Based On The Topological Structure Of The Feature Extraction And Recognition Research Of Weed Seeds

In recent years, in the fields of computer pattern recognition and image processing,automatic classification and recognition is a hot research topic. Topology structure is oneof the essential attributes of images. Compared to feature points of pure angular point andinflection point, it often contains rich and more accurate image information. The mainresearch of this paper is a pattern recognition method based on topology relationshipmodeling. It is based on the topology pattern recognition theory and from the perspectiveof human to know things, The completion of identification and classification is throughthe analysis of the relationship between the overall topology and the local structurecharacteristics, and we set up a framework of image recognition. This recognitionframework is composed of three links: local structure feature extraction, the structure ofthe spatial topology relation, the local structure and global structure modeling.Farmland weed is a kind of exotic harmful organisms and the purpose plant, in theprocess of the growth of crops, it causes serious interference and harms to the growth ofcrops. In the world, weeds can lead to agricultural production, so it causes huge losses,and threatens the safety of agroforestry. Weed seeds spread to our country mainly throughmixed with into seeds such as food. In order to limit weeds and reduce the weeds fromthe source, rapid and accurate identification of weed seeds, to strengthen the automaticidentification and research work of weed seeds are very important.According to the demands of seeds identification of the entry-exit inspection andquarantine, the identifying objects of this article are bean class weed seed images, weestablish a bean class weed seeds images databases. According to the recognitionframework proposed in this paper, we use the method of sliding window to converttwo-dimensional static image into a character sequence, and we extract every localstructure characteristics of sliding window such as moment invariant feature extraction methods.As the process of sliding from left to right, and have the order, it can be constructedthe overall spatial topology relationship, finally, we use artificial neural network (ANN)model to simulate the local structure feature modeling and Hidden markov model (HMM)to simulate the overall spatial topology relationship. Artificial neural network model has astrong learning ability and good adaptability. At the same time, it has the characteristicsof variation and resisting noise, and its recognition speed is quick, therefore, it is verysuitable for the local structure modeling. Hidden markov model has a strong temporalprocessing power, so we use it to the whole space topology relation model. By combiningthe artificial neural network model and hidden markov model, we complete the model ofimage of bean class weed seed. The experimental results show that the extraction andrecognition research based on the topological structure characteristics can effectivelyimprove the classification efficiency of weed seeds.