Grain Storage Microorganisms Microscopic Image Processing And Recognition

China is the largest country of grain production in the world, the stored grain in national grain depot is above 100 billion kilograms. As grain has a wide variety of microbial and contains many nutrients which are good natural nutrient medium for microbe, in the right condition, the microbial in the grain will be moldy metamorphism and will affect the security of human consumption. So, it is very necessary and important to help grain manager to take some measures to develop a kind of scientific, precise and simple detection technology for stored-grain microbial.This paper makes some related studies of the stored grain microbial identification system by the technology of image processing, pattern recognition, neural networks and so on; Especially in-depth studied feature extraction, compression and classification of microbial. This paper focuses on the following aspects:1. Enhancing and segmenting stored grain microbial image Making use of the median filtering method which sliding window is cruciform to smoothing handle microbial image; Making use of Edge Detection Algorithm which based on Iterative threshold and mathematical morphology to make edge detection of microbial image.2. Forming stored grain microbial features Extracts 18 features of stored grain microbial binary image, which includes morphological features and texture features, all the features are normalized.3. Selecting stored grain microbial features On the basis of systematically analyzing the two combinatorial optimization methods between genetic algorithm and simulated annealing algorithm, proposes genetic algorithm based on simulated annealing technology according to the characteristics of stored-grain microbial microscopic image, i.e. genetic simulated annealing algorithm, in-depth studies the feature, realization approach, parameter analysis and material realization of genetic simulated annealing algorithm. Makes use of genetic simulated annealing algorithm to choose features from the selected 18 features, then choose 10 better features, for example, area, perimeter, complexity, and so on.4. Compressing stored grain microbial features Analyses the traditional methods of features compression, such as the method that based on Category of the inter-distance, based on K-L translation, based on neural network and based on Wavelet Analysis; Making use of K-L translation compress based on the global in-class discrete matrix, the 10 dimension features are compressed into 5 dimensions in order to reduce the amount of computation classifier and to improve the overall performance of the system.5. Designing grain microbial classifier Designs improved BP neural network classifier and fuzzy classifier, Uses neural networks and fuzzy technology to achieve the classification of stored grain microbial, and in-depth studies the design of the BP network classifier. For the problems that the traditional BP algorithm easily goes into local minima and convergence speed is slow, analyses three kinds of improving methods to rapidly accelerate the network convergence. The off-line classification ratio of designed BP Network Classifier and fuss classifier to stored grain microbial reaches over 82.3%.6. Carrying out the detection system Develops grain microbial identification system software with Visual C++.net 2003 development tools.