The Agricultural Navigation System Based On Machine Vision

With the continuous development of the machine vision technology,agricultural robot using visual navigation technology has become an important research direction of the modern intelligent agricultural machinery,and has a wide application in the automatic spraying fertilizer,harvesting,farming,weeding and pests detection and so on.According to the current development of machine vision technology in agricultural field,and the relevant research results at home and abroad,this paper design and develop a set of agricultural navigation system based on machine vision.The detailed contents about research are as follows:(1)In view of the variety and complexity of agricultural environment,in order to ensure the success rate of segmentation in various environments,this paper splits image based on the level of local feature.Due to the complicated calculation process of hierarchical segmentation structure and the consumption of time and memory,the prediction algorithm based on discrete and continuous optimal gradient direction signal,compared to those who did not use prediction algorithm,can reduce 40% of the computation time and 10% of memory use.(2)Due to the crop drilling method,and only two crop rows used in navigation line extraction,creating region of interest(ROI)is used to simplify the computation in image processing.In view of the situation of the manual creation of ROI,this paper propose the automatic ROI.Compared with the manual creation of ROI,automatic ROI can be self adjusted according to changes in conditions,such as crop species,shooting angle and so on,which is no longer confined to a particular crop or scene.(3)In view of weeds affecting the accuracy of the navigation line extraction,this paper identify weed by the machine learning,in the use of artificial neural network,support vector machine(SVM)and random forests.existing weeds in the farm,this paper use machine learning methods for weeds identification,including artificial neural network,support vector machine and random forest.The experimental results show that the three classifiers can identify weeds,and the effect of random forest isthe best,so random forest is selected as the classifier of weed identification.(4)This paper have created the appropriate ROI and excludes the impact of the weed.In the view of crop recognition problem,horizontal and linear regression combination algorithm is simple and effective and ensure the efficiency and accuracy.Finally,this paper choose the horizontal and linear regression combination algorithm to detect the crop row,and conduct the experiment and analysis.(5)According to the farmland terrain environment and the actual demand,the robot control system was designed and developed based on the HBE-ROBOCAR model of the Korean Electronic Technology Research Institute.In order to overcome the distortion of the camera,the OpenCV calibration method is used to calibrate the experiment.Finally,the simulation experiment is carried out in the school.