Plastic bag quality detection system based on image recognition

Abstract

The invention discloses a plastic bag quality detection system based on image recognition, and relates to the technical field of quality detection, and the system comprises an image acquisition module which is used for obtaining an original image sequence on a production line; the strategy execution module is used for outputting a final quality detection result according to the dynamic strategy coefficient; the result archiving and feedback module is used for aggregating the final quality detection result and outputting an early warning instruction for prompting an operator to intervene; the core state sensing module is used for solving a system risk index based on the historical structured quality data; and the dynamic decision and adjustment module is used for inputting the system risk index into a preset nonlinear adjustment function so as to generate a dynamic strategy coefficient. According to the method, the detection model is adaptively adjusted through the dynamic strategy coefficient, the risk index is fused to perceive the quality trend, a closed-loop control architecture is constructed, collaborative optimization of the speed and precision is achieved, system oscillation is avoided, the quality risk is warned in advance, and the production efficiency and stability are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of quality inspection, and particularly relates to a plastic bag quality inspection system based on image recognition. Background Art

[0002] In the automated production process of plastic bags, quality inspection is an important link to ensure the qualified rate of products. With the improvement of the production line speed and quality standards, the inspection system based on image recognition is widely used in the real-time inspection of quality indicators such as surface defects and dimensional accuracy of plastic bags due to its non-contact and high-efficiency characteristics. This technology collects images through devices such as industrial cameras and realizes quality judgment through algorithm analysis, which can become an important means to replace manual sampling inspection.

[0003] The existing image recognition inspection systems have obvious deficiencies. Traditional image recognition inspection systems mostly adopt detection models with fixed thresholds, and it is difficult to achieve a dynamic balance between detection speed and accuracy. There are often system oscillation problems caused by threshold switching. At the same time, they also lack the ability to perceive the quality change trend, and can only make judgments based on the current defect rate, unable to predict potential batch quality risks in advance, resulting in lagging production adjustment and affecting production efficiency and product quality stability. Summary of the Invention

[0004] The purpose of the present invention is to provide a plastic bag quality inspection system and method based on image recognition, which solves the problems existing in the background art.

[0005] To solve the above technical problems, the present invention provides a plastic bag quality inspection system based on image recognition, including: an image acquisition module for acquiring an original image sequence on a production line; A strategy execution module for adaptively selecting a detection model from a preset speed-priority model and a precision-priority model according to a dynamic strategy coefficient; the strategy execution module is also used to process the original image sequence with the detection model to output a final quality inspection result; A core state perception module for calculating a recent comprehensive defect rate and a change rate of the recent comprehensive defect rate based on the historical structured quality data; the core state perception module is also used to calculate a system risk index by combining the recent comprehensive defect rate and the change rate; A result archiving and feedback module for aggregating the final quality inspection results to generate historical structured quality data recorded according to a detection cycle; the result archiving and feedback module is also used to compare the system risk index with a preset warning threshold, and generate and output a warning instruction for prompting an operator to intervene when the system risk index is higher than the warning threshold in a preset number of consecutive detection cycles.

[0006] A dynamic decision-making and adjustment module for inputting the system risk index into a preset non-linear adjustment function to generate the dynamic policy coefficient; Preferably, the image acquisition module is configured to capture the original image sequence through an industrial camera and a line array sensor at an adjustable frame rate and resolution.

[0007] Preferably, the core status perception module is specifically configured to: Within a preset sliding time window, sum the total number of final defects in the historical structured quality data and sum the total number of final detections; The core status perception module is further configured to divide the sum of the total number of final defects by the sum of the total number of final detections to calculate the recent comprehensive defect rate.

[0008] Preferably, the core status perception module is specifically configured to: Multiply the recent comprehensive defect rate by a preset defect rate weight to obtain a weighted defect rate; Multiply the change rate by a preset change rate weight to obtain a weighted change rate; And add the weighted defect rate and the weighted change rate to calculate the system risk index.

[0009] Preferably, the non-linear adjustment function is an S-shaped function; The dynamic decision-making and adjustment module is configured to subtract a preset risk adjustment center point from the system risk index to obtain a risk deviation; The dynamic decision-making and adjustment module is further configured to multiply the risk deviation by a preset adjustment curve steepness and transform the result through the S-shaped function to generate the dynamic policy coefficient.

[0010] Preferably, the policy execution module is specifically configured to: When the dynamic policy coefficient is lower than a preset speed priority threshold, select the speed priority model as the detection model; When the dynamic policy coefficient is higher than a preset accuracy priority threshold, select the accuracy priority model as the detection model; When the dynamic policy coefficient is between the speed priority threshold and the accuracy priority threshold, adjust the internal parameters of the speed priority model, the internal parameters of the accuracy priority model, or simultaneously adjust the internal parameters of the above two models as the detection model, and the internal parameters include a confidence threshold and an IOU threshold.

