Search

CN-121982014-A - Method for detecting health product package integrity through cooperative efficiency and intelligent detection

CN121982014ACN 121982014 ACN121982014 ACN 121982014ACN-121982014-A

Abstract

The invention provides a method for detecting the package integrity of a health product with collaborative efficiency and intelligent detection, which comprises the steps of S1, detecting system space-time synchronization and hardware module calibration, S2, packaging sample feature preliminary screening and suspected sample extraction, S4, suspected sample space vision three-dimensional point cloud accurate reconstruction, S5, multi-mode sensor data and three-dimensional point cloud model fusion mapping, S6, defect feature accurate identification and feature gradient feedback, S7, defect level judgment and full-dimension cross check, S8, detecting model increment training and full-flow algorithm optimization, S9, product line collaborative execution and detecting result full-flow feedback. According to the invention, through multi-algorithm bidirectional closed loop fusion and full flow collaboration, the accuracy and reliability of package defect identification are improved, the real-time detection requirement of an industrial production line is adapted, the full traceability of detection data and the continuous optimization of detection performance are realized, and the packaging quality of health products is ensured.

Inventors

  • GAO XIAOWEI
  • WEN XUAN

Assignees

  • 高胜智造生物科技(上海)有限公司

Dates

Publication Date
20260505
Application Date
20260210

Claims (10)

