Low-rank-constraint-based machine vision algorithm for chaffer-sieve-clogging recognition of corn harvester

Abstract

Although a chaffer sieve is used to separate impurities from kernels during corn harvesting, it is often clogged by impurities, and this has a negative impact on the separating performance. Accurate image recognition is the primary step in automatic working parameter adjustment that helps avoid clogging. Unfortunately, meshes of sieve underneath the impurities cannot be recognized using existing algorithms, and the clogging area of mesh and the impurity in background cannot be distinguished easily. To address this issue, a low-rank-constraint-based sieve clogging recognition (LSCR) algorithm is proposed in this study. Unlike existing algorithms, the position and shape of meshes are accurately estimated using the low-rank optimization strategy, and there is no need of training samples or complete information related to the mesh outline from the target images. The clogging areas are then determined based on the difference in relative reflectance. The experimental results demonstrate that the overall recognition accuracy in pixel level using the LSCR algorithm reaches 0.943 for the test scenes, which is significantly higher than that of the existing algorithms. LSCR can be potentially used for online chaffer-sieve-clogging detection in corn harvesters.