Global optimization using memetic differential evolution with applications to low level machine vision

Abstract

In recent years meta-heuristic optimization has gained popularity in industry as well as academia due to increased computational resources and advances in the algorithms employed. As it has become apparent that no single algorithm can be declared the best in all cases, a rise in hybrid, or memetic algorithms that can be designed on case by case basis can be observed. In this work memetic algorithms are studied in depth in the context of Differential Evolution Algorithm that is among the best modern generic optimization algorithms. A whole setting of Memetic Differential Evolution methods is discovered and empirically validated with the inclusion of fitness diversity based co-ordination and stochastic adaptation schemes. This work also studies the industrial problem of real-time paper web defect detection. This is a problem that is extremely constrained in time and as such permits no complex runtime solution. Thus, the problem has been formulated as an optimization problem utilizing a simple and efficient run-time model that is tuned to precision with rather more complex methods before applying it in the industrial field. The tools used for this are meta-heuristic optimization and especially, memetic optimization.