【作者】 王勇；

【导师】 殷涌光；

【作者基本信息】 吉林大学 ， 农业机械化工程， 2014， 博士

【摘要】 金黄色葡萄球菌、蜡样芽孢杆菌、沙门氏菌是食物中毒事件最常见的三种致病菌，其检测目前还是依靠传统的培养方法来确定，需要进行增菌、分离、纯化、生化鉴定和血清学实验等一系列步骤，这些方法最大的问题是耗时长，通常需要几天，不能满足食品生产企业和质量监督检验部门快速检测的要求，因此，寻找一种快速简便的检测方法就显得非常重要。本论文开展了对金黄色葡萄球菌、蜡样芽孢杆菌、沙门氏菌经过染色处理后利用计算机视觉技术进行快速检测的研究，这种方法完成相应致病菌检测的时间很短，最长需要5h，最短仅需2h，非常具有应用和推广的价值。试验研究了三种致病菌的前处理方法、染色和对应的计算机视觉识别技术，然后用本实验室自行研制的食品微生物快速检测系统进行检测，达到了快速检测的试验目的。相关的研究工作和结论如下：（1）研究了金黄色葡萄球菌的计算机视觉识别快速检测方法，包括金黄色葡萄球菌基础培养基的选择、抑制剂和促进剂的筛选、选择培养基的优化、金黄色葡萄球菌的染色、金黄色葡萄球菌图片的去噪、分割、形态学运算、提取、识别等方面内容，并且确定了计算机快速检测方法的检测限，和国标检测方法进行了比较，得到如下结论：确定了选择培养基的优化配方，即在基础培养基的基础上加上抑制剂和促进剂，植物蛋白胨3.0g/L，胰蛋白胨17.0g/L，氯化钠68g/L，葡萄糖2.5g/L，磷酸氢二钾2.5g/L，亚碲酸钾60mg/L，丙酮酸钠5g/L，甘氨酸9g/L，苯乙醇3.8ml/L。利用微生物快速检测系统进行检测的步骤是：取样品25g，剪成小块，为进一步破碎，加入灭菌砂进一步磨碎，将碎屑与放入均质袋内，加入225mL生理盐水，均质1～2min，用中速滤纸进行过滤，除去颗粒物，对滤液进行10倍稀释。取0.9mL选择培养基，加入到1.5mL离心管中，再取0.1mL的金黄色葡萄球菌菌悬液进行接种，将离心管在37℃120rpm的条件下培养4h。先对离心管中的菌液做10倍的稀释，然后用一次性注射器将这10mL液体进行过滤浓缩，浓缩使用一次性的针式过滤器，其孔径为0.45μm，可以有效将菌液进行过滤，并达到浓缩的目的。用微量取液器吸取2μL经过过滤浓缩的菌液，移至载玻片上，吹干。然后用无菌水冲洗一下干燥后的菌液，滤纸吸走多余水分，最后将该载玻片插入食品微生物快速检测仪中进行检测。图像处理采用采用中值滤波降噪、全局阈值分割、开运算平滑处理和边缘提取的步骤和方法，利用BP神经网络4-5-1模型进行图像的判别。快速检测法与国标法两种方法不存在显著差异，检测时间在5h以内，检测限为10～107cfu/mL。（2）研究了蜡样芽孢杆菌的计算机视觉识别快速检测方法，包括研究蜡样芽孢杆菌性质、杂菌的去除、菌悬液的制备、微波萌发、短期发酵、染色、蜡样芽孢杆菌图片的去噪、分割、形态学运算、特征值提取和图像的识别等方面内容，并且和国标检测方法进行了比较，得到如下结论：样品通过65℃水浴热处理30min杀灭非目标菌，以液体LB培养基为基质，添加萌发剂组合硫酸锌0.05%、氯化镁0.1%、氯化锰0.15%。菌悬液用微波处理60s以促进萌发，并经过2h的短期发酵。利用微生物快速检测系统进行检测的步骤是：称量25g样品，加入到锥形瓶中，同时加入55mL磷酸钠缓冲液，混合均匀后水浴65℃加热30min，然后冷却到室温。取1mL样品菌液，加入到1.5mL离心管中，11000rpm离心后收集芽孢，加入0.85mL萌发液，混匀后，放入微波炉中，在高档下加热60秒钟，然后冷却。取经过微波萌发的菌液0.1mL，加入到离心管中，同时加入短期发酵液，在37℃150rpm的条件下培养2小时。取出离心管，先对离心管中的菌液用0.9mL蛋白胨溶液做10倍的稀释，然后用一次性注射器将这1mL液体进行过滤浓缩，浓缩使用一次性的针式过滤器，其孔径为0.45μm，再用蛋白胨溶液冲洗一次并过滤。用微量取液器吸取5μL经过过滤浓缩的菌液，移至载玻片上，吹干。滴加6～8μLGluc染色液，加盖盖玻片后，避光保存，37℃等待10～20min，然后用无菌水冲洗一下，滤纸吸走多余水分，最后将该载玻片插入食品微生物快速检测仪中进行检测。图像处理采用中值滤波去噪、RGB彩色空间分割、开+闭运算平滑处理、形态和颜色特征提取等步骤，用以进行图像识别的BP神经网络结构为8-6-1，测试样本的识别准确率达到95%以上。快速检测方法与国标检测方法之间无显著差异，检测时间在5h以内，检出限为50～1×106cfu/mL。(3)研究了沙门氏菌的计算机视觉识别快速检测方法，包括胶体金的制备、金标抗体的制备、沙门氏菌染色制片、计算机获得图像的去噪、分割、形态学处理、特征值提取等内容，并且和国标检测方法进行了比较，得到如下结论：确定了合成胶体金溶液和金标抗体的条件：将0.01%氯金酸水溶液加热，加入还原剂柠檬酸三纳，连续煮沸后合成粒径为15nm～25nm胶体金溶液。将胶体金溶液调节pH为9.0，加入二抗，合成标记物，采用高速离心法对合成溶液进行纯化。利用微生物快速检测系统进行检测的步骤是：称取25g样品，剪碎成小块，加入盛有175mL含1%BSA的PBS-T溶液无菌均质杯中，均质1min～2min，样品匀液采用定量的中速滤纸进行过滤，除去大颗粒物质。然后用50mL含1%BSA的PBS-T溶液冲洗，收集滤液。用注射器吸取10mL滤液注入膜过滤器中浓缩，吸取2μL浓缩液于干净载玻片上，室温或37℃下晾干，再经75%的乙醇溶液固定6～9min，用1%牛血清白蛋白的PBS-T溶液冲洗掉固定液，滤纸吸干后，加入2μL浓度为1:2000的沙门氏菌一抗溶液，37℃下作用30min后，用洗涤液冲洗掉未结合的一抗，滤纸吸干后，再滴加2μL浓度1:40的金标物溶液，37℃下作用30min后，用洗涤液冲洗掉未结合的二抗，滤纸吸干后，分次加入2μL增菌液、第一次作用2min、第二次作用时间控制在6min内，先用洗涤液冲洗，再用去离子水冲洗，风干后送入快速检测系统中进行检测。图像处理选用顶帽交换结合灰度窗口变化法对进行去噪，用迭代阈值法进行分割处理，采用开闭运算和连通区域标记对图像进行形态学处理，应用填充空洞腐蚀法提取图像边缘，应用八链码跟踪方法进行特征值提取，BP神经网络结构为4-3-1。快速检测方法与国标方法进行比较不存在显著性差异，其相关性高，检测结果准确，检测时间短（2h），检出限为10～107cfu/mL，可以应用于实际检测。（4）对金黄色葡萄球菌、蜡样芽孢杆菌、沙门氏菌的生物学、培养性质、检测方法等进行了比较，三种致病菌的计算机视觉检测方法各有其原理和特点，但都是基于经过前处理后，将对应菌体进行染色，达到图像识别要求后，用计算机通过设定的程序进行识别，三种方法检测时间都很短，都能保证在5小时以内完成检测。更多还原

