We present an image analysis algorithm for flocculation quality estimation in high‐solids slurries, and demonstrate its performance using inline process images of oil sands tailings flocculation. While a skilled human operator can often successfully evaluate such images, variations in feed as well as the lack of isolated flocs or spatial reference‐points inherent in a high‐solids slurry can cause conventional image analysis techniques to fail. We overcome these challenges by recasting the images in Fourier space, discarding phase information, and applying an eigenfaces‐inspired image recognition algorithm to the resulting spectra. Each image is represented using a few projection coefficients onto an orthogonal basis and evaluated using likelihood‐based classification schemes. This algorithm shows a high degree of success evaluating the flocculation quality of 129 batch and inline flocculation experiments (5,610 images total) utilizing feed tailings from two different oil sand producers at a variety of feed dilutions and flocculant dosing levels.

中文翻译：

---

本征谱用于絮凝质量评估

我们提出了一种图像分析算法，用于估算高固含量浆液的絮凝质量，并使用油砂尾矿絮凝的在线过程图像展示了其性能。尽管熟练的操作人员通常可以成功地评估此类图像，但进料变化以及高固含量浆料中固有的孤立絮凝物或空间参考点的缺乏会导致常规图像分析技术失败。我们通过在傅立叶空间中重铸图像，丢弃相位信息并将本征面启发的图像识别算法应用于结果光谱，来克服这些挑战。每个图像都使用几个投影系数正交表示，并使用基于似然度的分类方案进行评估。