Optimization of LIDAR measurements via machine learning tools

Assessing the air composition is a key task both for civil and military purposes, to protect civilians and/or to safely deploy troops. LIDAR/DIAL measurements of remote sensing have demonstrated in the last years to be both trustworthy and to be productively applied in wide and open areas in order to detect pollutants and chemical components at different wavelengths. This contribution discusses the need of optimizing the success rate of the identification of different chemical components, while minimizing the number of required spectra or wavelengths to be acquired by experimental devices; in other words to provide faster and cheaper responses without lacking in accuracy. To achieve this goal, two robust machine learning techniques have been compared using the HITRAN database: the Classification And Regression Trees (CART) and the Support Vector Machine (SVM). The results obtained and discussed in this contribution reflect considerably the influence of the added synthetic noise, implicitly representative of a real experimental condition, on both the training and the test sets and allow pointing out the advantages or the disadvantages of using one technique with respect to the other. (PDF) Optimization of LIDAR measurements via machine learning tools. Available from: https://www.researchgate.net/publication/318746327_Optimization_of_LIDAR_measurements_via_machine_learning_tools [accessed Jun 18 2018].