【Abstract】 Staphylococcus aureus, Bacillus cereus and Salmonella are considered the mostcommon cause of food-borne illness worldwide. The plate count method is acommonly accepted, reliable method for the detection of them, but is timeconsuming for needing a series of steps such as enrichment, separation andpurification, biochemical identification and serological test and cannot meet therequirements for achieving rapid analysis. Thus, in order to shorten the analyticaltime requirement, simplify testing procedures, and meet the necessity of rapiddetection of S. aureus, B. cereus and Salmonella for the modern food industry, wedeveloped a rapid detection method using staining and computer vision technology.Compared to the plate count method, which requires3to7d, this new detectionmethod offers great time savings. The total analysis time was2to5h. Our new,rapid detection method for microorganisms in foods has great potential for routinefood safety control and microbiological detection applications. In this paper, thepretreatment methods, staining methods and computer vision technologies for S.aureus, B. cereus and Salmonella were studied using the Food MicrobiologicalRapid Detection system (FMRDS) manufactured by our laboratory.The main research work and results are as follows:1. In this section, we studied a rapid detection method based on computer visionfor selective isolation and identification of S. aureus from foods. The researchincluded preparation of selective medium, selection of inhibitors and accelerators,optimizing of selective medium, staining of S. aureus, image denoising,segmentation, morphological operation treatment, characteristic parametersextraction and artificial neural network identification. The detection range for S.aureus was confirmed and this method was compared with the plate count method.The conclusions are as follows:Through single factor and response surface analysis experiments, the formula ofS. aureus selective medium was determined to be17.0g/L tryptone,3.0g/L phytone,68.0g/L NaCl,2.5g/L KH2PO4,2.5g/L glucose,5g/L sodium pyruvate,9.0g/Lglycine,60mg/L K2TeO3, and3.8g/L phenethyl alcohol.Rapid detection by the FMRDSSolid sample (25g) was placed in a sterile homogeneous bag containingphysiological saline (225mL) and flapped for1to2min with a flapping homogenizer to create a liquid bacterial sample (1:10). To minimize large particles interference andfacilitate subsequent enrichment process, medium speed quantitative filter paper wasused to remove large particles. The large particulate matter was removed and filterpaper was rinsed with physiological saline (100mL) in order to minimize bacterialloss.A0.1-mL aliquot of the S. aureus suspension was added into a centrifuge tube(1.5mL) containing S. aureus selective medium (0.9mL). It was then fermented at37C for4h, and an aliquot of the fermentation solution (0.1mL) was taken anddiluted to10mL with sterile distilled water.The10-mL dilution was concentrated using a syringe filter (Ф13/0.45μm).Using disposable syringe,50μL of air was drawn and pushed through the filter3times to mix. The concentrated fermentation solution (2μL) close to the filtermembrane was then transferred onto the slide and air-dried. Sterile distilled water wasdripped