Research on Infrared Image SVM Segmention Algorithm for Mine Drilling and Rescue

Abstract

In order to identify human body targets and determine miners' locations with infrared images of life information during mine rescue by drilling, the infrared images should be accurately segmented. However, the current segmentation methods fail to meet the requirements due to the inherent limitations and the complex environment of post-disaster roadways. For this, this paper proposes a method of infrared image segmentation based on the SVM (Support Vector Machine) theory. It divides the videos obtained by XKQY-1 infrared life information detectors into positive, negative, and test samples, enhances the infrared images by the top-hat/bot-hat transformation method, adopts the cross validation method to select the optimum parameters C and G, trains the data set according to the RBF core function and obtains the SVM classifier to finally achieve the infrared image segmentation of the test samples. After that, this paper compares and analyzes the image segmentation effects with those of the traditional image segmentation methods such as edge detection algorithm, Otsu threshold segmentation method, K-mean clustering method, and morphological watershed method. The results indicate that the SVM-based infrared image segmentation method does not require priori knowledge and pre-processing programs such as the threshold optimization, thus it only takes 0.190s to compute, 24.33% of that in the QGA algorithm; the misclassification error rate is 0.06, 55.05% of that in the QGA algorithm; and the anti-noise capacity is stronger. In a word, this algorithm can be effectively applied in the segmentation of infrared images in mine rescue by drilling and provide great support for subsequent human target identification.