AI framework with computational box counting and Integer programming removes quantization error in fractal dimension analysis of optical images
AI framework with computational box counting and Integer programming removes quantization error in fractal dimension analysis of optical images
复制标题
DOI：
10.1016/j.cej.2022.137058
复制
发表时间：
2022-05-28
影响因子：
15.1
通讯作者：
You, Fengqi
中科院分区：
文献类型：
--
作者：
Liang, Haoyue;Tsuei, Michael;You, Fengqi
由于许多化学工程现象和过程涉及独特的形状和结构，分形维数（FD）分析是化学工程界的极大兴趣，因为它提供了一个形状复杂性的统计指标，不仅量化图像结构，而且还有助于解释功能特性。由于图像旋转和平移引入的量化误差（QE），过去用于估计FD的盒计数（BC）方法不一致。在这项工作中，我们提出了一个系统和自动的人工智能（AI）框架，通过整合图像预处理，基于集合覆盖优化的BC和回归分析，在没有QE的情况下，一致地估计与化学工程不同领域相关的FD。作为确定性优化技术的结果，FD估计对于所有图像保持一致，而不管图像旋转和平移。从界面科学、生物医学工程、人体解剖学等领域获得的图像数据集的结果证明了有效的框大小规格，以最大限度地提高各种图像的回归效率和FD计算精度。总的来说，拟议的人工智能框架提供了一种新的方法，可以准确有效地估计化学工程界感兴趣的光学图像的FD。
Because many Chemical Engineering phenomena and processes involve distinctive shapes and structures, fractal dimension (FD) analysis is of great interest to the Chemical Engineering community due to its utility in providing a statistical index of shape complexity, which not only quantifies image structure but also helps explain functional properties. Past box counting (BC) methods for estimation of FDs are inconsistent due to quantization error (QE) introduced from image rotation and translation. In this work, we propose a systematic and automatic artificial intelligence (AI) framework that consistently estimates FDs of relevance to different fields of Chemical Engineering without QE by integrating image preprocessing, set-covering optimization-based BC, and regression analysis. As a result of the deterministic optimization technique, FD estimations remain consistent for all images regardless of image rotation and translation. The results of image datasets obtained from fields such as interfacial science, biomedical engineering, human anatomy, among others, demonstrate efficient box size specification to maximize regression effectiveness and FD calculation accuracy for a variety of images. Overall, the proposed AI framework offers a new means of estimating FD accurately and efficiently for optical images of interest to the Chemical Engineering community.