This paper investigates feature selection for data dimensionality reduction in the binary classification of high-resolution remote sensing images. For this aim, a Teaching-Learning-Based Optimization (TLBO) algorithm is used within a wrapper feature selection (WFS) framework to pick the optimal features for the classification. In this work, the TLBO is used in conjunction with the support vector machine (SVM) classifier to constitute a machine learning paradigm. Compared to other evolutionary optimization algorithms, the proposed TLBO framework is characterized by less computational effort and no algorithm-specific parameter requirements. In the end, experimental tests are conducted to show the effectiveness of the proposed TLBO-based WFS approach by using multi-spectral data from Earth Observing-One Advanced Land Imager. The comparative results demonstrate the advantages of the proposed TLBO algorithm in achieving high classification accuracy in comparison to the genetic algorithm (GA) and particle swarm optimization (PSO).

Structure from Motion (SfM) is a proficient technique for 3D reconstruction from multiple views. However, its potentials could only be seen after the development of computer and digital photograph especially in the developing countries. Nowadays, SfM is applied in many applicative scenarios ranging from earth observation to heritage documentation. An important task in SfM that has been solved just recently in practice is bundle adjustment, which is accomplished through nonlinear least-squares optimization based on Levenberg-Marquardt (LM) algorithm. The aim of this paper is to present an experimental validation of the most popular open-source optimization frameworks, which implanted LM algorithm for solving bundle adjustment in SfM. For this purpose, the comparison among their performance is conducted by using bundle adjustment in the large benchmark.