[0011] Preferably, the policy execution module is further configured to generate a hardware adjustment instruction in response to the dynamic policy coefficient; The image acquisition module is further configured to receive the hardware adjustment instruction and dynamically adjust the acquisition frame rate and acquisition resolution of the original image sequence based on the hardware adjustment instruction.

[0012] Preferably, the historical structured quality data includes records in units of detection cycles; Each record includes the total number of final detections and the total number of final defects corresponding to the detection cycle.

[0013] Compared with the prior art, the present invention has the following beneficial effects: 1. By adaptively adjusting the detection model with a dynamic policy coefficient, a dynamic balance between detection speed and accuracy is achieved. When the risk is low, the speed-priority model is automatically adopted to improve the detection speed; when the risk increases, it is automatically switched to the accuracy-priority model to improve the detection accuracy, avoiding performance fluctuations caused by the traditional fixed-threshold method and solving the contradiction between speed and accuracy.

[0014] 2. Through the design of fusing the defect rate and its change rate with the risk index, the quality change trend can be sensed in real time. By means of the S-shaped nonlinear adjustment function conversion strategy, system oscillation is fundamentally avoided, and the response sensitivity of the detection system to production quality fluctuations is improved.

[0015] 3. By constructing a "perception - decision - execution" closed-loop control architecture, the detection strategy can be continuously optimized according to the historical structured quality data. From the dynamic adjustment of image acquisition parameters to the adaptive selection of the detection model, a data-driven intelligent adjustment mechanism is formed, realizing the transformation from passive detection to active adaptation, and providing technical support for early prevention of batch quality problems and stable production efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts; Figure 1 It is the logical block diagram of the system of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0017] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some of the embodiments of the present invention, rather than all of them. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0018] Example 1: Please refer to Figure 1 , the present invention provides a plastic bag quality detection system based on image recognition, including: an image acquisition module for acquiring an original image sequence on a production line; A strategy execution module for adaptively selecting a detection model from a preset speed-priority model and a precision-priority model according to a dynamic strategy coefficient; the strategy execution module is also used to process the original image sequence by using the detection model to output a final quality detection result; A core state perception module for calculating a recent comprehensive defect rate and a change rate of the recent comprehensive defect rate based on the historical structured quality data; the core state perception module is also used to calculate a system risk index by combining the recent comprehensive defect rate and the change rate; A result archiving and feedback module for aggregating the final quality detection results to generate historical structured quality data recorded according to a detection period; the result archiving and feedback module is also used to compare the system risk index with a preset warning threshold, and when the system risk index is higher than the warning threshold in a preset number of consecutive detection periods, generate and output a warning instruction for prompting an operator to intervene; A dynamic decision-making and adjustment module for inputting the system risk index into a preset non-linear adjustment function to generate the dynamic strategy coefficient; This embodiment provides a modular solution. Through the close cooperation of the image acquisition module, the strategy execution module, the result archiving and feedback module, the core state perception module, and the dynamic decision-making and adjustment module, a complete closed-loop control system from data acquisition, state perception, dynamic decision-making to strategy execution is constructed; the design of this system architecture aims to achieve dynamic adaptive detection of the production quality of plastic bags. By actively perceiving the quality change trend and real-time adjusting the detection strategy, it fundamentally solves the contradiction that is difficult to reconcile between speed and precision in traditional detection methods, and can quickly respond to the trend of quality problems, thereby improving the intelligence and automation level of the entire production process.

[0019] Example 2: The image acquisition module is used to capture the original image sequence through an industrial camera and a line array sensor at an adjustable frame rate and resolution; In this embodiment, the image acquisition module can, by configuring the industrial camera and the line array sensor, capture the original image sequence at an adjustable frame rate and resolution Capture the plastic bags flowing at high speed on the production line to generate an original high-frame-rate image sequence; this configuration ensures that the original data containing the complete surface details of the plastic bags can be obtained, providing a high-quality input basis for subsequent precise detection. The adjustability of its parameters is the hardware basis for implementing the system's adaptive detection strategy, and it responds to the instructions of the downstream module for dynamic adjustment to ensure that the data acquisition link can provide optimal support whether pursuing speed or accuracy.

