| Based on crop row extraction technology,this thesis analyses the research results of crop row extraction methods at home and abroad,and addresses the problem of autonomous walking during weeding operations of organic soybeans(first leaf stage)in the Northeast.The control system was designed through a study of visual pair row recognition methods for soybean seedling belts.It is conducive to reducing labour costs and improving work efficiency;relieves the heavy workload of drivers and reduces repetitive work;promotes the development of mechanised weeding technology for soybeans,and plays a vital role in the development of intelligent agricultural equipment.The main work accomplished and conclusions are as follows:(1)To study the method of soya bean seedling strip segmentation based on common colour imagesThe image acquisition time of the soybean seedling strip was determined,and according to the information characteristics of the soybean seedling strip image,the soybean seedling strip in the farmland image was segmented out,and by comparing the super green feature(2G-R-B)greyscale method with the YC_rC_ggreyscale method,the super green feature(2G-R-B)method was finally chosen to greyscale the image,and the shadows,dead grass and soil in the image were suppressed,and the background of the soybean seedling strip was Separation was carried out,and the grey scale values of the image were split into two parts with the best threshold using the Maximal Variance Between Clusters(OTSU)method,and the noise in the image was effectively filtered out by the open operation and the small area removal method.(2)Study of soybean seedling with row line extraction methodThe study of soybean seedling strip line extraction method based on probabilistic Hough transform and soybean seedling strip line extraction method based on SUSAN corner point detection;by using deep learning algorithm to extract soybean seedling strip lines and fusing attention mechanism,a soybean seedling strip line extraction method based on improved Inception V3-SE is proposed,and the analysis results show that the improved Inception V3-The analysis results show that the average detection time for processing a 1280×720 pixel image is120.2ms and the average seedling band line extraction accuracy is 91%,which is a highly reliable field route detection method.(3)Research on soybean field-to-row control systemBased on the improved Inception V3-SE method of soybean seedling strip row line extraction,the software and hardware experimental platform was built.Based on the original rice transplanter chassis,the structure of the mechanical chassis was improved,the vision sensor hardware selection was completed and the mounting height and mounting angle of the camera was determined according to the soybean planting environment and image acquisition requirements.Research on the row control system was completed.(4)Soybean seedling band-to-row test and analysisThe calibration principle of the colour camera is investigated and the camera is calibrated.Simulation of offline processing of soybean image sequences and analysis of the lateral differences and angular differences between the navigation lines generated by the vision algorithm and the manual navigation lines.Soybean seedlings with line to line control tests were carried out in the laboratory and the results showed that the mean error of angular deviation was0.82°with a standard deviation of 0.87°and that the soybean field to line control system could accurately extract the navigation straight line by testing its angular deviation error and transverse line offset error.When the chassis is moving at a speed of 1m/s,the alignment control system can work smoothly and meet the requirements of alignment operation. |
PDF Full Text Request