Loading...
Publication
Identificação de minas terrestres em imagens de infravermelho térmico
| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 2.64 MB | Adobe PDF |
Authors
Abstract(s)
Esta dissertação destina-se ao desenvolvimento de uma metodologia para a deteção de minas terrestres em imagens térmicas adquiridas em contexto militar. A deteção das minas em imagens térmicas é condicionada por diversos fatores, tais como: as características das minas, as características do terreno e profundidade a que as mesmas são colocadas. As condições em que as imagens são obtidas influenciam diretamente as metodologias utilizadas para efetuar a deteção automática das minas, através de técnicas de processamento de imagem.
A metodologia seguida neste trabalho é composta por duas fases: obtenção de imagens térmicas e o seu processamento. Na primeira fase foram elaborados diversos campos minados para averiguar quais os fatores que influenciam a deteção das minas. Na segunda fase foi feito o processamento de um conjunto de imagens para detetar as assinaturas das minas. Este processamento inclui a classificação com os algoritmos K-Nearest Neighbours (KNN) e máquina de suporte vetorial e parametrização recorrendo a validação cruzada com o método 10-fold.
Foi implementada uma segunda abordagem baseada em segmentação com recurso a limiares. Assim, foi possível efetuar uma comparação relativa entre os algoritmos de classificação e o de segmentação. A análise dos resultados obtidos permite identificar como fatores que condicionam a deteção das minas a emissividade das mesmas, a sua profundidade de enterro, a presença de vegetação na superfície do solo e o período do dia em que as imagens são obtidas.
A melhor classificação das imagens foi obtida com o classificador KNN para 40 características, selecionadas com Sequential Backward Selection (SBS), recorrendo à métrica de distância da correlação.
This paper aims to develop a methodology for the detection of landmines with thermal images acquired in military context. Thermal images enable the study of several factors that condition the detection of landmines such as their characteristics, the soil characteristics and burial depth. The conditions under which the images are obtained directly influence the methodologies used to perform the automatic detection of landmines through image processing techniques. This paper’s methodologies follow two phases: acquisition of thermal images and its processing. In the first phase were prepared some minefields to identify the factors that influence the quality of detection. In the second phase was conducted the image processing on a set of images for landmine detection. This processing was based on classification using K-Nearest Neighbor (KNN) and Support Vector Machine (SVM) algorithms and parameterization using the 10-fold cross validation technique. A second approach was implemented based on segmentation using thresholds. Thus, it was possible to make a relative comparison between the classification and segmentation algorithms. The obtained results allow us to identify as factors that influence the detection of thermal signatures of landmines the emissivity of the mines, the burial depth, the presence of vegetation on the surface of the soil and the time of the day at which images are obtained. The optimal classification was obtained with the KNN classifier to 40 features selected with Sequential Backward Selection (SBS), and using the distance metric of correlation.
This paper aims to develop a methodology for the detection of landmines with thermal images acquired in military context. Thermal images enable the study of several factors that condition the detection of landmines such as their characteristics, the soil characteristics and burial depth. The conditions under which the images are obtained directly influence the methodologies used to perform the automatic detection of landmines through image processing techniques. This paper’s methodologies follow two phases: acquisition of thermal images and its processing. In the first phase were prepared some minefields to identify the factors that influence the quality of detection. In the second phase was conducted the image processing on a set of images for landmine detection. This processing was based on classification using K-Nearest Neighbor (KNN) and Support Vector Machine (SVM) algorithms and parameterization using the 10-fold cross validation technique. A second approach was implemented based on segmentation using thresholds. Thus, it was possible to make a relative comparison between the classification and segmentation algorithms. The obtained results allow us to identify as factors that influence the detection of thermal signatures of landmines the emissivity of the mines, the burial depth, the presence of vegetation on the surface of the soil and the time of the day at which images are obtained. The optimal classification was obtained with the KNN classifier to 40 features selected with Sequential Backward Selection (SBS), and using the distance metric of correlation.
Description
Keywords
Termografia classificadores Minas terrestres Segmentação