Research On Identification For Dynamic In-field Weed Based On Image Processing Technology

Recently how to prevent and control the weeds in the field has been one of the important research topics in the development of agricultural production.As the global science develops and progresses,the research of preventing and curing the weeds in fields has the trend to technology diversification.However,as far as the practical application is concerned,the most effective way is still to use the chemical agent which causes some related problems such as the food safety and ecological pollution problems because of the long-time over-using herbicide and the increasing drug resistance of the weeds.Therefore,variable rate spraying technology has become the hot topic in agricultural research home and abroad in order to use the chemical herbicide properly and prevent insecticide sprayers from spraying indistinguishably in the large area.Among the research work,recognizing the weeds in the field and judge the density of weeds is the important step to realize the variable rate spraying.This article selected the main cash crop-corn in northeast of China as the research object,and took the intensive study of the weed identification method in the maize seedling stage based on image processing techniques.Here are the contents:(1)Analysis of dynamic scene of farmland.The image acquisition is the foundation of the following weed recognition work.It requires the following work: firstly,to simulate the dynamic scene of field work and to compare the crops' pictures under static state and dynamic state.Secondly,to analyze the influence of agricultural machinery's speed on imaging.Finally the proper debounce plan should be selected to eliminate the boresight shake and the influence of external environment on image acquisition and thereby,the real scene should be restored.(2)Extraction of color's feature.As there are no difference between the foreground information of the weed and seedling crop with the soil background color,we can extract the characteristic parameter of the image's color by comparing the most-frequent color model with newly-built color model and to eliminate the unrelated information in the external environment and reduce the light's influence.(3)Image segmentation.This article raised a kind of new segmentation algorithm of threshold sequential search based on the particle swarm optimization which combined the particle swarm optimization and classical threshold segmentation algorithm.This kind of algorithm can reduce the segmentation time of pictures and improve the processing time by simulating the way of particle optimization on the premise of ensuring the accuracy of the threshold.(4)Weed identification.Random H of transform has been used to extract the crop's ridge line and the position filling method based on the line width to collect the interline weeds' density.The result should be put into estimating the whole weeds density and finally the weed identification can be realized.The algorithm can keep the target crop's information clearly and completely and reduce the time of image acquisition,image process and weed identification and improve the real-time of weeds recognition.It can also provide some guiding significance and higher practical application value to realize the field real-time variable spray technology.