Support vector machine based classification of seafloor rock types measured underwater using Laser Induced Breakdown Spectroscopy

Laser Induced Breakdown Spectroscopy (LIBS) is an analytical technique for rapid chemical sensing that is becoming increasingly in field applications. LIBS is a form of atomic emission spectroscopy. When a high-powered pulsed laser beam is focused on the sample, it produces a plasma that emits wavelengths of light that can be observed using a spectrograph to identify the composition of the target. One of the applications of LIBS is for in situ analysis in the ocean, where recently remotely operated vehicles equipped with a LIBS devices have demonstrated in situ chemical analysis of mineral deposits on the seafloor at depths of >1000m. In this paper, an automated method to detect the class of hydrothermal deposits is developed that uses Support Vector Machine (SVM) algorithm. The 10 rock samples used in this experiments consist of 9 rock samples collected from the seafloor and 1 rock sample collected from land. Powder pressed pellets were prepared from these samples and underwater LIBS measurements were made. The high dimensional spectral dataset was normalized and preprocessed using Principal Component Analysis (PCA) to reduce the dimensions of the data and then classification was performed using SVM. Linear and Radial Basis Function kernels were used to analyze data and the classification accuracy was determined using independent sets of training and test data. The effect of increasing features on the classification accuracy was analyzed. The 10 rock samples were classified with an accuracy of ~91%, implying that the technique may be useful for in situ classification of seafloor deposits. In our future work, we will investigate the effects of increasing the number of samples on the accuracy of classification.