on the slide for rinsing, and excess moisture on the slide was absorbed withfilter paper. Finally, the slide was analyzed using the rapid detection system.The image identification program of the food microbiological rapid detectionsystem included the following steps: median filtering denoising, the global thresholdsegmentation, open operation smooth processing and edge extraction, using BPneural network model for image discriminant4-5-1.Compared to the Baird-Parker plate count method, this rapid detection methodoffers great time savings. The total analysis time was5h. Results using the rapiddetection method agree with those using the traditional Baird-Parker method with acorrelation coefficient of>0.99in the range of10-107cfu/mL.2. In this section, we researched the rapid detection method of B. cereus basedshort-term fermented. The research content including B. cereus nature, heattreatment, bacteria suspension preparation, microwave germination and short-termfermentation, bacteria liquid concentration and dyeing, image denoising,segmentation, morphological operation treatment, characteristic parametersextraction and artificial neural network identification. The rapid method wascompared with the plate count method. The conclusions are as follows:Sample was made into solution and heated it at65℃for30min to sterilize nontarget bacteria. Based on LB medium, the germination agent combination wasdetermined to be0.05%ZnSO4,0.1%MgCl,0.15%MnCl. The spore suspensionwas used microwave to germinate the spores for60s under high grade condition. Thegerminated suspension was fermented for2h and filtered with needle filter.Rapid detection by the FMRDSSolid sample (25g) was taken which was crumbled fully and put into conical flask which contained225mL physiological saline and shook up. Put conical flask inwater bath at65℃for30min, and cooled it to room temperature.The bacteria solution above (1mL) was absorbed with pipette and put into1.5mL centrifugal pipe and then centrifuged for10min at11000rpm. Thesupernatant (0.85mL) was discarded and the germination solution (0.85mL) wasmixed and shook up. The spore suspension was put in microwave oven andgerminated for60s under the condition of high-grade and then cooled to roomtemperature rapidly.0.1mL germinated suspension was drew and put into1.5mLcentrifugal pipe which contained0.9mL short-term fermented solution. Afterfermented for2h at150rpm at37℃,0.1mL fermented solution was drew and madegradient dilution in turn with0.9mL peptone solution.1mL was absorbed by a disposable syringe and inserted it in a needle filter(Ф13/0.45μm), the bacteria liquid was concentrated with the needle filter before thecentrifugal pipe was flushed and filtered twice with1.0mL peptone solution (0.5%).Then disposable syringe was drawn out and1mL air was indrawn into filter bottomand5μL bacteria liquid closed to filter membrane was drew. Finally, the bacterialiquid was put on the slide and air-dried. Dripped6-8μL dye liquor (X-Glu) on theslide and covered glass, kept at37℃for10-20min under conditions of avoidinglight, then