Font Size: a A A

Research On Support Vector Machine Based Weed Discrimination

Posted on:2011-10-13Degree:MasterType:Thesis
Country:ChinaCandidate:K P QuFull Text:PDF
GTID:2178330332460251Subject:Computer application technology
Abstract/Summary:PDF Full Text Request
As the quick development of Computer Vision Technology, weeds recognition by using Computer Vision Technology has been a research hotspot. By using the advantages of Support Vector Machine on solving small samples, and nonlinear problems, this paper uses Support Vector Machine in weed identification, to improve the accuracy of small samples recognition of weeds. This paper choose the northeast sugar beet and soybean fields as the main objects, analyzed the images taken by the digital camera in natural light, and developed a weed recognition system which can intelligently classify 30 species of weeds.The first step is to normalize the images of weeds, and after the normalization the size of weeds image is 256×256. In order to recognize the weed plants from the complex background environment, Excess Green Method is used for gray scaling the weed images. During the process of image binarization, an improved Otsu method was proposed. The algorithm is stable,and able to obtain the region of weed area accurately. This paper proposes an integrated use of morphological processing, hole filling algorithm and the small area of elimination algorithm to solve the problem that some binary images have little small white areas in black background or have small black holes in plant region. Then, this paper extracted the effective shape characteristics of the plant, the skeleton of plant area ratio and skeleton circumference ratio. After the target binary image is mapped to gray-scale image, we extracted the co-occurrence matrix and Gabor wavelet texture features from the gray-scale image. Put the data set into Stprtool for processing, after changing the extracted shape and texture features into the data set format of Stprtool required. Finally, we compared the combined kernel with Polynomial kernel and RBF kernel. The simulation experiments proved that the combined kernel with the appropriate parameters can enhance the performance of Support Vector Machine.
Keywords/Search Tags:Weed discrimination, Image processing, Feature extraction, Support Vector Machine(SVM), Combined kernel
PDF Full Text Request
Related items