Multi-Objective Model Selection for Unmanned Aerial Vehicles Automatic Target Recognition Systems

• Published in: IFAC-PapersOnLine Vol. 50; no. 1; pp. 11607 - 11612
• Main Authors: Alves Ribeiro, Victor Henrique, Reynoso-Meza, Gilberto, dos Santos Coelho, Leandro
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.07.2017
• Subjects: Artificial neural networks; Classifiers; Electromagnetic signals; Genetic algorithms; Multi-objective optimization; Optimization; Signal processing; Classifiers; Electromagnetic signals; Artificial neural networks; Signal processing; Multi-objective optimization; Optimization; Genetic algorithms
• ISSN: 2405-8963; 2405-8963
• DOI: 10.1016/j.ifacol.2017.08.1652

Within unmanned aerial vehicles on-going research topics, automatic target recognition is acquiring relevance. This is due to the easiness with which it is possible to acquire such devices. In order to do this, signal processing and classification techniques could be adopted. Also, in order to improve the classification ratio, optimization techniques could be used. This subject is object of study for many researchers, but the authors worry about the lack of information while using single-objective optimization techniques. The proposed work comprises the application of multi-objective optimization design in order to create an automatic target recognition system to discriminate different types of unmanned aerial vehicles. In order to accomplish this, a K-band radar system is used to send and receive electromagnetic signals, which are then processed with feature extraction techniques, and finally, applied on an artificial neural network system. In order to improve the system’s classification ratio, the classifier is defined as a multi-objective problem, and evolutionary multi-objective optimization techniques are applied. Finally, in order to select the best possible trade-off, the level diagrams multi-criteria decision making methodology is used to compare different solutions.