Research On Rapid Detection Of Total Bacteria In Food By Digital Micro-image Identification Technique

Food is material foundation that human beings rely on for existence inactivities of life. It is also vital energy for human beings' life. The relationshipbetween food hygiene and people's health is extremely closed. By examiningmicroorganism in food, environment of food processing and hygienic conditionsof food can be judged. Basing on it, polluted degree of food by germs can beexactly evaluated, which provides scientific proofs for various jobs. According tothem, people can provide measures of prevention and cure for poisoning foodwhich has been eaten by people and animals Total bacteria is one of the mostimportant standards in food quarantine. At present, it will take 48 hours to getaccurate results by using the traditional Culture Counting Method, which can'tsatisfy the rapid pace of life. Other various methods of examination at homeand abroad also have some disadvantages that prevent expanding and completelyreplace the traditional Culture Counting Method. Therefore, the dissertationputs forward a design-rapid detection of total bacteria in food by digitalmicro-image identification technique, which is based on Machine Visionand Artificial neural network.The dissertation designs pre-manufacture handling organization andautomatic analyzing devices. The former is used to do centrifugal settling andcoloration and the latter is used to withdraw pictures of bacteria and analyzecharacteristics and count. Use a camera to take the sample germs picture and usepicture operation, smooth and filtering etc. to remove backgrounds of originalpictures and to get rid of noise etc. Basing on these, mainly focus onwithdrawing characteristics and put forward using characteristics, such asshapes of bacteria and colors of bacteria etc. to identify and input them intoclassifier of artificial neural network to make mode's identify and then giveexamination results.According to the thoughts mentioned above, following jobs have been done tomake a research on examining total bacteria in food.1. Design hardware device. Divide the automatic picture identification andchecking system into two parts: pre-manufacture handling organization andautomatic analyzing devices. Transport Methylene blue staining solution tothe microscope before processing by a slice of automatically providingequipments in order to make the system work high -efficiently. In automaticanalyzing devices, use Microscope XSP-8CA(16×100 )with three oculars , WV-CP240 type colorful CCD camera of Panasonic, SDK-2000 acquisition card ofUNIKA and a picture identification and controlling software designed by oneselfto make sure to get clear pictures of bacteria. Use color picture acquisition cardto input the pictures of bacteria into computer which can analyze the total bacteriaand count.2. Pre-process pictures. Noises in collecting pictures and unequal shining etc.influence the accuracy of partition processing, so we must use picture operation,smooth and filtering etc. means to get rid of backgrounds and noises etc. Useopening and shutting etc, operation of morphology and the method of artificialedition to get rid of the bad influences that caused by partition processing, such asthe points of spattering noises and burr and attachment and so on in order toacquire a good binarization picture.3. Picture partition. Explore various techniques of picture partition and theirapplication in picture partition and compare many operations of partition whichare used according to characteristics of microscope pictures. At last, chose to useiterating strategy and it makes a great improvement in accuracy of bacteriapartition.4. Choose characteristics. Choose characteristics among the characteristics ofbacteria and successfully get six valid characteristics basing on indexes ofperimeter, area, shape factor, rectangle degree, Stretch and gray feature. Atthe same time, choose some non-germs individuals and withdraw theircharacteristic parameter in order to provide foundation for identification.5. Identify bacteria and count. According to the research object of thedissertation and six valid characteristics, design and train the BP network theclassification machine. Establish to use sigmoid function of transfer function.Design the biggest training times is 1000 times and minimum of trainingvelocity is 0.9. The dynamic state parameter is 0.6. Allowed error is 0.0001.Perimeter area, shape factor, rectangle degree, Stretch and gray feature asnodes of the nerve network and nodes of output layer is 2 which is shown by 1and 0 of binary. Guiding Signal of bacteria is 1 while guiding signal ofnon-bacteria is 0. The nodes of the implicit layer are 6. Therefore, the networkstructure is 6-6-2. Apply nerve network to train the experiment data and the errorin training is 9.99635×10-5 and the error in test is 1.7065×10-4.6. Use the Visual C++6.0 languages to carry out the functions mentionedabove and develop a software package for identification and counting whichmatches with the hardware device system designed by the researcher. A picture cancollect 0-500 bacteria and error is±2%. It takes no more than 30s to analyze agroup of pictures.7. Contrast experiment. Apply the principle of statistics to compare the newmethod with the traditional method in contrast experiment of bacteria liquid, milk,juice and beef. The differences between the two are not obvious.There is no similar research on rapid detection of total bacteria in foodlike the dissertation. The dissertation is an experiment. Though there existmany flaws in the design of software and hardware, the results aresatisfying. There are many qualify control stations which belong to towns,cities or provinces in our country. Rapid detection of total bacteria in foodis an important job and the system talked in the dissertation has a goodmarket. We should make further improvement in speed of identification,withdrawing characteristics and design of hardware to improve accuracyand try to apply the system in the market.