|
China is the world's largest food production, storage and consumption country, do a good job in grain storage is related to a colossal task. In recent years, China's total grain reserves of up to 500 million tons. In order to ensure the safe storage of food, the annual national grain reserves subsidies cost of several hundred billion yuan, but there are still a lot of food because of poor management decisions and other reasons suffer losses. Which Treasury grain storage loss rate of about 0.2%, the loss is very alarming, and pests is one of the main factors. China's \Sampling method at home and abroad, the acoustic method, near-infrared method detection method can not accurately online pests in stored grain type, density, and other information. In addition, with the increase of resistance of the pests in stored grain, the type and density of recent years are on the rise, which put forward higher requirements for automatic detection of pests in stored grain. Therefore, the development of scientific and practical, accurate and convenient pests in stored grain line detection system is necessary, it is extremely urgent. Image recognition method of line detection of pests in stored grain, with high accuracy, low cost, high efficiency, no pollution, a small amount of labor, easy-to-computer and grain storage grain condition detection system connected advantages contribute to food the library management personnel for scientific decision-making, as well as when to take reasonable preventive measures to achieve food shelf, quantity, purpose of preservation. Pests in stored grain image acquisition and pre-processing is the basis of the follow-up work. In machine vision system using a CCD camera to obtain images of pests in stored grain, but also due to the the image large amount of data during transmission distortion using singular value decomposition image compression, using the method of singular value decomposition and generalized inverses of matrices image restore, and due to the influence of noise using a gray-scale transformation, wavelet transform and neighborhood filtering method for image enhancement and smoothing, and a variety of simulation results are analyzed and compared. In order to extract the stored grain pests in order to correctly identify the target, and then uses a fixed threshold method, the experience threshold method, iterative threshold method, the maximum variance threshold method, fuzzy clustering method, simulated annealing algorithm to remove food background. On this basis, the extraction of stored grain pests geometric features invariant moments and texture features. Simulated annealing algorithm for feature selection, three optimal feature. Designed on the basis of the above work, the nearest neighbor classifier and weighted Euclidean distance classifier to identify correctly identify the major pests in stored grain, the recognition rate of around 95%.
|