关键词: 种薯, 激光点云, 三维重建, 特征提取, 芽眼识别

Abstract: Accurate recognition of seed potato bud eyes is an important prerequisite for realizing intelligent cutting. In order to solve the problems of misjudgment and difficulty in obtaining 3d position information of seed potato bud eye directly due to the influence of light in machine vision recognition, a new method of seed potato bud eye recognition based on laser 3D reconstruction was proposed. First, the ROI area in the process of the point cloud was determined to eliminate the influence of the background in the acquisition process by industrial camera to match the line laser continuous acquisition mobile chips of laser light image, using the triangulation principle to obtain the depth of the information found on the surface of a potato, light gray centroid method was utilized to extract the center, to get point cloud data found on the surface of a potato. Then, according to the point cloud sparsity, random noise and skirt noise in the point cloud were removed and obtained to improve the quality of the high point cloud and reduce the bud misjudgment rate. On the premise of retaining the features of the eyes, the voxel filtering algorithm was used to sparse the point clouds to improve the efficiency of the eyes recognition. Finally, the point cloud normal vector was obtained by plane fitting to the local neighborhood of arbitrary point on the seed potato surface, and the weighted covariance matrix was built to parameterize the seed potato surface point cloud. According to the dynamic threshold set by the matrix eigenvalue size, the surface point clouds of seed potato were initially screened, and the candidate points for seed potato sprout eye discrimination were obtained. European clustering algorithm was used to obtain the point cloud clusters of candidate points, and the largest eigenvalue point in each point cloud cluster was selected as the key point. The Angle cosine value between the center line vector and normal vector composed of the key points and other points in the neighborhood was used to screen the key points again and finally determine the location of each eye of the seed potato. The experimental results showed that the recognition rate of bud eye was 95.13% and the recognition rate of bud eye error was 4.87%, which could provide reference for bud eye recognition in intelligent cutting of potato seed potatoes.

Key words: seed potato, laser point cloud, three-dimensional reconstruction, feature extraction, bud eye recognition