Research On Autonomous Operation Perception Technology In Unstructured Farmland Environment

Since the 21st century,China has focused on accelerating the development of the agricultural machinery industry and promoting the level of agricultural mechanization.In order to improve the efficiency of agricultural production and alleviate the shortage of rural labor resources,agricultural machinery and equipment have been listed as the top ten research areas of "Made in China 2025".The environmental perception technology of agricultural machinery is an important content of intelligent research on agricultural equipment.The agricultural machinery perception task in unstructured farmland environment has mainly two directions:navigation line detection and obstacle detection.This paper aims to design a perception system with high robustness to the farmland environment,and studies the key technologies involved.The main contents are as follows:(1)Design an agricultural machinery environment perception system based on visual perception and supplemented by radar.The principle and implementation of the vision perception module,radar detection module and main controller module are studied;the communication schemes between different modules are studied,and the purpose of remotely controlling the main controller by wireless communication is achieved.(2)Aiming at the problem that the visual detection navigation line is easily affected by changes in natural light,two algorithms are proposed to remove light components from farmland images.A multi-scale reflection model of farmland images based on Retinex theory is used to eliminate light components from the image frequency domain using wavelet transform.A color camera imaging model is constructed to eliminate natural light components through the relationship between color value channels.Experiments show that both algorithms can effectively remove the illumination components in the image and enhance the robustness of the visual detection route algorithm.(3)The core content of the visual detection route algorithm is studied:image segmentation.In order to solve the problem that traditional image segmentation algorithms rely too much on image color features and cannot effectively segment fields,a segmentation algorithm based on feature design and machine learning is proposed.Superpixel segmentation algorithm preprocessing is introduced for farmland images,the color features and texture features of superpixels are designed and extracted,and the farmland parts are classified based on the support vector machine model.The experiment shows that the proposed algorithm can effectively segment different parts of farmland and complete the task of detecting the border line of the field.(4)Aiming at the problem that obstacle detection based on a single sensor is easily affected by environmental changes,a vision-radar fusion detection scheme is proposed.The fusion of vision and radar in time and space is realized.The effective target selected by the radar is used as the seed point,and the task of detecting the size of the obstacle is completed in the visual depth map.Experiments prove that this scheme can accurately detect the spatial position and size information of obstacles in front of the agricultural machinery.