Deep-Learning Segmentation of Clogging Patterns of Cylindrical Drip Emitters with Varied Geometric Features

Abstract

Three sets of drip emitter samples that were used in agricultural farms for 3–5 years were examined using a Computed Tomography (CT) scanner. The 2D slices and 3D images obtained were processed using Dragonfly 2020.1 software. Clogging material that was deposited gradually over the years on the emitter geometry was segmented using three different methods: (1) intensity thresholding, (2) machine learning (ML), and (3) deep learning (DL). The DL method not only delivered a more precise estimation of the quantity of clogging material, but also eased the segmentation process. Various measurements of emitter geometry, covering flow path and outlet areas, were taken and compared for three sample emitters. Clogging material got deposited predominantly on the outlet areas for all three samples, irrespective of their different usage times and emitter geometries. Efforts to optimize the design of emitters against clogging need to take this finding into consideration. Sample Emitter 2 with distinctly narrower flow path and smoother curved flow boundaries was found to have the least deposition of clogging material on its surface. Further study with a larger data set is required to establish a definite relationship between the geometric features and clogging intensity of drip emitters.