Font Size: a A A

Study Of Recognition Method Of Leguminous Weed Seeds Image

Posted on:2010-08-06Degree:DoctorType:Dissertation
Country:ChinaCandidate:C J ShiFull Text:PDF
GTID:1118330332964971Subject:Physical oceanography
Abstract/Summary:PDF Full Text Request
Weed is the plant that can disturb the condition which the primary exist on. As the alien living, it is very harmful. Through being mixed in agriculture and forestry crops seeds and grain, it is introduced into China, as a result, it has brought huge economic losses for the national economy and agriculture and forestry industry. To protect the economy security of agriculture and forestry industry, fast and accurately identifying the weed seeds is an effective means.At present, the traditional identification way of seed weed for the port is the manual retrieval method which needs long time to identify and has a low recognition rate. And currently there is no begin about the study of weed seed image recognition. According to the need of port about weed quarantine and identification, this article designs a computer-vision-based image recognition system of the category of leguminous weed seeds, and establishes the image database and feature database of weed seeds, through the analysis of biological characteristics and machine features which can be used to recognize and identify the leguminous weeds. Experiment proved that the system has improved weed seed identification and identification of efficiency and accuracy. In this paper, the main work is the following:1) To determine the basis of image characteristics of weed seeds and study the shape feature extraction methods. Recognition system through the machine, instead of the traditional taxonomic classification,to analyze and study the genetic stability of weed seeds morphological feature. Based on the analysis of the legume weed seed and the shape characteristics of umbilical, in this paper there are 9 regional identity and 7 invariant moment worked out. Experiments show that 16 feature recognition and classification is efficient. An improved pulse coupled neural network (PCNN) as an image segmentation algorithm, using the maximum entropy image as a fitness function, through the genetic algorithm to optimize PCNN connection coefficient, threshold and attenuation parameters, is used to carry on the image segmentation and binary image output to leguminous seeds and hilum, to extract the shape features. The experimental results show that the method is feasible and effective.2) The study of the fixity vision character of the image. Segmentation and put out the binary image by use of PCNN network, selection the ratio of the geometrical features between seeds and hilum, and HU's square to let it have fixity vision character on the scale, rotation and translation. Incomplete block of weed seeds in the identification, the use of feature points as the local characteristics of weed seeds, the feature has the location and orientation of weed seeds unrelated to meet the scale, rotation and translation invariance. So as to ensure the stability of the image features with the same kind of weed seeds.3) The study of the image effectiveness feature analysis and classification results. Assessment the relationship between the combination of image features and recognition rate by use of BP neural networks. Through principal component analysis and kernel principal component analysis to identify the image features with the redundancy is small, the characteristics of correlation is not strong. For the weed seeds and small samples of multi-class classification by support vector machine (SVM) method of the experimental results obtained to verify that the method of SVM recognition is effictive. At the same time used the particle cluster analysis algorithm class, but the results show the effectiveness of weed seeds classification and recognition is not strong and the limited scope of application.4) The study on method of image recognition for incomplete and mask weed seeds. The polygon expression method of boundary division is used as contour shape of approximate weed seeds. The contour feature points are used as local feature, and is inputted to self-organization delayed neural networks, Using the angle characteristic value of characteristic point and the space syntopy between them, the comparison, the match and the recognition is carried on. The experiment proved that this method may recognite the incomplete mask weed seed image efficiently.5) Design and establish the database of weed seed and the system platform of image recognition. This paper collects 69 families,808 species,5181 images totally, and storages the information of the corresponding weed seed's name, the shape, the color, the texture, the growth environment, the distribution area and image feature of leguminosae class weed seed. The platform of image recognition system contain image acquisition, pretreatment, image segmentation, binaryzation, filter, feature extraction and computation, image feature classification, pattern recognition, et al.This topic come from Inspection and Quarantine Bureau of Shandong research projects" Weed seeds of legumes classfical image Recognition System", Issue number: SK07-2004, And get in Shandong, Inspection and Quarantine, "science and technology location Award" prize. Use of the research grant weed seeds detection in soybean seeds from the United States imports, effectively improving the level of weed seed identification and enforcement of quarantine checking efficiency.
Keywords/Search Tags:weed seeds, machine recognition, pulse coupled neural network, visual invariance, BP neural network, support vector machine
PDF Full Text Request
Related items