AM - DC - DCEE - EEM - Outras Publicações
Permanent URI for this collection
Browse
Recent Submissions
- Image fusion for the detection of camouflaged peoplePublication . Bento, Nádia Alexandra FerreiraThe use of thermal imaging is a benefit for the Armed Forces. Due to their advantages, they have a large number of applications, including the detection of camouflaged people. For better results, the thermal information can be merged with the color information which allows a greater detail, resulting in a greater degree of security. The present study implemented as pixel level image fusion methods: Principal Components Analysis; Laplacian Pyramid; and Discrete Wavelet Transform. A qualitative analysis concluded that the method which performs better is the one that uses Wavelets, followed by the Laplacian Pyramid and finally the PCA. A quantitative analysis was made using as performance metrics: Standard Deviation, Entropy, Spatial Frequency, Mutual Information, Fusion Quality Index and Structural Similarity Index. The values obtained support the conclusions drawn from the qualitative analysis. The Mutual Information, Fusion Quality Index and Structural Similarity Index are the appropriate metrics to measure the quality of image fusion as they take into account the relationship between the fused image and the input images.
- Identification of landmines in thermal infrared imagesPublication . Pimenta, Jorge LeitãoThis paper explores the detection of landmines using thermal images acquired in military context. The conditions in which the images are obtained have a direct influence on the methods used to perform the automatic detection of landmines through image processing techniques. The proposed methodology follows two main phases: acquisition of thermal images and its processing. In the first phase, four different experiences were prepared to analyze the factors that influence the quality of the detection. In the second phase was conducted the image processing on a set of images based on classification techniques using the K-Nearest Neighbor (KNN) and Support Vector Machine (SVM) algorithms. The classification was performed on a set of features extracted from ROI’s obtained by a sliding window. A second approach was also implemented based on segmentation using thresholds. The results achieved allow to identify factors that influence the detection of the mines: the burial depth, the presence of vegetation on the surface and the time of the day at which images were obtained. The optimal classification was obtained with the KNN classifier with 40 features selected with Sequential Backward Selection (SBS), and using the distance metric of correlation.
- Automatic classification of green areas for cartography purposesPublication . Dias, Joel Augusto Joanaz D’AssunçãoThis paper exposes a study with the purpose of discriminating the different types of vegetation. One way to get information about the land is through cartography, where aerial photogrammetry is one of the most used techniques. The main objective of this study is the development of a tool capable of processing aerial photographs on the visible spectrum and for it to be able to discern and classify the different types of vegetation. The adopted methodology comprises into three major steps: feature extraction, feature selection and image classification using two classifiers, K-Nearest Neighbors and Support Vector Machine. The first step extracts statistical features and features of texture, the second step implements a technique that allows the selection of the most relevant features and the last step is divided in the optimization of the classifiers input parameters and subsequent image classification. It was not possible to use the eight classes pre-defined due to the similarity between some of them, which led to the merge of some, resulting in four new classes. The images were classified according the new classes and the performance of the two classifiers was compared. It was found that the best classifier is the Support Vector Machine using the function of kernel Radial Basis Function showing 89,8% of correct classifications. The influence of the feature selector was tested and it was concluded that it led to an average increase of 8,25% in the classifier’s performance. It was also concluded that the best results were achieved with 10 features for the KNearest Neighbors and with 20 features for the Support Vector Machine.