Automatic Detection of Fine Roots in Minirhizotron Images

Abstract

Minirhizotrons and specialized camera equipment have been widely adopted for in situ observation of fine root dynamics in horticultural settings. However, the laborious nature of data collection from minirhizotron images limits the number and size of experiments that can reasonably be analyzed. Here we present an algorithm for the automatic detection and measurement of roots in minirhizotron images, including the discrimination of light-colored roots from bright background objects. First, two-dimensional matched filtering and local entropy thresholding are used to produce binarized images from which roots are detected. Next, a strong root classifier based on geometric and intensity features is used to discriminate roots from unwanted background objects. A labeling algorithm identifies each individual root in the image, and root lengths and diameters are measured using Dijkstra's algorithm and the Kimura–Kikuchi–Yamasaki method for obtaining the length of a digitized path. This approach allows us to identify and measure fine roots as individuals, rather than simply measuring the aggregate root length in an image. Experimental results from a collection of 250 peach (Prunus persica) root images demonstrate the effectiveness of the approach. The algorithm is able to detect and measure a variety of roots of different shapes, sizes, and orientations, with a detection rate of 92%, a false–positive rate of 5%, and an average measurement error of 4.1% and 6.8% for length and diameter, respectively. Current work involves improving the efficiency of the algorithm and incorporating it into an application. We are also exploring algorithms for tracking the location of a root over time as it grows darker in color and blends with the surrounding soil.