IMAGE MONITORING USING MULTIVARIATE CONTROL CHARTS

Abstract

This study focuses on developing a Multivariate Hotelling's T2 control chart to monitor image data. Dimensionality reduction of the image data is achieved through Principal Component Analysis (PCA) and Sparse Principal Component Analysis (SPCA). Control charts were then built using different number of Principal Components (PCs), and their performance was evaluated using different run length metrics, including ARL, SDRL, MDRL, and percentile values. The results indicated that both models were effective in detecting shifts in the data; however, performance improved as more variance was incorporated into the model. The study concludes that SPCA is preferable due to its sparsity, which allows some loadings to be zero, thereby enhancing the interpretability of variables that affect image features. Additionally, SPCA and Hotelling T2 control chart are applied in a real-life context to detect apple rot, demonstrating efficient detection outcomes. Recommendations for future research and further evaluation of the SPCA model are presented.