Weight Vector Adjustment-Based Multi-objective Segmentation of Reconstructed Thermal Images

Complex damage has special segmentation needs during reconstructed thermal image segmentation. In this section, three segmentation objective functions oriented towards the needs of noise cancellation, detail preservation and edge retention are considered simultaneously to solve the complex damage segmentation problem. However, in practical multi-objective problems, the Pareto Fronts are not ideally continuous and uniform. This chapter therefore proposes two methods of weight vector adjustment for non-regular Pareto Front surfaces. One is based on the crowding degree adaptive weight vector adjustment method from the perspective of population individual distribution; the other is based on the effective region incremental learning and PDM adjustment method from the perspective of Pareto Front surface shape learning.