Monitoring of Drift Patterns in Image Data

TL;DR

This paper introduces a novel method using oblique-axis regression trees for monitoring drift patterns in image sequences, effectively detecting gradual changes and abrupt shifts across various applications.

Contribution

It proposes a new decision tree-based approach for drift monitoring in images, capable of characterizing complex drift patterns and detecting abrupt changes.

Findings

Method effectively detects drift patterns in diverse simulations

Captures discontinuities in spatial and temporal image data

Remains capable of detecting abrupt step changes when no drift occurs

Abstract

Sequential monitoring of images has broad applications across various domains, including climate science, ecosystem monitoring, medical diagnostics, and so forth. In many such applications, images acquired over time exhibit gradual changes, referred to as drifts, which pose significant challenges for monitoring. Rather than detecting only abrupt step changes, it is crucial to monitor and characterize these drift patterns. Despite its practical importance, the problem of drift monitoring in image sequences has received limited attention. This paper addresses this gap by proposing a novel drift monitoring method based on an oblique-axis regression tree. It is particularly effective for monitoring drift patterns in the jump location curves present in the image intensity functions. By leveraging a decision tree framework, the method captures discontinuities both in spatial image intensity and temporal progression. A key advantage of this method lies in its flexibility: in the absence of drift, it remains capable of detecting abrupt step changes. Theoretical properties and numerical performance in diverse types of simulation settings indicate its broad applicability.