SVM-Based Root Identification Algorithm Using Ridge Features and Eroded Masks

Abstract

In botanical image processing, it is challenging to identify roots automatically since roots occupy only a very small part of the image and their colors are similar to those of soil. To address this, we apply three strategies. First, to fit the characteristics of roots, many ridge-like filters are applied. Second, the machine learning method of the support vector machine is applied and the boost mechanism is adopted. Also, morphology is adopted to avoid applying the pixels in the border as the training data. With these techniques, the roots with a variety of colors and widths can be accurately extracted. ? 2020 IEEE.