Rapid Detection Research For Microorganisms In Food Based On Biotechnology And Computer Vision

Food safety is a major public safety issue, which has a direct relationship with the people's lives, health and social stability, and microbial contamination problem is a very important factor of it. Both of aerobic bacterial counts and total coliforms counts are the important microbiological parameters which were used as the national food standards and hygiene evaluation in many countries. Moreover, the controlling of these two parameters is an important method to reduce the human health risk caused by food-borne microbe. However, the traditional detection methods were both time-consuming and complex operation, so it will not be beneficial to on-site monitor. For food safety supervision staff, whether can find the suspicious problem on scene quickly and take corresponding measures has an important significance for improving the efficiency and intensity of supervision to ensure food hygiene safety. In addition, the rapid detection method can be applied to monitor the food production processes, which is conducive to find problems directly and resolve them in time. Therefore, many rapid detection methods were proposed. A new automatic and rapid detection system for enumeration of total viable bacteria and total coliforms in food was developed based on comprehensive utilization of biotechnology technology and computer vision technology. The detailed research work and conclusions obtained of the dissertation are as follows:1 Develop a rapid detection system based on computer vision technologyThe construction of rapid detection system is the foundation of this rapid detection method in this thesis. A rapid detection system for microbiological parameters in food was developed based on computer vision technology. The realized functions of this system included: auto-focus of microscope, auto-capture bacterial images, brown-out reset, analyze and process of bacterial images, output the detection results and so on. The system utilized a normal microscope to large the bacterial images, and a color digital camera with 5.1 million effective pixels was used to capture the images of bacterial staining effect. Moreover, three motors were used to drive the object stage move along X, Y, Z-axis direction and the corresponding MCU program was designed. The stepper motor reciprocating moved along the X, Y-axis direction to capture images. One picture was captured when motor moved one step. The paces along X, Y-axis were 0.32 mm and 0.23 mm, respectively. Total area on the slide needed to be scanned is 5.12×5.12 mm. Therefore, there are 16×23 = 368 images need to be collected. These images will be analyzed one by one to determine the count by the detection system.Both rough focusing and accurate focusing were integrated in auto-focusing program of detection system. Definition evaluation function of the rough focusing method is:where,g ( x,y)is the gray value at ( x ,y) point. Accurate focusing program adoptSUM fr xySUMfgxySUMfbxywhere, fr ( x,y), fg ( x,y), fb ( x,y)described the color feature value at ( x ,y) point. The system can get good focusing effect.In addition, the system adopted 24c02 chip to memorize the position of X,Y,Z-axis. It can effectively avoid interruption caused by power down, make sure that the motors go back to the start point when each collection finished.2 Determination of viable staining method and sample pre-treating methodThe methylene blue staining method was used to distinguish the live bacteria and dead bacteria, and a color digital camera with 5.1 million effective pixels was used to capture the images of bacterial staining effect. The principle for distinguishing the live bacteria and dead bacteria was proposed. The methylene blue solution is blue when it is in an oxidizing environment, but it will turn colorless when it exposed to a reducing agent. It can be used as an indicator to determine if a bacteria is alive or not, because of the reductase in the viable bacteria. After staining, the live bacteria was colorless in the interior of a dark blue circle (the cytoderm of the bacteria was thicker than the other parts) and the dead bacteria was light blue in the interior of a dark blue circle. The samples' pre-treating is an important part of rapid detection method, it related to the accuracy and sensitivity of whole rapid detection method. The most appropriate pre-treating method for four ordinary kinds of food samples was determined, and the specific pre-processing steps were as follows: Added 175 ml sterilized normal saline into 25 g solid sample or 25 mL liquid sample (if there were some big impurity particles in liquid sample, which maybe will build up filter membrane), stirred them enough by magnetic stirrer. Then, the mixture was filtered through medium speed quantitative filter paper in order to remove the impurity particles. The bacteria were in the filtrate because most bacterial actual size is between 0.5μm to 5μm and the filtration opening size is 0.45μm. Washed the materials on the filter paper with 50 mL sterile saline, continued to collect filtrate, then 1:10 diluted solution for detection was made. Got 5 mL filtrate above using sterile injector and concentrate it by ourselves-made filter. Then, 2 mL air was insufflated into the filter from its back. 5μL liquid was taken on the filter membrane to smear the slide. Air-dried fixation method was used to keep the bacteria alive. Then, 5μL methylene blue staining solution was taken to stain the bacteria on the slide. Staining time was 2 min. Washed the stain solution on the slide slowly, and air-dried again. Then, took this slide into the rapid detection system. In addition, according to the experimental requirements of the detection system, two slide templates were designed in this paper. They were used to starting point orientation.3 Bacterial image processing and feature extraction of bacterial cellsAccording to characteristic of viable bacteria image, combined with the traditional methods, a new image processing method for viable bacteria detection was developed. In order to improve image quality and reduce noise, process the bacteria images with median filter method. The captured image was 24 bit RGB image. Therefore, three images based on R,G,B color component treated with median filter respectively, and then rebuild the filtered RGB color image.Comparing the segmentation effect of dynamic threshold segmentation method based on R,G,B,H,I,S color component respectively, the best