Weed Recognition System Based On Binocular Vision

In order to ensure normal growth and high-yield of crops, weeding is necessary at the seedling stage. The currently large area spraying of herbicides will waste too much herbicide and result in the environment pollution.This article takes 2-5 leaves of maize and weed during seedling stage as the research object and Study on technology of weed accurate detection based on binocular stereo vision. We have found that maize seedlings are generally higher than the weeds during the same period and present a neural network classification method based on binocular stereo vision, a technique combines the features of the actual height of the plant. The main research contents of this subject include:(1) Through accuracy analysis, built and calibrated binocular vision system, design an image acquisition system and grab binocular image of crops and weeds.(2) Find a suitable method for monocular image preprocessing. Find a calculation method for extracting the shape and texture features of crops and weeds.(3) Study several kinds of height feature extraction method and present a method for extracting features of the plant height based on binocular stereo vision, a technique combines the features of the actual height of the plant and take the characteristics of the image acquisition and calculation accuracy into account.(4) The 2-5 leaves of maize weeding period can be divided into three stages and we have build SVM classification model for each stage. The three models achieve the 98.333 per cent average recognition accuracy rate. Height feature caused average recognition accuracy rate to improve by 5 per sent. This research can achieve precise weeding and can detect weeds with high accuracy.