Evolving Processing Pipelines for Industrial Imaging with Cartesian Genetic Programming

This study presents an approach based on Miller's Cartesian Genetic Programming (CGP)1 which is adapted for industrial image processing. Unlike competing methodologies that depend on large datasets, our refined CGP-IP approach allows for training filter pipelines using smaller datasets, therefore rendering systems more resilient to environmental changes and machine settings. We introduce a dependency graph to rule out invalid pipeline solutions and examine strategies to reapply previous solutions. Our modifications are designed to increase the likelihood of early convergence and improvement in the fitness indicators to allow for a more resource-efficient configuration of image filter pipelines.