Research On Weed Classification Based On Fuzzy Classification And Feature Fusion

As the plant diseases and insect pests, the weed is one of the most important crops in the disaster. The weeds and crops compete for space, sunlight, water and fertilizers so that the crop growth inhibition and yield. According to statistics, our country suffers harm the annual loss of weeds about 200 billion kilogram of grain, rapeseed and peanut is about 2 billion kilogram. In weed control methods, chemical weeding method is popular because of its high herbicidal ability. But the use of chemical herbicides is generally extensive large area spraying. Spraying crops and weed free herbicide soil region is not only wasteful, but also cause the ecological environment potential hazard. Variably spraying the field can significantly reduce the amount of herbicides to reduce the harm to the environment, as only spray herbicide in the area of weed. At present in the weed recognition, mainly use the feature position, plants feature (including the shape, color, texture, spectra) and 3D images and other methods. In this paper, we use four kinds of common weeds in the field as the research object, propose the fuzzy neural network technology method of color, texture, height features of weeds, to realize the effective identification of field weed.Weed identification is the process of a qualitative analysis. It is different from the general quantitative analysis, the characteristics of the weed itself is fuzzy. Defining weeds in some absolute quantity classification index is not scientific. So this paper applies the fuzzy weed characteristic parameters as the input of the network for weed identification. In order to reduce the complexity of hierarchical network structure and training, we divide the network into four parts according to the characteristics of the weed species and its influence factors on the recognition. They are color feature sub network, texture feature sub network mainly, texture feature sub network secondary, and the height of feature sub network. And then make them the automatic identification of integrated network for weed four sub networks. And the structure of the network is optimized by genetic algorithm.The new fuzzy genetic neural network has the stronger robust and adaptive.Automatic classification method of weed put forward in this paper provides support and the feasibility of scheme theory spray for the following variable herbicide.. And put height information and spectrum information into the network. There is some reference to the development of automatic identification system of weeds.