Study Of Water Floating Bottle Detection Algorithm With Machine Vision

Water environment problem has always been the subject of much attention. The surface wastes not only have negative visual effect, but also may cause deterioration of the water body. To remove rubbish from the surface of the water timely and effectively at the same time try to reduce the manpower, or in the case of certain resources to effectively improve the quality of the water environment, reasonable choosing the floating garbage removal time becomes the key. Apparently the garbage automatic monitoring will provide guidance for the water waste removal time. For machine vision has many successful cases in the target detection, the water floating wastes detection algorithm with machine vision is presented in this thesis.The algorithm mainly according to the process of machine vision(image acquisition, image segmentation, target extraction and pattern recognition) to be studied,its purpose is to use the machine vision to judge the amount of floating wastes and determine whether the garbage in the image need to be cleared to salvage or not. Because of the uneven luminance of image background caused by the sunshine, this thesis puts forward two correction algorithms to solve this problem, which will pave the next step of image segmentation. One is top-hat transform algorithm of grayscale morphology,another is the image background illumination correction algorithm based on probability method. Through experimental analysis, combination of the two algorithms can get more good segmentation effect. Using method of histogram twin peaks, the threshold method based on maximum entropy principle, the one-dimensional histogram maximum entropy method, fuzzy threshold method, k-means clustering method, the Otsu method and edge segmentation method on the two lighting corrected images respectively for image segmentation and merging some of them as a new segmentation image. Do 5 Binary morphological operations(Dilating, Closing, Skeleton, Dilation-skeleton and Closing-skeleton) on the new segmentation image can generate 5 kinds of binary images,along with the original binary image, there are 6 kinds of binary images. Extract the target areas as the characteristic of those images. Normalize the Characteristics of samples, use BP neural network, BP neural network optimized by particle swarm optimization, Bayesian classification algorithm, Support Vector Machines, Decision Tree and KNN algorithm 6 recognition methods to image classification recognition. Compare and analyze the image classification: need to salvage the water images and do not need to salvage the surface image. Experiments show that the recognition rate of KNN algorithm is highest, using leave-one-out method to verify the effectiveness of the algorithm and achieve satisfactory results and verify the feasibility of the water floating wastes detection algorithm with machine vision.