With the continuous promotion of new industrialization, product informatization and population urbanization in China, the cluster model of using multiple agricultural machines for joint operation is more in line with the development trend of modern agriculture. The research of cooperative operation between agricultural machinery is the foundation of realizing large-scale cluster operation. The traditional grain harvesting process is separated from the loading process. The harvester needs to stop to unload the grain after it is fully loaded. The efficiency of grain harvesting is low when the driver identifies the loading status of grain containers and controls the movement of grain transport vehicle. The method of combining harvester and grain transport can realize unloading without stopping, speed up the operation and improve the performance of harvester. In order to improve the efficiency and monitoring ability of master-slave cooperative operation, the dynamic identification of grain loading in the grain containers of the combine harvester-grain transport vehicle system should be emphasized. At present, there is little research on the method of identifying the loading state of grain containers in the cooperative operation system of combine harvester. There are three-dimensional laser scanning sensor, ultrasonic sensor and pressure sensor. These sensors can detect the height, weight and other information of the grain, but the spatial distribution information of the grain in the grain box is difficult to identify, and there are problems such as high failure rate and poor stability. To solve the above problems, this paper adopts machine vision technology to identify the loading state of grain in the grain containers and obtain the spatial distribution state information of grain, so as to adjust the relative position between harvester and grain transport vehicle, which has better applicability for collaborative control. The research contents of this paper mainly include: (1) An improved contact line detection method was proposed for the cooperative operation of combine harvester and grain transport vehicle. The distance between the convex point on the 2-d convex hull of grain and the line of grain containers border is used to judge the loading state. Compared with the traditional method of identifying contact lines only by edge information, it is more intuitive. A Cellular Neural Network algorithm was used to detect the edge of grain transport vehicle. A method of parameter design of CNN edge template based on ant lion algorithm is presented. The iterative process of edge detection is simplified by analyzing the state equation of CNN discretization. The straight lines of grain containers border after edge detection were screened through the characteristics of angle and distance, and the grain containers area was segmented as the region of interest. Based on the yellow feature of grain, a grain area segmentation method combining RGB color space and HSV color space was designed. The convex point coordinates of grain area were detected, and the distance between grain convex point and grain containers border line was obtained. (2) In order to solve the problem that two-dimensional contact lines cannot represent complete spatial distribution information of grain containers, a 3d reconstruction method of grain area based on multiple depth cameras was proposed. The depth camera is designed in the diagonal position of the grain containers to make the most of its effective field of view. According to the camera installation location, a calibration method of multiple depth cameras based on 3D grain containers model was designed. Firstly, the 3D model diagram of the grain containers was constructed, and the template library was formed by multi-angle grid scanning at its corners. Taking the center position of the grain containers as the origin of the coordinate system, the depth camera at the diagonal position is converted to the grain containers coordinate system by the method of template matching. The point cloud information collected by two depth cameras was filtered and denoised, and the iterative nearest point algorithm was used for fusion to obtain the three-dimensional model and spatial distribution information of grains in the grain containers . (3) The simulation experiment platform was designed according to the geometric model of cooperative operation of combine harvester and grain transport vehicle. On the simulation platform, the grain loading state was identified by using the contact line obtained by the color camera and the 3d model established by the depth camera.