swilled and put it in the rapid detection system to identify.The image identification program of the system included the following steps:median filtering denoising, RGB color space segmentation, open and close operationsmooth processing and morphological and color feature extraction, using BP neuralnetwork model for image discriminant8-6-1.Compared to plate count method, there was a good linear relationship comparedwith plate method. The total analysis time was5h. The range of detection was50-1×106cfu/mL.3. In this section, we researched the rapid detection method of Salmonella. Theresearch content including the preparation of immunogold nanoparticles andantibodies, Salmonella liquid concentration, dyeing, image denoising, segmentation,morphological operation treatment, characteristic parameters extraction and artificialneural network identification. The rapid method was compared with the plate countmethod. The conclusions are as follows:Determines the conditions of the immunogold nanoparticles and antibodies,heated the0.01%gold chloride acid aqueous solution, add reducing agent citrate,after continuous boiling, the size of15nm-25nm colloidal gold solution wascompounded. Adjusted pH to9.0, mixed second antibody, and then purified thesynthetic solution with high-speed centrifugation. Rapid detection by the FMRDSSolid sample (25g) was placed in a sterile homogeneous bag containing PBS-T(175mL) and flapped for1to2min with a flapping homogenizer to create a liquidbacterial sample. To minimize large particles interference and facilitate subsequentenrichment process, medium speed quantitative filter paper was used to remove largeparticles. Then use flush the filter paper with50mL PBS-T solution and collected.10mL was absorbed by a disposable syringe and inserted it in a needle filter,2μLbacteria liquid closed to filter membrane was drew and put on the slide and air-dried.Dripped75%ethanol and kept6-9min, flushed with PBS-T solution, mixed2μlSalmonella primary antibody and immunogold nanoparticles respectively, kept at37℃for30min. And then4μl growth solution is dropped into the specimen in twicewith a total reaction time4-8min. Then swilled and put it in the rapid detection systemto identify.The image identification program of the system included the following steps:spatial domain filtering method combined with gray level window changes denoising,iterative threshold method segmentation, open and close operation smoothprocessing and fill the hole corrosion method extraction, using BP neural networkmodel for image discriminant4-3-1.Compared to plate count method, there was a good linear relationship comparedwith plate method. The total analysis time was2h. The range of detection was50-1×106cfu/mL.4. Compared the nature, training, detection methods of Staphylococcus aureus,Bacillus cereus and Salmonella. Three methods have their principles andcharacteristics, they are all based on the processing of pretreatment, staining andimage recognition. The detection time of these methods are very short which onlycost2-5hours.更多还原