Research On Key Technology Of Foreign Body Detection In Large Infusion

Large infusion is one of the five important preparations in the field of medicine.In the production process,foreign matter occasionally enters the liquid reagent,which has great potential harm to infusion patients.At present,the detection of foreign bodies in large infusions mostly adopts the combination of machine vision and image processing.Although the detection speed is greatly improved compared with manual visual inspection,there are still many problems.The main reasons are that some foreign objects are missed due to their small size,and bubbles and random noises are mistakenly detected as foreign objects.In view of these problems,this paper proposes a large infusion foreign body detection system,and studies the key technologies.The main contents of this paper are as follows:(1)In view of the detection requirements of foreign bodies in large infusions,an image acquisition system was built by selecting appropriate industrial cameras,optical lenses and light sources.According to the different optical properties of foreign objects in the liquid,two different optical imaging schemes are studied and divided into two detection areas,bright field and dark field.Compared with the conventional imaging scheme,the contrast of foreign objects in the image is significantly improved.From the image acquisition level area to reduce the difficulty of subsequent foreign body detection.(2)Aiming at the situation that the conventional median filtering algorithm will weaken the details of small foreign objects,a switching median filtering method based on extreme value detection is studied for filtering.The experimental comparison shows that the filtering method can effectively retain the details of small foreign objects while filtering out the noise in the image.Aiming at the situation that the original ORB algorithm has poor registration effect in the dark image,this paper improves the ORB algorithm.An image enhancement link based on Laplace operator is added to improve the dark image features;In the feature point description stage,the brief descriptor is improved by 3 × 3 neighborhood replaces the original single point sampling method to avoid noise point interference;In the feature point matching stage,the block matching strategy is introduced to improve the matching efficiency.The experimental comparison shows that the improved registration algorithm has good registration effect while maintaining rapidity.In the ROI setting,this paper studies a method based on the sum of the pixel values of the rows and columns of the image.The experimental comparison shows that this method has a faster processing speed than other methods when it meets the set accuracy requirements.(3)Aiming at the detection of moving foreign bodies in large infusions,this paper studies a moving object extraction method based on three-frame difference and Fisher evaluation function.By introducing the Fisher evaluation function as the threshold setting criterion for image binarization,the moving object and the background can be better segmented,and the outline and details of the moving foreign object are preserved.The current moving position of the moving target is extracted by three-frame difference,which effectively avoids the phenomenon of "ghosting" and "hollowing" in the conventional difference method.By analyzing the motion characteristics of the foreign object,the difference in the motion trajectories of foreign objects,bubbles and noise is summarized.On this basis,multiple eigenvalues of the moving object are obtained through Blob analysis,and the tracking of the moving object is realized according to the feature matching.The experimental test proves that the algorithm has a good tracking effect of moving targets,and can eliminate the interference of bubbles and noise through different motion trajectory characteristics.(4)In this paper,the upper computer software is written for the large infusion foreign body detection system.Through this software,600 bottles of large infusion products were tested for foreign body detection,and the detection accuracy was 95.67%,which met the actual detection requirements.There are 49 figures,11 tables and 52 references in this thesis.