Research On Vision Navigation System Of Paddy Field Weeding Robot Based On Image Understanding

With the development of robot technology, intelligent agricultural machinery is one of the development trends in the future. How to make agricultural machinery work automatically is the current research focus. Weed, duckweed and cyanobacteria in paddy field are competing with rice seedlings for nutrient. This is a significant reason for affecting the yield and quality of rice. At present, the weeding robots and navigation algorithms are not suit for the paddy field. Therefore, the research on the visual navigation system for weeding robots on paddy field is carried out. The main research contents are list as follow:(1) To paddy field in South China, a method to extract the navigation line of seedlings is proposed. Due to the complex environment of paddy field, single gray scale method cannot get good results. So in this paper, 4 kinds of commonly used method of gray scale include G-R、2G-R-B、ExG-ExR and S component are compared and analyzed. An evaluating method of gray scale based on threshold segmentation is proposed. And the validity of the evaluation method is verified by the actual comparison. To solve the problem that the characteristic point of rice seedlings often affected by the paddy field background, this paper take the corner of the image as the charateristic point. The extraction effects as the time complexity of the algorithm of Harris corner, FAST corner and SUSAN corner are discussed. The SUSAN corner is more suitable for extrating the rice seedling characeristic points compared with FAST corner and Harris corner. And the SUSAN corner of the threshold value has been modified, making it more suitable for the extraction the corner of rice seedling. In order to reduce the time comsuming of the sequential clustering algorithm, the idea of hierarchical clustering algorithm is used to improve the algorithm. Without reducing the accuracy of clustering, the time comsuming of the sequential clustering algorithm was been greatly reduced. Then the navigation line is detected by the least square method and Hough transform based on the known point from the clustered feature point.(2)The fuzzy control rules of weeding robot are designed. After extracting the navigation line, the position and pose of weeding robot are calculated using geometric methods. The fuzzy control rules of weeding robot are designed. Then the state equation of the weeding robot is obtained by the system identification method. The controllability and observability of the state equation are verified. And the fuzzy control algorithm is verified by using Simulink.(3)A visual navigation system for paddy field weeding robot is developed and tested. The error of position deviation and angle deviation of weeding robot was tested. Then the experiment of the autonomous navigation of the weeding robot is carried out without noise in the background. We also carry out an experiment to test the autonomous navigation performance in the situation that the background color is similar to the rice seedlings. The result shows that the navigation system is still effective even when the background color is similar to the rice seedling.Research and experimental results show that the weeding robots on paddy field visual navigation algorithm developed in this paper can be used in complex environment of southern paddy field. The navigation system has a good performance in the simulative automatic row-following tests.