Citrus Image Segmentation Based on Genetic Algorithm
|School||Changsha University of Science and Technology|
|Course||Computer Software and Theory|
|Keywords||citrus threshold segmentation interclass variance improved genetic algorithm|
Computer vision technology is widely used in agricultural products inspection in recent years, which makes modern agriculture more specialized. The thesis mainly focuses on the concrete algorithm of citrus image segmentation, which is a very important link in subject research named automatic inspection system of citrus coating damage. The automatic inspection system of citrus coating damage adopts computer vision technology. Compared with manual selection, the inspection precision of computer vision technology has improved a lot. Whether the Algorithm of Citrus image segmentation is good or not, it will directly influence damage precision of citrus coating inspection. The thesis puts forward a full set of citrus image segmentation algorithm.Image segmentation is an important and primary problem in the field of computer vision, which means decomposing image into regions of various characteristics, and extracting the interesting target. Image segmentation is a key step from image process to image analysis, which is always one of the hot problems in the research of image technology.First of all, the thesis studies the common image segmentation technologies, compares their advantages and disadvantages, and makes further research on these segmentation methods combining with genetic algorithm. Furthermore, the thesis puts forward a new algorithm for multicolor citrus image segmentation, which adopts improved genetic algorithm combining with improved threshold method. The thesis has improved the genetic algorithm because of the problem of poor convergence and premature occurrence in the traditional genetic algorithm, and works out an improved genetic algorithm which can automatically adjust crossover probability and mutation probability according to the fitness values of individuals .In addition, the thesis is based on the following characteristic: the more difference between divergent types of objects is, the better image segmentation effect will be. But to the same type of object, the image segmentation effect will be better with little difference. Combing between-class distance with interclass variance, the thesis advances an effective algorithm of optimum threshold.The thesis, through simulation experiment, brings forward threshold scope which is more stable, and makes the image segmentation edges more dedicated, Which further satisfies these requirements putting forward by automatic inspection system of citrus surface damage, such as citrus image segmentation quality, the speed of segmentation and immediacy. At the same time, the algorithm can be used in kinds of image real-time processing and analyzing, which boasts highly practicability and popularity.