The rule storage module is configured to store at least one set of identification rules; The matching analysis module is configured to analyze the matching degree between the current identification rule and the current conveying parameters, including the conveying cycle time and the conveying speed. The confidence assessment module is configured to: analyze the confidence of the initial recognition result based on the matching degree, wherein the initial recognition result is obtained by analyzing the identifier image based on the current recognition rules, and the confidence is determined based on the feature consistency analysis between the matching degree and the initial recognition result; The result determination module is configured to: determine whether to output the initial recognition result based on the confidence level; output the result when the confidence level meets the preset output conditions, otherwise do not output the result. The rule adjustment module is configured to: adjust the current recognition rule to obtain the new recognition rule when the initial recognition result is not output; The loop control module is configured to repeatedly analyze the identification image and evaluate the confidence level using the new recognition rule until the confidence level meets the preset output conditions and then output the corresponding final recognition result. The sorting execution module is configured to perform sorting operations on the pallets based on the output of the final identification result.

2. The machine vision-based tray sorting system according to claim 1, characterized in that, When the matching analysis module analyzes the matching degree, it includes the following steps: A1: Get the processing time of the current recognition rule per cycle; A2: Calculate the ratio of the single processing time to the conveying cycle time to obtain the time matching coefficient.

3. The machine vision-based tray sorting system according to claim 2, characterized in that, When the matching analysis module analyzes the matching degree, it also includes the following steps: A3: Obtain the minimum image clarity requirement corresponding to the current recognition rule; A4: Calculate the difference between the minimum image sharpness requirement and the actual image blur amount associated with the transmission speed to obtain the sharpness matching coefficient; A5: The time matching coefficient and the sharpness matching coefficient are weighted and fused, and the weighted fused value is defined as the matching degree.

4. The machine vision-based tray sorting system according to claim 3, characterized in that, The actual image blur amount is the product of the conveying speed and the exposure time of the image acquisition module, and the minimum image clarity requirement is the maximum allowable blur amount preset by the current recognition rule.

5. The machine vision-based tray sorting system according to claim 4, characterized in that, The identification rules include preprocessing rules, feature extraction rules, and classification rules; The preprocessing rules include at least the filtering method and grayscale conversion parameters; The feature extraction rules include at least the type of feature to be extracted and the feature quantization dimension; The classification rules include at least the classification model type and the category determination threshold.

6. The machine vision-based tray sorting system according to claim 1, characterized in that, When the confidence assessment module analyzes the confidence level, it includes the following steps: B1: Obtain the feature consistency parameter of the initial recognition result. The feature consistency parameter is obtained by: processing the same identified image continuously at least three times using the current recognition rule, extracting the feature vector obtained from each processing, analyzing the corresponding similarity between all feature vectors, and defining the normalized value of the similarity as the feature consistency parameter. B2: The matching degree is weighted and fused with the feature consistency parameter of the initial recognition result, and the weighted fusion value is defined as the confidence degree.

7. The machine vision-based tray sorting system according to claim 5, characterized in that, When the rule adjustment module adjusts the current recognition rules, it includes adjusting at least one of the preprocessing rules, adjusting the feature extraction rules, and adjusting the classification rules.

8. The machine vision-based tray sorting system according to claim 1, characterized in that, The sorting execution module includes at least a gripping mechanism, a sorting conveying mechanism, and a sorting and storage mechanism; The gripping mechanism is equipped with an actuator that adapts to the material tray; The sorting and conveying mechanism includes at least two sorting channels, each corresponding to a different sorting category; The sorting and storage mechanism has storage areas that correspond one-to-one with each of the sorting channels.

9. The machine vision-based tray sorting system according to claim 7, characterized in that, When the rule adjustment module adjusts the current recognition rule, it first analyzes the specific influencing factors that cause the confidence level of the initial recognition result to not meet the output conditions, and then adjusts the corresponding type of recognition rule for the specific influencing factors, including: When the specific influencing factor is that the time matching coefficient is less than a preset time threshold, the feature quantization dimension in the feature extraction rule is adjusted. When the specific influencing factor is that the sharpness matching coefficient is less than the preset sharpness threshold, the filtering method in the preprocessing rule is adjusted. When the specific influencing factor is that the feature consistency parameter is less than the preset consistency threshold, the category determination threshold in the classification rule is adjusted.

10. A machine vision-based tray sorting method, applied to the machine vision-based tray sorting system as described in any one of claims 1-9, characterized in that, The method includes the following steps: S1: Collect the identification image of the material tray; S2: Store at least one set of identification rules; S3: Analyze the matching degree between the current identification rules and the current conveying parameters, including the conveying cycle time and the conveying speed; S4: Analyze the confidence level of the initial recognition result based on the matching degree. The initial recognition result is obtained by analyzing the identifier image based on the current recognition rules. The confidence level is determined based on the feature consistency analysis between the matching degree and the initial recognition result. S5: Determine whether to output the initial recognition result based on the confidence level. Output the result when the confidence level meets the preset output conditions, otherwise do not output the result. S6: Without outputting the initial recognition result, adjust the current recognition rule to obtain the new recognition rule; S7: Repeatedly analyze the identified image and evaluate the confidence level using the new recognition rule until the confidence level meets the preset output conditions, and then output the corresponding final recognition result; S8: Perform a sorting operation on the pallet based on the final identification result output.

Images