Study On Automatic Non-destructive Inspection System Of Vaccinated Egg Survival Based On Machine Vision

Vaccine-virus immunization is the main therapy for avian flu pandemicpreparedness. The quality of bird flu vaccine is directly related to poultry and human’slife. Now bird flu vaccines are generally produced by brewing up live flu virus strainsin specific pathogen free eggs. The ability to automatically discriminate survival ofvaccinated SPF eggs prior to virus proliferative cultivation, which can protect thenormal eggs from bacterial infections and obtain sterile virus strains, is conducive toremoving the dead vaccinated eggs timely This paper develops a non-destructivedetection system based on machine vision for vaccinated eggs survival identificationbefore virus cultivation to replace human visual inspection, which observedangiogenesis in vaccinated eggs all by human vision. Image processing techniques andpattern recognition techniques both are used to realize a fully functional solution forvaccinated eggs survival identification The main study contents and achievements inthis paper are as follows:(1) Vaccinated egg image acquisition system is established to acquire vertical eggimage, then a160×70rectangular region is segmented as ROI for imagepreprocessing after image auto-cropping on angiogenesis area in the image. Thisoperation can not only remove background interference, but also greatly reduce thecomputation of image processing.(2) For the frustrating problem that previous studies cannot process brown eggimages with lots of discrete high-brightness speckle noise as a result of hightransmittance of holes in eggshell, this paper respectively introduces Harris algorithmand SUSAN algorithm to distinguish the speckle noise pixels, then accordinglyproposes symmetrical-neighbor mean filtering and USAN local gray mean filteringalgorithm to remove the intrinsic speckle noise. Finally, this study has compared thedenoising effects with common filtering algorithms, the experimental results show thatthe proposed denoising algorithm can not only eliminate the speckle noise pixels completely but maintain detail features of egg image.(3) According to the principle that manual candling confirms survival ofvaccinated eggs base on their angiogenesis, this study uses Canny edge detectionalgorithm and mathematical morphology operation to process the preprocessed eggimages, then a binary image including main blood vessels was extracted accurately.Finally, labeling method is used to obtain value of blood vessel area.(4) The percentage of blood vessel area from egg ROI area is selected as thecharacteristic parameter for egg survival classification after analyzing its spatialdistribution pattern, which can increase the robustness of survival classification.Finally, nearest neighbor classification algorithm is applied for the real-time onlinesurvival identification.Taking360brown vaccinated eggs for survival non-destructive inspectionexperiment, the experimental results showed that the designed system can achieveidentification accuracy rate of96.94%, false-positive identification rate of3.06%,false-negative identification of0, average identification time of0.642s, which couldmeet the need of practical production. It is proved that the designed system is feasiblefor vaccine manufacturing to judge the survival of vaccinated eggs real-timely.