Fisher Discriminant Analysis (FDA), a supervised feature reduction method in seismic object detection

Automatic processes on seismic data by using pattern recognition is one of the interesting fields in geophysical data interpretation. One part is the seismic object detection using different supervised classification methods that finally has an output as a probability cube. Object detection process starts by generating a pickset of two classes; labeled as object and non–object then selecting a set of attributes that are inputs to a classifier. As a significant step before classification, it is better to perform a feature extraction algorithm to transfer data from input space to feature space; which results can be used to reduce the dimensions by eliminating less important features in the new feature space. In this paper, our goal is to investigate and propose a proper feature extraction method for the seismic object detection process. For this purpose, we propose the Fisher Discriminant Analysis (FDA) feature extraction as a suitable method for seismic object detecting and compared this method with the Principal Component Analysis (PCA) that is very popularly used method for feature extraction. The seismic object in this study is fluid migration pathways in the North Sea and Support Vector Machine (SVM) classifier with Radial Basis Function (RBF) kernel used for classification. Finally, the obtained results show that the minimum average classification error after 30 repeats by PCA is 0.152 with 3 features but for FDA it is very smaller (.078) with only 1 dimension. The second and most important result is a posterior probability in the physical domain obtained by FDA is more interpretable than PCA. In conclusion, based on the results of the FDA for seismic object detection in this study, it is recommended that the FDA as a feature extraction method (FDA in this case automatically performs feature selection and reduces the dimension) be performed before classification in seismic object detection.