返回Google圖書搜尋

Automatic Explosive Hazard Detection in FL-LWIR and FL-GPR Data

出版	University of Missouri--Columbia, 2012
URL	http://books.google.com.hk/books?id=jf23oAEACAAJ&hl=&source=gbs_api

註釋Detection of land mines and other buried explosive hazards, has been, and continues to be, a serious issue for military and civilian organizations. A vast amount of work has gone into developing detection and remediation technologies for these objects over the past quarter century, and a wide variety of techniques and algorithms have been described in the literature. This work encompasses five years of research into algorithms for the automated detection of explosive hazards in forward-looking long-wave infrared (FL-LWIR) and forward-looking ground-penetrating radar (FL-GPR) data; two technologies that offer the possibility for faster forward progress than current downward-looking systems. It addresses the problems of geo-referencing of individual cameras on moving platforms assuming a flat earth model, allowing transformation between FL-GPR and FL-LWIR coordinate spaces, effectively handling and exploiting multiple look sensor data when those looks have variation in perspective, the design of complex, yet efficient, prescreening algorithms, and target recognition using cell-structured image features with highly unbalanced class distributions. Many algorithms from the areas of machine learning, computer vision, and computational intelligence have been utilized in creating these algorithms, such as support vector machine (SVM) classification and cell-structured image feature extraction using image features such as the local binary pattern (LBP), the histogram of oriented gradients (HOG), a modified version of the edge histogram descriptor (EHD), and maximally stable extremal region (MSER) segmentation based shape features. Other techniques include genetic algorithm training of ensembles of size-contrast filters under multiple objective criteria and covariance matrix adaptation evolution strategy (CMA-ES) optimization for learning perspective transformation parameters when there is uncertainty in the locations.