A Two-stage Hybrid Algorithm for Optic Camouflage Performance Assessment Based on DEA and ANN

This paper assessed optic camouflage performance by a two-stage hybrid model combined DEA and ANN. The assessment indexes of optic camouflage performance were firstly constructed. Then we proposed the hybrid algorithm by the following two stages: (1) Conventional CCR model was improved by super-efficiency issue with non-Archimedean infinitesimal; (2) ANN was combined with super-efficiency DEA to form a hybrid model. The objectivity of teacher value in ANN which is defined as the relative efficiencies of the optic camouflage performance has a bearing on convergence speed and learning precision of network. DEA is applied to calculate teacher values, which requires no assumption on the appearance of the frontier surface as well as makes no hypothesis concerning the internal operations of a decision making unit. Meanwhile, the super-efficiency scores will help to distinguish the efficient decision making units as well. A simulation test shows that the convergence speed of the new model is increased by 57.69%, and the learning error is improved by 98.47%. The improved hybrid model has higher convergence speed and better learning precision than traditional one to assess optic camouflage performance.