Automatic Plant Branch Segmentation and Classification Using Vesselness Measure

摘要

Remote monitoring of plant vegetation is an effective method to save time and to improve production efficiency. Modern agriculture techniques utilise mobile robot and machine vision for automated image acquisition and analysis. The Identification of plant parts such as leaves, stem, branches and flowers is important for assessing plant growth, irrigation strategy and plant health. In this paper, automatic segmentation and counting of plant branches based on vesselness measure and Hough Transform techniques is presented. Frangi 2D filter, based on Hessian matrix eigenvalues has been used to classify image pixels as either tube-like or blob-like. First the input image was converted to the gray scale image and used as input to the Frangi 2D filter. Size filter was used to eliminate non-branches and small objects from the image. Hough Transform was applied to detect and draw lines on the stem and branches on the image. The developed method can detect and count the branches automatically and was applied on different sides of view and different illumination conditions for the same plant. The results show a high percentage of branches segmentation for clear side views of the plant. However, branch segmentation was affected by low illumination conditions.