Field weed is the big enemy of agricultural production, and also is one of the key problems that blocked the crop growth in Chinese agriculture. Accurate positioning weeds and realizing the variable precision applying pesticide or herbicide are particularly important. To solve various field weeds positioning difficult problems, a multi-feature based weed reverse positioning method was proposed. By taking the field crops as the research object, the multi-objective weed positioning issue was transformed into single objective crop recognition problem. Firstly, seven moment invariants and eight shape feature parameters were extracted from many of the individual soybean crop leaves, and the mean value of moment invariants and shape features were taken as standard soybean leave feature value. Secondly, after a series of image preprocessing such as image segmentation, regional separation feature match and connected component analysis, multi-feature recognition method with HU invariant moments and shape features of crops were utilized to accurately locate each crop plant. Finally, based on color feature, the green plants outside of the crops region were treated as weeds. Furthermore, a small weed positioning device was designed based on this method, which was applied to wide pesticide spraying machine. Field experiment results showed that weed recognition accuracy of this system for weed in soybean field was more than 90% when the spraying machine working speed was 5km/h, hence weed positioning and pesticide accurate spraying problems could be well settled.