segmentation method was ascertained. The dynamic threshold segmentation based on S component can obtain the best segmentation effect. The segmentation result demonstrated that the bacteria and the background were accurately separated and the viable bacteria and the non-viable bacteria can be separate well and truly. At that moment, non-viable bacteria were all solid, while the viable bacteria were hollow. Then, we can judge hollow regions by detecting the total pixel value of the nine points in the center. If a region was hollow, this region would be filled with black (0). If the region was solid, this region would be filled with white (255). And the outer boundaries of bacteria images were extracted in order to easy to feature extraction. But before that, in order to save the processing time, some big impurity particles and small noise points should be wipe off firstly. Image morphological opening and closing operation algorithm were used to smooth the outer and inner boundary of bacteria and eliminate the isolated point. Moreover, the connected region which area was bigger thanπ×12.5×12.5 mm2 would be eliminated.A new feature was brought up, regular degree (G) and compact degree (J), which can be used to distinguish the bacteria and non-bacteria. It can be used to describe the complexity of images. Moreover, according to the bacteria morphological features, several geometrical feature parameters and moment invariant feature parameters of the bacteria binary image were extracted. The one which would not change with the change of translation, rotation and size was chosen and used to identify the bacteria. They were: shape factor (C), eccentricity (E), rectangular degree (R). At the same time, extracted bacterial color features (g,b,H,S) and their statistics(mean and standard deviation), gmean,bmean,Hmean,Smean,gRsd,bRsd,HRsd,SRsd by the center position of the labeled region.4 Design of bacterial recognition classifierError back propagation artificial neural network model was designed. Optimization of the network predictability and performance was carried out by changing the number of hidden layers, number of neurons in the hidden layer, type of learning rule and the transfer function, in order to get the best network configuration and network parameters. A variable learning rate momentum gradient descent algorithm was adopted for the network training. The best network configuration was: The input layer consisted of threeteen neurons which corresponding to the threeteen features parameters obtained above. The output layer had one neuron representing the bacterial tutor signal. The tutor signals of viable bacterial and non-viable bacterial were 1 and 0, respectively. Moreover, one hidden layer consisted of ten neurons was the most suitable ANN structure. The maximum number of learning epochs for training was set to 50000, and allowing error was 0.0001. Moreover, tansig and tansig transfer functions were adopted for input and output layer, respectively.By using the rapid detection system integrating this classifier, total viable bacteria counts in samples could be accurately enumerated within 1 h, which was much less than 48 h by using the traditional aerobic plate count method. Furthermore, comparisons of detective results of total bacteria counts by rapid automatic detection system and aerobic plate count method were made, they were closely correlated, all of the coefficient of determination (R2) of these four kinds of samples were larger than 0.99. And the t test results also showed that there was no significant difference (P>0.05) between these two detection results. Therefore, the rapid detection system developed in the paper will greatly meet the requests of on-site rapid detection technique for the safety of agricultural products.In addition, the minimum system detection limit for total viable bacteria can arrive at 1 cell/mL. As for the solid samples, the minimum method detection for total viable bacteria can arrive at 10 cells/mL. But for most liquid samples, it can arrive at 1 cell/mL. The detection range of this rapid detection system is from 1 cell/mL to 106 cells/mL.5 Determination of predicted model of total coliformsAccording to the flora distribution characteristics of some detection samples, several mathematical models based on statistics was proposed for estimating total coliforms counts. The vegetable samples were chosen to be detected to validate the feasible of this close to real-time predicted method. Enumerated total viable bacteria counts and total viable bacilli counts by the food microbiological parameters rapid detection system. Simultaneously, total coliforms were enumerated by multiple-tube fermentation technique (traditional method). Based on the detection results, multiple regression model analysis, BP neural network analysis, and trend surface analysis were used to do the data processing. According to comparison predicted accuracy and predicted errors of these models, BP neural networks'accuracy is some what higher than that of other prediction methods, R2 = 0.9716, the mean of predicted error was 0.01314. Therefore, the BP trained neural network can be used to predict total coliforms. It will play an important role in the monitoring of water quality.6 Application prospectsIn the evaluation of food hygienic standard, the inspection results of total bacterial count and coliform group of food are two important indexes which need be strictly detected before the product factory granted. Corresponding detection time of the device has significant influence on product quality. There is no doubt that short time can release logistic pressure and decrease product number in storage, so as to accelerate the turnover of funds. Rapid detection method is very important, especially for these enterprises which running HACCP. By using the rapid detection method, these enterprises could instantly monitor pollution situation of raw material and semi finished articles, and therefore, take measure to make improvement of the over-proof products of microbe. The detection system developed in this paper can make product detection rapidly, so as to find problem product in time and make adjustment in convenient. The system also has the characteristic of high degree of automation and simple interactive operation rules; therefore it could save the cost of human resource effectively. Based on illustrated above, conclusion can be drawn that the rapid detection system developed in the paper has vast potential market space and good application prospects.