Detection Of Industrial Microbes Based On Digital Image Processing Method

Industrial cooling water is a particular environment which is suitable for microorganisms'growing. When the pH value, temperature, illumination of water agrees, they are able to grow and reproduce rapidly. Once they propagate excessively in the industrial cooling water system, biofouling could develop. This causes many problems such as increase of the frictional resistance in tubes and deposit corrosion, which result in perforation of equipments and influences the regular production. Consequently, it is absolutely necessary to detect microbes in electric power, chemical industry and metallurgy industry. Heterotrophic bacteria is usually detected and counted by National Standard Method of China at present, i.e. flat dish numberation, which has the disadvantages of subjectivity, big error and low efficiency because of relying on manual counting. Digital image processing method was applied in this paper to the detection of heterotrophic bacteria in industrial cooling water. The detecting process consists of getting the amount of heterotrophic bacteria colony and the recognition of bacteria in industrial cooling water. On basis of lots of experimental data, shape invariant moment, fractal method, neural network, support vector machine, simulated annealing algorithm and the genetic algorithm were applied to the recognition of the microbes in the industrial cooling water. The recognition process was discussed systematically from the two aspects of theory and experiments.First, after the acquisition of the bacteria colony image, pre-processing, binarization, overlapping segmentation, region labeling, statistical modification was carried out to obtain the amount of the bacteria colonies. In order to decrease the count error further, recognition of heterotrophic bacteria colony based on support vector machine method was then presented. The recognition accuracy reached 98.7%. Then shape invariant moment, gray level co-occurrence matrix, statistical texture, fractal dimension method were adopted to extract the features of bacteria, and 15 parameters were obtained. To realize the classification and recognition of the bacteria, the simulated annealing method combined with the genetic method was used to optimize and choose the features extracted and 11 features were attained. Eventually, the training samples after the above feature extraction process were fed into the BP Neural Network, Elman Neural Network and Support Vector Machine respectively to carry out the recognition process. The result indicated that the recognition precision of the support vector machine was the maximum. A new recognition method of the bacteria was put forward from the aspects of theory and technology in this paper.