  1. 1. The method for detecting the package integrity of the health care product by using the cooperative efficiency and the intelligent detection is characterized by comprising the following steps of: S1, performing space-time synchronization and precision calibration on a hardware module of a detection system to obtain a calibration parameter set of the detection system; s2, based on a calibration parameter set of a detection system, performing global space-time synchronous data acquisition on the health product packaging sample, and binding a unique sample identification code to obtain a global original data set; s3, performing feature extraction and sample preliminary screening on the global original data set by adopting an improved lightweight convolutional neural network algorithm to obtain a suspected sample data set; S4, performing point cloud preprocessing and three-dimensional point cloud accurate reconstruction on the suspected sample data set by adopting a radial basis function implicit curved surface reconstruction algorithm to obtain a spatial vision accurate model set; s5, performing multi-mode data depth fusion by adopting a tensor kernel space fusion algorithm and combining a sensor feature set extracted from a suspected sample data set and a space feature set extracted from a space vision accurate model set to obtain a fusion data set; s6, performing defect characteristic accurate identification on the fusion data set by adopting a capsule network algorithm with enhanced image attention to obtain a defect identification data set; And S7, performing defect grade judgment and full-dimension cross checking based on the defect identification data set to obtain a defect grade judgment and checking data set.
  2. 2. The method for detecting the package integrity of a health product by cooperatively and intelligently detecting the efficiency of the cooperation according to claim 1, further comprising: S8, constructing an incremental training sample set based on the defect grade judgment and verification data set, and performing incremental training and parameter optimization on an improved lightweight convolutional neural network algorithm, a radial basis function implicit surface reconstruction algorithm, a tensor kernel space fusion algorithm and a graph injection meaning force enhanced capsule network algorithm to obtain a model optimization parameter set; and S9, based on the defect grade judging and checking data set and the model optimizing parameter set, the detection system and the production line control system are cooperatively executed, and the full-flow detection result is fed back to each step from S1 to S8, so that a full-flow closed loop for detecting the package integrity of the health product is formed.
  3. 3. The method for detecting the package integrity of the health care product by cooperative efficiency and intelligent detection according to claim 1, wherein S5 generates a spatial registration error in a multi-mode data depth fusion process, the spatial registration error is fed back to S4 for adjusting parameters of a radial basis function implicit surface reconstruction algorithm, S6 generates a defect feature gradient in a defect feature accurate identification process, and the defect feature gradient is fed back to S5 for adjusting parameters of a tensor integration space fusion algorithm.
  4. 4. The method for detecting the package integrity of the health care product by using the cooperative efficiency and the intelligent detection according to claim 3 is characterized in that the specific process of feeding back the space registration error to the S4 is that the S5 calculates the space registration error of the two types of features based on the sensor feature set and the space feature set, the space registration error is transmitted to a radial basis function implicit curved surface reconstruction algorithm of the S4 in real time, the S4 adjusts basis function weight coefficients and supporting radiuses of the algorithm based on the space registration error, the algorithm after the adjustment carries out three-dimensional point cloud accurate reconstruction on a suspected sample data set again, an optimized space vision accurate model set is generated, and the space vision accurate model set is input into the S5, so that dynamic optimization of reconstruction accuracy is realized.
  5. 5. The method for detecting the package integrity of the health care product by using the collaborative efficiency and the intelligent detection according to claim 3 is characterized in that the specific process of feeding back the defect characteristic gradient to S5 is that when S6 carries out defect characteristic identification, a digital capsule vector set is generated through the capsule layer operation of a capsule network algorithm with enhanced image attention, meanwhile, the first partial derivative of the digital capsule vector set by the total loss of a model is calculated, the defect characteristic gradient is obtained, the defect characteristic gradient is transmitted to a tensor kernel space fusion algorithm of S5 in real time, the S5 adjusts the kernel weight matrix of the algorithm based on the defect characteristic gradient, the adjusted algorithm is recombined with the sensor characteristic set and the space characteristic set to carry out multi-mode data depth fusion, an optimized fusion data set is generated and is input into S6, and the defect recognition degree of fusion characteristics is improved.
  6. 6. The method for detecting the package integrity of the health care product by cooperative efficiency and intelligent detection according to claim 3, wherein S4, S5 and S6 form a bidirectional interactive closed loop, and the termination condition of the closed loop is that when the spatial registration error generated by S5 is reduced to a preset matching threshold value and the full-dimensional cross check verification of S7 is effective, the bidirectional interactive closed loop is terminated, and S4, S5 and S6 respectively output a final spatial vision accurate model set, a fusion data set and a defect identification data set for defect grade judgment, algorithm incremental training and cooperative execution of a production line.
  7. 7. The method for detecting the package integrity of the health care product by using the collaborative efficiency and the intelligent detection according to claim 1 is characterized in that the specific implementation process of the improved lightweight convolutional neural network algorithm in the step S3 is that spatial visual features and sensor physical features are extracted from a global original data set, the two types of features are spliced to form a global lightweight feature vector, the global lightweight feature vector is input into a trained improved lightweight convolutional neural network model, the model outputs the class prediction probability of a sample through feature operation, samples which cannot be definitely judged to be qualified or definitely defective are screened out according to a preset probability threshold, and relevant data of all the samples are integrated to form a suspected sample data set.
  8. 8. The method for detecting the package integrity of the health care product by the cooperative efficiency and the intelligent detection according to claim 1 is characterized in that the specific implementation process of the implicit curved surface reconstruction algorithm of the radial basis function in S4 is that three-dimensional rough scan point cloud data is extracted from a suspected sample data set, point cloud preprocessing operation of outlier rejection and key point extraction of a defect potential area is firstly carried out on the three-dimensional rough scan point cloud data, then a tight support radial basis function is constructed based on the preprocessed point cloud data, implicit curved surface model parameters are solved, then a laser three-dimensional scanner is controlled to carry out high-precision three-dimensional fine scan on the suspected sample to obtain fine scan point cloud data, the fine scan point cloud data is fitted to an implicit curved surface model to complete three-dimensional point cloud reconstruction, and finally spatial feature binding is carried out by combining six-view visual features of the suspected sample and the reconstructed point cloud model to generate a spatial visual accurate model set.
  9. 9. The method for detecting the package integrity of the health-care product by the collaborative efficiency and the intelligent detection according to claim 1 is characterized in that the full-dimension cross check in the step S7 comprises three mutually independent check layers, namely sensor data check, space vision data check and fusion feature check, wherein the sensor data check verifies the physical matching of defect features based on a sensor feature set extracted by a suspected sample data set, the space vision data check verifies the space matching of defect features based on a space vision accurate model set, the fusion feature check verifies the fusion matching of defect features based on a fusion data set, when the check results of the three layers are all passed, the defect identification data set is judged to be valid, the corresponding defect grade judgment and check data set is generated based on the defect identification data set, and when any one layer fails to check, relevant data of a corresponding sample is reclassifying the suspected sample data set, and a secondary detection flow is started.
  10. 10. The method for detecting the package integrity of the health care product with cooperative efficiency and intelligent detection according to claim 2 is characterized in that the construction and increment training process of the increment training sample set in the step S8 is that samples with effective judgment results are selected from defect grade judgment and verification data sets, suspected false detection samples are removed, relevant data of the effective samples are classified and integrated according to algorithm types, an increment training sample set corresponding to a lightweight convolutional neural network algorithm, a radial basis function implicit surface reconstruction algorithm, a tensor kernel space fusion algorithm and a capsule network algorithm with enhanced drawing and meaning force one by one is constructed, small-batch increment training is respectively carried out for four types of algorithms, core parameters of each algorithm are adjusted, and optimization parameters of all algorithms are integrated after training is completed, so that a model optimization parameter set is formed.