[0020] Example 3: The core status perception module is specifically used to sum the total number of final defects in the historical structured quality data within a preset sliding time window and sum the total number of final detections; the core status perception module is also used to divide the sum of the total number of final defects by the sum of the total number of final detections to calculate the recent comprehensive defect rate. The core status perception module is specifically used to multiply the recent comprehensive defect rate by a preset defect rate weight to obtain a weighted defect rate; multiply the change rate by a preset change rate weight to obtain a weighted change rate; and add the weighted defect rate and the weighted change rate to calculate the system risk index. The core status perception module obtains historical structured quality data from the result archiving and feedback module and performs operations within a preset sliding time window T to quantify the current production quality status; this module calculates the recent comprehensive defect rate and its change rate to further calculate the system risk index ; The recent comprehensive defect rate is calculated based on the statistics of historical data within the sliding time window T, and its calculation formula is:

[0021] is the recent comprehensive defect rate of the current detection cycle; is the current detection cycle; is the size of the preset sliding time window; is the number of final defects in the i-th detection cycle; is the total number of final detections in the i-th detection cycle; This calculation method can reflect the immediate status of the current production quality in real time; The system risk index is calculated by multiplying the recent comprehensive defect rate by its change rate Perform weighted summation to fuse the immediate state and the change trend of quality. The calculation formula is as follows:

[0022] is the system risk index for the current detection period; is the defect rate weight (a preset parameter); is the recent comprehensive defect rate for the current detection period; is the change rate weight (a preset parameter); is the defect rate change rate; is the prediction time constant, a pre-set engineering parameter with a time dimension used to adjust the response sensitivity of the system to the quality change trend; Change rate is approximately calculated as , where is fixed to 1 detection period; This comprehensive risk assessment method enables the system to not only focus on the current defect level but also predictively capture the signs of quality deterioration or improvement, providing a deeper and more forward-looking basis for decision-making.

[0023] Example 4: The non-linear adjustment function is an S-shaped function; The dynamic decision-making and adjustment module is used to subtract a preset risk adjustment center point from the system risk index to obtain a risk deviation; The dynamic decision-making and adjustment module is also used to multiply the risk deviation by a preset adjustment curve steepness and transform the result through the S-shaped function to generate the dynamic policy coefficient; The dynamic decision-making and adjustment module receives the system risk index calculated by the core state perception module , and uses a preset S-shaped non-linear adjustment function to transform it into a smooth dynamic policy coefficient ranging from 0 to 1 ; The preset S-shaped function simulates the expert decision-making logic, enabling smooth and non-linear policy transitions according to the risk level and avoiding system oscillations that may be caused by traditional threshold switching methods; The formula for this transformation process is:

[0024] is the dynamic policy coefficient (value range [0,1]); is the adjustment curve steepness (a preset parameter, controlling the aggressiveness of policy conversion); is the system risk index; is the risk adjustment center point; Through this function, the system risk this quantitative index is mapped to a guiding policy parameter , where the value is set by experts according to the desired aggressiveness of the policy transition, and is the risk balance point defined by the business, at which the system speed and accuracy reach the most balanced state (corresponding to ), thus providing the core driving force for the smooth switching of the downstream module between the two strategies of "speed priority" and "accuracy priority".

[0025] Example 5: The policy execution module is specifically used to select the speed priority model as the detection model when the dynamic policy coefficient is lower than the preset speed priority threshold; select the accuracy priority model as the detection model when the dynamic policy coefficient is higher than the preset accuracy priority threshold; when the dynamic policy coefficient is between the speed priority threshold and the accuracy priority threshold, adjust the internal parameters of the speed priority model, the internal parameters of the accuracy priority model, or adjust the internal parameters of the above two models simultaneously as the detection model, and the internal parameters include the confidence threshold and the IOU threshold; The policy execution module is also used to generate a hardware adjustment instruction in response to the dynamic policy coefficient; the image acquisition module is also used to receive the hardware adjustment instruction and dynamically adjust the acquisition frame rate and acquisition resolution of the original image sequence based on the hardware adjustment instruction; The policy execution module is the ultimate executor of the system decision. It receives the dynamic policy coefficient generated by the dynamic decision and adjustment module , and converts it into specific software algorithms and hardware parameter configurations; at the software level, when approaches 0, this module selects and applies a lightweight detection model with speed priority to maximize the processing efficiency; when approaches 1, it switches to a deep learning model with accuracy priority to ensure the highest detection accuracy; for values between the two, the module realizes the seamless integration of the two strategies by smoothly adjusting the key parameters inside the model; at the hardware level, this module also maps the value to specific instructions for the image acquisition module. When the value increases (biased towards accuracy), it will instruct the image acquisition module to moderately increase the acquisition frame rate and resolution , providing richer data support for high-precision algorithms; conversely, reducing these parameters to lighten the system load; this software and hardware collaborative regulation mechanism enables the system to act as an organic whole, dynamically and finely allocate computing resources according to real-time risks, and realizes the optimization of system energy efficiency on the premise of meeting quality control requirements; Embodiment 6: The historical structured quality data includes records in units of detection cycles; each of the said records contains the total number of final detections and the total number of final defects corresponding to the detection cycle; The result archiving and feedback module is the only source for generating the system's historical data, and is responsible for real-time aggregation and archiving of each detection result output by the policy execution module; this module At the end of each detection cycle, it will count the "total number of final detections" and the "total number of final defects" , and this set of data is stored as a structured record in the historical database; through this continuous recording process, the system constructs a traceable historical structured quality database characterized by time series; these records are not only the basis for evaluating the long-term performance of the system, but more importantly, they form the data cornerstone of the decision feedback closed-loop, providing the necessary and authoritative input data for the upstream core state perception module to calculate the recent defect rate and risk index, thus ensuring that the decisions of the entire system are always based on the real and continuous production quality history, realizing truly data-driven intelligent control.

