Stratified Aircraft Recognition: A Two-Step Classification Approach

The method of two-step multi-class classification was developed. Its essence is sequential binary clustering, which allows us to determine the order in which classes are compared to obtain improved final metrics. In order to get meaningful results seven classifiers were used: Hist Gradient Boosting, Random Forest, Ada Boost, MLP, SVM, KNN, and Logistic Regression. The method iteratively performs binary classifications and determines their most efficient sequence. Thus, aircrafts and non-aircrafts classification were performed first. Then, another binary classification was performed for non-aircrafts images. Numerical experiments were conducted and F1-score was calculated. The determination of the minimal representative volume for the training dataset has been accomplished. Features where extracted. The developed optimization method was checked. All metrics were cross-validated. The method was applied to recognize aircrafts, drones, and helicopters.