Description

Method for detecting health product package integrity through cooperative efficiency and intelligent detection Technical Field The invention relates to the technical field of health product packaging detection, in particular to a health product packaging integrity detection method with synergistic efficiency and intelligent detection. Background In the current health product package integrity detection industry, the traditional detection mode mainly relies on manual visual inspection, and has the problems of low detection efficiency, strong subjectivity and high false detection rate of missed detection, and cannot adapt to the high-speed detection requirement of a modern production line. The existing automatic detection scheme mainly adopts a single sensor detection or simple data fusion technology, and has the defects that firstly, the three-dimensional point cloud reconstruction precision is insufficient, the fine defects of irregular curved surfaces of the health-care product package are difficult to capture, reconstruction distortion is easy to occur, secondly, the multi-mode sensor data and the space vision data are not fused deeply, the characteristic loss is serious, the physical characteristics and the space characteristics of the package cannot be comprehensively reflected, thirdly, the defect identification algorithm depends on the traditional convolutional neural network, the space structure information of the defects is difficult to be reserved, the slight defects are missed to be detected, fourthly, the detection flow is unidirectional data transmission, a closed-loop optimization mechanism is not adopted, the detection precision and the detection efficiency cannot be continuously improved along with the detection times, and fifthly, the detection result and the execution link of the production line are poor in cooperativity, the detection-execution-optimization integrated management and control cannot be realized, and the strict packaging quality standard of the health-care product industry is difficult to be met. Disclosure of Invention The invention provides a method for detecting the package integrity of a health product by cooperative efficiency and intelligent detection, which aims at the defects in the prior art, realizes the automation, high precision and intelligent management and control of the whole process of the package detection of the health product, solves the problems of low manual detection efficiency, insufficient automatic detection precision, lost characteristics, no closed loop optimization and poor product line synergy, improves the accuracy and reliability of package defect identification by multi-algorithm bidirectional closed loop fusion and the cooperation of the whole process, adapts to the real-time detection requirement of an industrial production line, realizes the full traceability and continuous optimization of detection data, and ensures the package quality of the health product. In order to achieve the above purpose, the invention adopts the following technical scheme: the method for detecting the package integrity of the health-care product by using the cooperative efficiency and the intelligent detection comprises the following steps: S1, performing space-time synchronization and precision calibration on a hardware module of a detection system to obtain a calibration parameter set of the detection system; s2, based on a calibration parameter set of a detection system, performing global space-time synchronous data acquisition on the health product packaging sample, and binding a unique sample identification code to obtain a global original data set; s3, performing feature extraction and sample preliminary screening on the global original data set by adopting an improved lightweight convolutional neural network algorithm to obtain a suspected sample data set; S4, performing point cloud preprocessing and three-dimensional point cloud accurate reconstruction on the suspected sample data set by adopting a radial basis function implicit curved surface reconstruction algorithm to obtain a spatial vision accurate model set; s5, performing multi-mode data depth fusion by adopting a tensor kernel space fusion algorithm and combining a sensor feature set extracted from a suspected sample data set and a space feature set extracted from a space vision accurate model set to obtain a fusion data set; s6, performing defect characteristic accurate identification on the fusion data set by adopting a capsule network algorithm with enhanced image attention to obtain a defect identification data set; S7, performing defect grade judgment and full-dimension cross check on the basis of the defect identification data set to obtain a defect grade judgment and check data set; S8, constructing an incremental training sample set based on the defect grade judgment and verification data set, and performing incremental training and parameter optimization on an improved lightweight convolut