Research On Visual Navigation Technology Of Broad-leaved Crop For High Ridge Cultivation Model

Visual navigation has stronger adaptability than satellite navigation in the agricultural denial environment where satellite signals are interfered and occluded.Navigation line extraction is the key technology of agricultural visual navigation.The existing navigation line extraction methods mainly focus on ridge-free or low-ridge environments such as wheat and corn,and most of them are aimed at single-period environments.There is no navigation line extraction algorithm adapted to different crop sizes for multiple periods in high ridge environment of broadleaved crop.In addition,the change of light conditions will greatly affect the imaging process of broad-leaved high ridge crop environment represented by tobacco,especially the difference of light intensity caused by time change and the shadow caused by high ridges blocking light will increase the difficulty of image preprocessing.Aiming at the special field environment of broadleaved high ridge crops in multiple periods,the related research on key technology of visual navigation was carried out,the main research contents are as follows:(1)The image preprocessing method suitable for environment of broad-leaved high ridge crops was determined.After comparing the timeliness of Maximum Class Square Error Algorithm and Maximum Entropy Threshold Algorithm,the Maximum Class Square Error Algorithm with better timeliness is selected as the subsequent binarization algorithm.Three grayscale methods,namely Method of extracting Ex G component from normalized RGB color space,Method of extracting H component from HSV color space,Method of extracting a component from CIE-Lab color space,were combined with the maximum between-class variance algorithm.Three combination methods are tested and compared under sunny and cloudy conditions.According to the test results,Method of extracting a component from CIE-Lab color space and the Maximum Class Square Error Algorithm were used as the grayscale method and binarization method in this study.In order to reduce the noise and holes in the binary images,the binary images are morphologically processed.The impact of morphological processing on subsequent operations is very limited.(2)A method of extracting navigation lines based on broad-leaved crop lines in high ridge environment is proposed.The method consists of four parts: image preprocessing,feature point extraction,navigation line calculation and feedback mechanism for dynamic number of horizontal strips.On the basis of image preprocessing,crop row feature points are obtained by equidistant segmented vertical projection method.The adjacent feature points are adaptively clustered,and the dynamic segmentation point clustering method is used to determine the cluster feature point set.The horizontal distance optimization and point-line distance optimization are performed on the cluster feature point set.The linear regression method based on Huber loss function is used to fit the optimized feature point set to obtain the crop lines,and the navigation line is calculated according to the crop lines.A horizontal strip dynamic quantity feedback mechanism is designed to improve the multi-period adaptability of the algorithm.Experiments have shown that the navigation line extraction method in this study can meet the basic efficiency requirements of visual navigation,and the relative least squares method can adapt to longer crop growth periods.(3)A visual navigation system based on fuzzy control is developed,and related experiments and analysis are carried out.A camera calibration program was developed using the Zhang calibration method to obtain camera parameters and correct distorted images.Based on the navigation line,the two pose parameters,distance deviation and angle deviation,are calculated according to the camera parameters and the principle of perspective transformation.The fuzzy control rules are formulated.the navigation software running on the host computer and the lower computer control circuit are developed,and the test vehicle model is assembled.The accuracy test of pose calculation was conducted to evaluate the pose parameter errors calculated by the system.The evaluation results showed that the errors of distance deviation and angle deviation are within acceptable ranges.Non-interference navigation experiments were conducted in three simulation scenarios to test the system’s performance in multiple period environments;Light-interference navigation experiments were conducted in two simulated scenarios with different lighting conditions,and the performance of the system was tested under different lighting conditions.The motion curves obtained from the above two performance tests show a convergence trend,and the vehicle can automatically navigate along the crop rows on both sides.The timeliness of the image processing program and the entire visual navigation system was tested,and the experimental data showed that both can meet the basic efficiency requirements.