[0026] The above are only the preferred embodiments of the present invention, and do not limit the present invention in other forms. Any person skilled in the art may use the disclosed technical content to make changes or modifications into equivalent embodiments with equivalent changes and apply them to other fields. However, as long as it does not depart from the technical solution content of the present invention, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention still fall within the protection scope of the technical solution of the present invention.

Claims

1. A plastic bag quality detection system based on image recognition, characterized in that It includes: An image acquisition module for obtaining the original image sequence on the production line; A policy execution module for adaptively selecting a detection model from a preset speed - priority model and a precision - priority model according to a dynamic policy coefficient; the policy execution module is also used to process the original image sequence by using the detection model to output a final quality detection result; A core state perception module for calculating a recent comprehensive defect rate and a change rate of the recent comprehensive defect rate based on the historical structured quality data; the core state perception module is also used to calculate a system risk index by combining the recent comprehensive defect rate and the change rate; A result archiving and feedback module for aggregating the final quality detection result to generate historical structured quality data recorded by detection cycle; the result archiving and feedback module is also used to compare the system risk index with a preset warning threshold, and generate and output a warning instruction for prompting an operator to intervene when the system risk index is higher than the warning threshold in a preset number of consecutive detection cycles; A dynamic decision - making and adjustment module for inputting the system risk index into a preset non - linear adjustment function to generate the dynamic policy coefficient.

2. The quality detection system of plastic bags based on image recognition according to claim 1, wherein The image acquisition module is used to capture the original image sequence through an industrial camera and a line - array sensor at an adjustable frame rate and resolution.

3. The quality detection system for plastic bags based on image recognition according to claim 1, characterized in that, The core state perception module is specifically used for: Summing up the total number of final defects in the historical structured quality data and summing up the total number of final detections within a preset sliding time window; The core state perception module is also used to divide the sum of the total number of final defects by the sum of the total number of final detections to calculate the recent comprehensive defect rate.

4. The plastic bag quality detection system based on image recognition according to claim 1, characterized in that, The core state perception module is specifically used for: Multiplying the recent comprehensive defect rate by a preset defect rate weight to obtain a weighted defect rate; Multiplying the change rate by a preset change rate weight to obtain a weighted change rate; And adding the weighted defect rate and the weighted change rate to calculate the system risk index.

5. A plastic bag quality detection system based on image recognition according to claim 1, characterized in that The non - linear adjustment function is an S - type function; The dynamic decision - making and adjustment module is used to subtract a preset risk adjustment center point from the system risk index to obtain a risk deviation; The dynamic decision - making and adjustment module is also used to multiply the risk deviation by a preset adjustment curve steepness and convert the result through the S - type function to generate the dynamic policy coefficient.

6. The quality detection system of plastic bags based on image recognition according to claim 1, characterized in that, The policy execution module is specifically used for: When the dynamic policy coefficient is lower than a preset speed - priority threshold, selecting the speed - priority model as the detection model; When the dynamic policy coefficient is higher than a preset precision - priority threshold, selecting the precision - priority model as the detection model; When the dynamic policy coefficient is between the speed priority threshold and the accuracy priority threshold, the internal parameters of the speed priority model, the internal parameters of the accuracy priority model, or the internal parameters of the above two models are adaptively adjusted simultaneously as the detection model, and the internal parameters include the confidence threshold and the IOU threshold.

7. A plastic bag quality detection system based on image recognition according to claim 1, characterized in that, The policy execution module is further configured to generate a hardware adjustment instruction in response to the dynamic policy coefficient; The image acquisition module is further configured to receive the hardware adjustment instruction and dynamically adjust the acquisition frame rate and acquisition resolution of the original image sequence based on the hardware adjustment instruction.

8. A plastic bag quality detection system based on image recognition according to claim 1, characterized in that, The historical structured quality data includes records in units of detection cycles; Each of the records contains the total number of final detections and the total number of final defects corresponding to the detection cycle.

Images