Detection of thermal infrared oil tank targets based on SVM classification

Abstract

In this paper, a method for selecting and testing the features of thermal infrared (TIR) remote sensing images of oil tank targets is proposed to address deficiencies in the recognition of TIR images due to their characteristic low contrast, striping noise, and low spatial resolution. An evaluator was used to select 22 out of 29 features including texture and geometry, and training was conducted for samples with respect to these features through the libSVM (a support vector machine). This could lead to an effective detection of quasi-circular oil tank targets with a certain degree of robustness. The one-dimensional statistical features (pixel value, mean value, variance, mid-value, and gray level histogram), LBP features, EOH features and invariant moment features are more meaningful for the recognition of TIR remote sensing images of oil tank targets; the proper selection of training samples of oil tank with background is quite important for the effectiveness of detection; difference in detection windows can also influence the effect of detection. A method for selecting and testing the features of oil tank targets in TIR remote sensing images is proposed for effective classification of such targets under certain conditions.